BERT_For_Question_Answering_Model_2.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "BERT For Question Answering - Model 2.ipynb",
      "provenance": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU",
    "widgets": {
      "application/vnd.jupyter.widget-state+json": {
        "a785fa04c6cd405e98a0c1ea48049a02": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "model_module_version": "1.5.0",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_d132c7bbc7f6489eb029e69ce185047b",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_45577540d3fb4254ac6d3d31fb270e43",
              "IPY_MODEL_955325f6ebd9477c825e897d06cf7472"
            ]
          }
        },
        "d132c7bbc7f6489eb029e69ce185047b": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "45577540d3fb4254ac6d3d31fb270e43": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "model_module_version": "1.5.0",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_2035f0a3d44543969f27acfba0bd9608",
            "_dom_classes": [],
            "description": "Downloading: 100%",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 231508,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 231508,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_6695c488a069478996bdd883c38810cd"
          }
        },
        "955325f6ebd9477c825e897d06cf7472": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_d51e4916e495421794832386cd779f51",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 232k/232k [00:08&lt;00:00, 27.3kB/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_9189de0b13a44ad5b010677eef6c2aa8"
          }
        },
        "2035f0a3d44543969f27acfba0bd9608": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "6695c488a069478996bdd883c38810cd": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "d51e4916e495421794832386cd779f51": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "9189de0b13a44ad5b010677eef6c2aa8": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "f12ef211c990423da7130c24b84c44e8": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "model_module_version": "1.5.0",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_c34de109390a4f42a61d27da847a3030",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_22ef926636b64d08a47b0e5e00fc781a",
              "IPY_MODEL_26291e2c0dbd4d428298e08e5ee37c2d"
            ]
          }
        },
        "c34de109390a4f42a61d27da847a3030": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "22ef926636b64d08a47b0e5e00fc781a": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "model_module_version": "1.5.0",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_d45a51c696974cedade00d6e45b21f19",
            "_dom_classes": [],
            "description": "Downloading: 100%",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 433,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 433,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_e4bccb1222cf4781bd305381b9a8955d"
          }
        },
        "26291e2c0dbd4d428298e08e5ee37c2d": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_5c4238fc2fd64b90b77511d0f5d10154",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 433/433 [00:06&lt;00:00, 71.8B/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_1feab8e88552475684e7f765cee2b8ee"
          }
        },
        "d45a51c696974cedade00d6e45b21f19": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "e4bccb1222cf4781bd305381b9a8955d": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "5c4238fc2fd64b90b77511d0f5d10154": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "1feab8e88552475684e7f765cee2b8ee": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "cc1b4717df2541b093f23bc6f0b8be3c": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "model_module_version": "1.5.0",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_34ef98e3087e478a94d4b72b665a167f",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_d3c8e90ba3d1492db3076dc3d997611b",
              "IPY_MODEL_afbd2d2df82a45bdbea2f1bf5540f203"
            ]
          }
        },
        "34ef98e3087e478a94d4b72b665a167f": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "d3c8e90ba3d1492db3076dc3d997611b": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "model_module_version": "1.5.0",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_318dbe0c15aa45398dd58d5e89a15d4d",
            "_dom_classes": [],
            "description": "Downloading: 100%",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 440473133,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 440473133,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_7ceec48fac05464fa4160f6255032f92"
          }
        },
        "afbd2d2df82a45bdbea2f1bf5540f203": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "model_module_version": "1.5.0",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_e7404fb6c6124a579bd11d7a6d9e6c76",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 440M/440M [00:05&lt;00:00, 76.6MB/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_79e14053b80d464287139d8bf72a3fc3"
          }
        },
        "318dbe0c15aa45398dd58d5e89a15d4d": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "7ceec48fac05464fa4160f6255032f92": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "e7404fb6c6124a579bd11d7a6d9e6c76": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "model_module_version": "1.5.0",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "79e14053b80d464287139d8bf72a3fc3": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "model_module_version": "1.2.0",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        }
      }
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "id": "a3s-IwAXoN1m",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 615
        },
        "outputId": "17448192-255a-4222-947a-88b6fb7645c0"
      },
      "source": [
        "!pip install transformers==2.11.0"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Collecting transformers==2.11.0\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/48/35/ad2c5b1b8f99feaaf9d7cdadaeef261f098c6e1a6a2935d4d07662a6b780/transformers-2.11.0-py3-none-any.whl (674kB)\n",
            "\u001b[K     |████████████████████████████████| 675kB 5.0MB/s \n",
            "\u001b[?25hRequirement already satisfied: packaging in /usr/local/lib/python3.6/dist-packages (from transformers==2.11.0) (20.4)\n",
            "Collecting tokenizers==0.7.0\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/14/e5/a26eb4716523808bb0a799fcfdceb6ebf77a18169d9591b2f46a9adb87d9/tokenizers-0.7.0-cp36-cp36m-manylinux1_x86_64.whl (3.8MB)\n",
            "\u001b[K     |████████████████████████████████| 3.8MB 22.0MB/s \n",
            "\u001b[?25hRequirement already satisfied: filelock in /usr/local/lib/python3.6/dist-packages (from transformers==2.11.0) (3.0.12)\n",
            "Requirement already satisfied: tqdm>=4.27 in /usr/local/lib/python3.6/dist-packages (from transformers==2.11.0) (4.41.1)\n",
            "Requirement already satisfied: regex!=2019.12.17 in /usr/local/lib/python3.6/dist-packages (from transformers==2.11.0) (2019.12.20)\n",
            "Collecting sacremoses\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/7d/34/09d19aff26edcc8eb2a01bed8e98f13a1537005d31e95233fd48216eed10/sacremoses-0.0.43.tar.gz (883kB)\n",
            "\u001b[K     |████████████████████████████████| 890kB 53.0MB/s \n",
            "\u001b[?25hRequirement already satisfied: dataclasses; python_version < \"3.7\" in /usr/local/lib/python3.6/dist-packages (from transformers==2.11.0) (0.7)\n",
            "Collecting sentencepiece\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/e5/2d/6d4ca4bef9a67070fa1cac508606328329152b1df10bdf31fb6e4e727894/sentencepiece-0.1.94-cp36-cp36m-manylinux2014_x86_64.whl (1.1MB)\n",
            "\u001b[K     |████████████████████████████████| 1.1MB 55.9MB/s \n",
            "\u001b[?25hRequirement already satisfied: numpy in /usr/local/lib/python3.6/dist-packages (from transformers==2.11.0) (1.18.5)\n",
            "Requirement already satisfied: requests in /usr/local/lib/python3.6/dist-packages (from transformers==2.11.0) (2.23.0)\n",
            "Requirement already satisfied: pyparsing>=2.0.2 in /usr/local/lib/python3.6/dist-packages (from packaging->transformers==2.11.0) (2.4.7)\n",
            "Requirement already satisfied: six in /usr/local/lib/python3.6/dist-packages (from packaging->transformers==2.11.0) (1.15.0)\n",
            "Requirement already satisfied: click in /usr/local/lib/python3.6/dist-packages (from sacremoses->transformers==2.11.0) (7.1.2)\n",
            "Requirement already satisfied: joblib in /usr/local/lib/python3.6/dist-packages (from sacremoses->transformers==2.11.0) (0.16.0)\n",
            "Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.6/dist-packages (from requests->transformers==2.11.0) (2020.6.20)\n",
            "Requirement already satisfied: idna<3,>=2.5 in /usr/local/lib/python3.6/dist-packages (from requests->transformers==2.11.0) (2.10)\n",
            "Requirement already satisfied: urllib3!=1.25.0,!=1.25.1,<1.26,>=1.21.1 in /usr/local/lib/python3.6/dist-packages (from requests->transformers==2.11.0) (1.24.3)\n",
            "Requirement already satisfied: chardet<4,>=3.0.2 in /usr/local/lib/python3.6/dist-packages (from requests->transformers==2.11.0) (3.0.4)\n",
            "Building wheels for collected packages: sacremoses\n",
            "  Building wheel for sacremoses (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Created wheel for sacremoses: filename=sacremoses-0.0.43-cp36-none-any.whl size=893257 sha256=df01d6df8d76badf0243ef5ccd7955a6b6753f7694cbefb23da6b326f7739a14\n",
            "  Stored in directory: /root/.cache/pip/wheels/29/3c/fd/7ce5c3f0666dab31a50123635e6fb5e19ceb42ce38d4e58f45\n",
            "Successfully built sacremoses\n",
            "Installing collected packages: tokenizers, sacremoses, sentencepiece, transformers\n",
            "Successfully installed sacremoses-0.0.43 sentencepiece-0.1.94 tokenizers-0.7.0 transformers-2.11.0\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "L_XUGqrDoWUa",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        },
        "outputId": "d48a31dc-9da8-4847-8131-5af0e58f60a1"
      },
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/drive')\n",
        "%cd '/content/drive/My Drive/Practical Topics in Machine Learning/Project'"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Mounted at /content/drive\n",
            "/content/drive/My Drive/Practical Topics in Machine Learning/Project\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KGAO4SMuULoY"
      },
      "source": [
        "import json\n",
        "import time\n",
        "import torch\n",
        "import datetime\n",
        "import numpy as np\n",
        "from torch import nn\n",
        "import torch.nn.functional as F\n",
        "from torch.utils.data import DataLoader\n",
        "from transformers import BertTokenizerFast, BertModel, BertPreTrainedModel\n",
        "from transformers import AdamW\n",
        "import matplotlib.pyplot as plt"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7ojxBv1yULs_"
      },
      "source": [
        "MODEL_FILE = \"./model2_File\"\n",
        "\n",
        "# Set the seed value all over the place to make this reproducible.\n",
        "SEED = 42\n",
        "\n",
        "np.random.seed(SEED)\n",
        "torch.manual_seed(SEED)\n",
        "torch.cuda.manual_seed(SEED)\n",
        "\n",
        "device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qz1dIWIsULwU",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 66,
          "referenced_widgets": [
            "a785fa04c6cd405e98a0c1ea48049a02",
            "d132c7bbc7f6489eb029e69ce185047b",
            "45577540d3fb4254ac6d3d31fb270e43",
            "955325f6ebd9477c825e897d06cf7472",
            "2035f0a3d44543969f27acfba0bd9608",
            "6695c488a069478996bdd883c38810cd",
            "d51e4916e495421794832386cd779f51",
            "9189de0b13a44ad5b010677eef6c2aa8"
          ]
        },
        "outputId": "0dc46139-23c6-4153-c5cd-09f611ff8880"
      },
      "source": [
        "# Load the BERT tokenizer.\n",
        "tokenizer = BertTokenizerFast.from_pretrained('bert-base-uncased')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "a785fa04c6cd405e98a0c1ea48049a02",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, description='Downloading', max=231508.0, style=ProgressStyle(descripti…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "eBFTENolBYqS"
      },
      "source": [
        "# Construct BertBaseForQA model - Version 2"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FT5jpoQbULzE"
      },
      "source": [
        "class BertBaseForQA_Version2(nn.Module):\n",
        "\n",
        "    def __init__(self):\n",
        "        super(BertBaseForQuestionAnswering_Version2, self).__init__()\n",
        "\n",
        "        # Load the pretrained BERT model of 12-layers with an uncased vocab.     \n",
        "        self.bert = BertModel.from_pretrained(\"bert-base-uncased\")\n",
        "        self.hidden_size = self.bert.config.hidden_size\n",
        "\n",
        "        # Task specific output layers - one for the start position and the second for the end position of the answer.\n",
        "        self.linear_start = nn.Linear(self.hidden_size, 1)\n",
        "        self.linear_end = nn.Linear(self.hidden_size, 1)\n",
        "\n",
        "        # Freezing BERT's embeddings.\n",
        "        for param in list(self.bert.embeddings.parameters()):\n",
        "            param.requires_grad = False\n",
        "                \n",
        "        # Freezing the first 8 layers of BERT.\n",
        "        for i in range(1, 9):\n",
        "            for param in list(self.bert.encoder.layer[i].parameters()):\n",
        "                param.requires_grad = False\n",
        "\n",
        "    def forward(self, input_ids, attention_mask, token_type_ids):\n",
        "        \n",
        "        bert_output = self.bert(input_ids=input_ids, attention_mask=attention_mask, token_type_ids=token_type_ids)\n",
        "\n",
        "        hidden_states = bert_output[0]\n",
        "\n",
        "        # Pass the hidden states through the linear layers to get the start and end logits.\n",
        "        start_logits = self.linear_start(hidden_states).squeeze(-1)\n",
        "        end_logits = self.linear_end(hidden_states).squeeze(-1) \n",
        "\n",
        "        # Producing predictive distribution for the start and end positions.\n",
        "        start_probs = F.log_softmax(start_logits, dim=1)\n",
        "        end_probs = F.log_softmax(end_logits, dim=1)\n",
        "\n",
        "        return start_probs, end_probs"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vxdqXVCpUL1u",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 115,
          "referenced_widgets": [
            "f12ef211c990423da7130c24b84c44e8",
            "c34de109390a4f42a61d27da847a3030",
            "22ef926636b64d08a47b0e5e00fc781a",
            "26291e2c0dbd4d428298e08e5ee37c2d",
            "d45a51c696974cedade00d6e45b21f19",
            "e4bccb1222cf4781bd305381b9a8955d",
            "5c4238fc2fd64b90b77511d0f5d10154",
            "1feab8e88552475684e7f765cee2b8ee",
            "cc1b4717df2541b093f23bc6f0b8be3c",
            "34ef98e3087e478a94d4b72b665a167f",
            "d3c8e90ba3d1492db3076dc3d997611b",
            "afbd2d2df82a45bdbea2f1bf5540f203",
            "318dbe0c15aa45398dd58d5e89a15d4d",
            "7ceec48fac05464fa4160f6255032f92",
            "e7404fb6c6124a579bd11d7a6d9e6c76",
            "79e14053b80d464287139d8bf72a3fc3"
          ]
        },
        "outputId": "e0a7d514-33c8-448e-ea4c-1b2f4d745513"
      },
      "source": [
        "model = BertBaseForQA_Version2()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "f12ef211c990423da7130c24b84c44e8",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, description='Downloading', max=433.0, style=ProgressStyle(description_…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "cc1b4717df2541b093f23bc6f0b8be3c",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, description='Downloading', max=440473133.0, style=ProgressStyle(descri…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4Lq0dStso5yn"
      },
      "source": [
        "# Preprocessing"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "X2hogQVFUL4K"
      },
      "source": [
        "# Takes a time in seconds and returns a string hh:mm:ss.\n",
        "def format_time(elapsed):\n",
        "\n",
        "    # Round to the nearest second.\n",
        "    elapsed_rounded = int(round((elapsed)))\n",
        "    \n",
        "    # Format as hh:mm:ss\n",
        "    return str(datetime.timedelta(seconds=elapsed_rounded))\n",
        "\n",
        "# Measure how long the preprocessing takes.\n",
        "t0 = time.time()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XS7pqYBYUL6t"
      },
      "source": [
        "def read_squad2_file(file_name):\n",
        "\n",
        "    # Load the training data.\n",
        "    with open(file_name, 'r') as file:\n",
        "        data = json.load(file)\n",
        "\n",
        "    # Create question, context and answer lists.\n",
        "    questions, contexts, answers = [], [], []\n",
        "\n",
        "    for topic in data['data']:\n",
        "\n",
        "        # For every context paragraph.\n",
        "        for paragraph in topic['paragraphs']:\n",
        "            \n",
        "            context = paragraph['context']\n",
        "\n",
        "            # For every question.\n",
        "            for qa in paragraph['qas']:\n",
        "                \n",
        "                question = qa['question']\n",
        "                answer = qa['answers'][0] if qa['answers'] else {}\n",
        "\n",
        "                # Store the question, the context paragraph and the answer.\n",
        "                questions.append(question)\n",
        "                contexts.append(context)\n",
        "                answers.append(answer)\n",
        "\n",
        "    return questions, contexts, answers\n",
        "\n",
        "train_questions, train_contexts, train_answers = read_squad2_file('train-v2.0.json')\n",
        "dev_questions, dev_contexts, dev_answers = read_squad2_file('dev-v2.0.json')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lA46xhLRUMDF"
      },
      "source": [
        "def compute_answer_end_index(contexts, answers):\n",
        "    \n",
        "    for context, answer in zip(contexts, answers):\n",
        "\n",
        "        if not answer:   # If the question is unanswerable.\n",
        "            answer['answer_start'] = 0\n",
        "            answer['answer_end'] = 0\n",
        "            continue\n",
        "\n",
        "        gold_text = answer['text'] \n",
        "        start_index = answer['answer_start']\n",
        "\n",
        "        # Compute the end index of the answer.\n",
        "        end_index = start_index + len(gold_text)\n",
        "\n",
        "        if context[start_index:end_index] == gold_text:\n",
        "            answer['answer_end'] = end_index\n",
        "        else:\n",
        "            raise ValueError\n",
        "\n",
        "compute_answer_end_index(train_contexts, train_answers)\n",
        "compute_answer_end_index(dev_contexts, dev_answers)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CuL5BLRfUL-S"
      },
      "source": [
        "# Pack every question-context in the training set.\n",
        "packed_train = [(context, question) for context, question in zip(train_contexts, train_questions)]\n",
        "\n",
        "# Tokenize all train examples.\n",
        "train_encodings = tokenizer.batch_encode_plus(packed_train, max_length=tokenizer.model_max_length,\n",
        "                                              pad_to_max_length=True, return_token_type_ids=True)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2ZQeVJM1wwV9"
      },
      "source": [
        "# Pack every question-context in the dev set.\n",
        "packed_dev = [(context, question) for context, question in zip(dev_contexts, dev_questions)]\n",
        "\n",
        "# Tokenize all dev examples.\n",
        "dev_encodings = tokenizer.batch_encode_plus(packed_dev, max_length=tokenizer.model_max_length,\n",
        "                                            pad_to_max_length=True, return_token_type_ids=True)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "amKsjrBWUsQD"
      },
      "source": [
        "def add_token_positions(encodings, answers):\n",
        "    \n",
        "    start_positions = []\n",
        "    end_positions = []\n",
        "    \n",
        "    for i in range(len(answers)):\n",
        "\n",
        "        # If the question is unanswerable, the ground truth is the [CLS] token which is in index 0. \n",
        "        if answers[i]['answer_start'] ==0 and answers[i]['answer_end']==0:\n",
        "            start_positions.append(0)\n",
        "            end_positions.append(0)\n",
        "        \n",
        "        else: # According to the start/end index, find the position of the suitable token.\n",
        "            start_positions.append(encodings.char_to_token(i, answers[i]['answer_start']))\n",
        "            end_positions.append(\n",
        "                encodings.char_to_token(i, answers[i]['answer_end'] - 1))        \n",
        "        \n",
        "        # If None, the answer passage has been truncated.\n",
        "        if start_positions[-1] is None:\n",
        "            start_positions[-1] = tokenizer.model_max_length\n",
        "            \n",
        "        if end_positions[-1] is None:\n",
        "            end_positions[-1] = tokenizer.model_max_length\n",
        "    \n",
        "    encodings.update({'start_positions': start_positions, 'end_positions': end_positions})\n",
        "\n",
        "add_token_positions(train_encodings, train_answers)\n",
        "add_token_positions(dev_encodings, dev_answers)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CaxwVLb9UsVE"
      },
      "source": [
        "class Squad2Dataset(torch.utils.data.Dataset):\n",
        "    \n",
        "    def __init__(self, encodings):\n",
        "        self.encodings = encodings\n",
        "\n",
        "    def __getitem__(self, idx):\n",
        "        return {key: torch.tensor(val[idx]) for key, val in self.encodings.items()}\n",
        "\n",
        "    def __len__(self):\n",
        "        return len(self.encodings.input_ids)\n",
        "\n",
        "train_dataset = Squad2Dataset(train_encodings)\n",
        "dev_dataset = Squad2Dataset(dev_encodings)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0IHQEQo1Usdk"
      },
      "source": [
        "# Create the DataLoaders for the training and dev sets.\n",
        "train_loader = DataLoader(train_dataset, batch_size=10, shuffle=True)\n",
        "dev_loader = DataLoader(dev_dataset, batch_size=10, shuffle=False)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CiOMMnFnUsaA",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "26bd7169-84a0-40c6-888e-2d97994125f9"
      },
      "source": [
        "# Report how long the preprocessing took.\n",
        "print(\"The preprocessing took {} (hh:mm:ss)\".format(format_time(time.time()-t0)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "The preprocessing took 0:00:50 (hh:mm:ss)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "P7jAWRxitC-p"
      },
      "source": [
        "# Evaluation Methods"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5VTH6b6C-sjJ"
      },
      "source": [
        "def get_questions_ids(file_name):\n",
        "    \n",
        "    # Load the dev data.\n",
        "    with open(file_name, 'r') as file:\n",
        "        data = json.load(file)\n",
        "\n",
        "    ids = []  # Create questions ids list.\n",
        "    \n",
        "    for topic in data['data']:\n",
        "\n",
        "        # For every context paragraph.\n",
        "        for paragraph in topic['paragraphs']:\n",
        "\n",
        "            # Store every question's id in the list.         \n",
        "            for qa in paragraph['qas']:\n",
        "                ids.append(qa['id'])\n",
        "\n",
        "    return ids\n",
        "\n",
        "ids = get_questions_ids('dev-v2.0.json')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Cb9zWFamU3l4"
      },
      "source": [
        "def evaluate(loader, ignored_index):\n",
        "    \n",
        "    # Will store the model predictions.\n",
        "    answers = []\n",
        "\n",
        "    # Reset the total dev loss.\n",
        "    total_dev_loss = 0.0\n",
        "\n",
        "    with torch.no_grad():\n",
        "\n",
        "        for batch in loader:\n",
        "\n",
        "            # Unpack the batch from the dataloader and copy each tensor to the GPU.   \n",
        "            input_ids = batch['input_ids'].to(device)\n",
        "            attention_mask = batch['attention_mask'].to(device)\n",
        "            token_type_ids = batch['token_type_ids'].to(device)\n",
        "            start_positions = batch['start_positions'].to(device)\n",
        "            end_positions = batch['end_positions'].to(device)\n",
        "\n",
        "            # Forward pass.\n",
        "            start_out, end_out = model(input_ids=input_ids, attention_mask=attention_mask,\n",
        "                                       token_type_ids=token_type_ids)\n",
        "\n",
        "            # Compute the Negative Log Liklihood loss.\n",
        "            start_loss = F.nll_loss(start_out, start_positions, ignore_index=ignored_index)\n",
        "            end_loss = F.nll_loss(end_out, end_positions, ignore_index=ignored_index)\n",
        "\n",
        "            # The total loss of the current batch is the average of the losses.\n",
        "            total_loss = (start_loss + end_loss) / 2\n",
        "\n",
        "            # Accumulate the dev loss.   \n",
        "            total_dev_loss += total_loss.item()\n",
        "\n",
        "            # Move to CPU.\n",
        "            input_ids = input_ids.detach().cpu()\n",
        "            start_out = start_out.detach().cpu()\n",
        "            end_out = end_out.detach().cpu()\n",
        "\n",
        "            # Predict the start and end positions of the answer.\n",
        "            start_predictions = torch.argmax(start_out, dim=1)\n",
        "            end_predictions = torch.argmax(end_out, dim=1)\n",
        "\n",
        "            cls_token_id = [tokenizer.cls_token_id]\n",
        "\n",
        "            # For each input id and the start and end predicted positions.\n",
        "            for current_input_ids, start_pred, end_pred in zip(input_ids.tolist(), start_predictions.tolist(),\n",
        "                                                              end_predictions.tolist()):\n",
        "\n",
        "                answer_tokens = current_input_ids[start_pred: end_pred + 1]\n",
        "\n",
        "                # If the model predicted the question is answerable with an answer span and it contains the [CLS] token \n",
        "                # within it, remove it from the answer span.\n",
        "                if cls_token_id != answer_tokens and cls_token_id[0] in answer_tokens:\n",
        "                    answer_tokens = answer_tokens[1:]\n",
        "\n",
        "                # Store predictions.\n",
        "                answers.append('' if cls_token_id == answer_tokens else tokenizer.convert_tokens_to_string(answer_tokens))\n",
        "\n",
        "    # Return the average dev loss over all of the batches and the model's predictions.\n",
        "    return total_dev_loss / len(loader), answers"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bw-gYp0q9qNF"
      },
      "source": [
        "def produce_predictions_file(questions_ids, predictions, epoch):\n",
        "    \n",
        "    preds = {key: val for key, val in zip(questions_ids, predictions)}\n",
        "\n",
        "    if epoch == 1:\n",
        "        with open('./preds_model2_epoch_1.json', 'w') as outfile:\n",
        "            json.dump(preds, outfile)\n",
        "\n",
        "        !python3 evaluate-v2.0.py ./dev-v2.0.json ./preds_model2_epoch_1.json\n",
        "    else:\n",
        "        if epoch == 2:\n",
        "            with open('./preds_model2_epoch_2.json', 'w') as outfile:\n",
        "                json.dump(preds, outfile)\n",
        "\n",
        "            !python3 evaluate-v2.0.py ./dev-v2.0.json ./preds_model2_epoch_2.json\n",
        "        else:\n",
        "            if epoch == 3:\n",
        "                with open('./preds_model2_epoch_3.json', 'w') as outfile:\n",
        "                    json.dump(preds, outfile)\n",
        "\n",
        "                !python3 evaluate-v2.0.py ./dev-v2.0.json ./preds_model2_epoch_3.json\n",
        "            else:\n",
        "                with open('./preds_model2_epoch_0.json', 'w') as outfile:\n",
        "                    json.dump(preds, outfile)\n",
        "\n",
        "                !python3 evaluate-v2.0.py ./dev-v2.0.json ./preds_model2_epoch_0.json"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "frcrYI8FrHul"
      },
      "source": [
        "# Training"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YpOf9FJxVpYb"
      },
      "source": [
        "if torch.cuda.is_available(): # Run on GPU\n",
        "    model.cuda()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1WvPg_XdVCvv"
      },
      "source": [
        "# Number of training epochs.\n",
        "EPOCHS = 3\n",
        "\n",
        "# Initialize the optimizer.\n",
        "optimizer = AdamW(model.parameters(), lr=6e-5)\n",
        "\n",
        "# Ignore this index when calculating the loss.\n",
        "ignored_index =tokenizer.model_max_length"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Jqiea6woVCzX",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "afd42ae9-d901-4914-b673-4c1696db2124"
      },
      "source": [
        "def train(train_loader, dev_loader, optimizer, epochs, ignored_index):\n",
        "\n",
        "    train_epochs_loss, dev_epochs_loss = [], []\n",
        "    train_steps_loss = []\n",
        "    \n",
        "    lr = optimizer.param_groups[0]['lr']\n",
        "\n",
        "    # Measure the total training time for the whole run.\n",
        "    total_t0 = time.time()\n",
        "          \n",
        "    print(\"\")\n",
        "    print('============================ Initial Evaluation ===========================')\n",
        "    print(\"\")\n",
        "    print(\" Evaluating...\")\n",
        "    print(\"\")\n",
        "\n",
        "    # Compute the loss and the model's predictions on the dev set, before training.\n",
        "    initial_dev_loss, preds = evaluate(dev_loader, ignored_index)\n",
        "    dev_epochs_loss.append(initial_dev_loss)\n",
        "      \n",
        "    print(\"   Initial Dev Loss: {:.3f}\".format(initial_dev_loss))\n",
        "    print(\"\")\n",
        "\n",
        "    # Report the EM and F1 score on the dev set.\n",
        "    print(\"   Initial Dev Score:\")\n",
        "    produce_predictions_file(ids, preds, 0)\n",
        "    print(\"\")\n",
        "    \n",
        "    # Measure how long the evaluation took.\n",
        "    evaluation_time = format_time(time.time() - total_t0)\n",
        "    print(\"   Initial Evaluation took: {}\".format(evaluation_time))\n",
        "    print(\"\")\n",
        "\n",
        "    # For each epoch...\n",
        "    for epoch in range(0, epochs):\n",
        "        \n",
        "        # ========================================\n",
        "        #               Training\n",
        "        # ========================================\n",
        "        \n",
        "        # Perform one full pass over the training set.\n",
        "\n",
        "        print(\"\")\n",
        "        print('=============================== Epoch {:} / {:} ==============================='.format(epoch + 1, epochs))\n",
        "        print(\"\")\n",
        "        print(' Training...')\n",
        "        print(\"\")\n",
        "\n",
        "        # Measure how long the training epoch takes.\n",
        "        t0 = time.time()\n",
        "\n",
        "        # Declare train mode.        \n",
        "        model.train()\n",
        "\n",
        "        # Reset the total loss for this epoch.\n",
        "        total_train_loss = 0.0\n",
        "\n",
        "        # Reset the loss for the next 1000 bathes.\n",
        "        sum_loss = 0.0\n",
        "\n",
        "        for batch_idx, batch in enumerate(train_loader):\n",
        "\n",
        "            # Zero the gradients from the previous iteration.\n",
        "            optimizer.zero_grad()\n",
        "\n",
        "            # Unpack the training batch from the dataloader and copy each tensor to the GPU.   \n",
        "            input_ids = batch['input_ids'].to(device)\n",
        "            attention_mask = batch['attention_mask'].to(device)\n",
        "            token_type_ids = batch['token_type_ids'].to(device)\n",
        "            start_positions = batch['start_positions'].to(device)\n",
        "            end_positions = batch['end_positions'].to(device)\n",
        "\n",
        "            # Forward pass.\n",
        "            start_out, end_out = model(input_ids=input_ids, attention_mask=attention_mask, token_type_ids=token_type_ids)\n",
        "\n",
        "            # Compute the Negative Log Liklihood loss.\n",
        "            start_loss = F.nll_loss(start_out, start_positions, ignore_index=ignored_index)\n",
        "            end_loss = F.nll_loss(end_out, end_positions ,ignore_index=ignored_index)\n",
        "\n",
        "            # The total loss of the current batch is the average of the losses.\n",
        "            total_loss = (start_loss + end_loss) / 2\n",
        "\n",
        "            # Accumulate the train loss.   \n",
        "            total_train_loss += total_loss.item()\n",
        "            sum_loss += total_loss.item()\n",
        "\n",
        "            # Backward pass to calculate the gradients.\n",
        "            total_loss.backward()\n",
        "\n",
        "            # Update the parameters.\n",
        "            optimizer.step()\n",
        "\n",
        "            # Progress update every 1000 batches.\n",
        "            if (batch_idx + 1) % 1000 == 0:\n",
        "                \n",
        "                # Compute the average loss of the last 1000 batches.\n",
        "                current = sum_loss / 1000\n",
        "                train_steps_loss.append(current)\n",
        "\n",
        "                # Calculate elapsed time in minutes.\n",
        "                elapsed = format_time(time.time() - t0)\n",
        "                \n",
        "                # Report progress.\n",
        "                print(\"   Batch: {:>6,} / {:>5,}...\".format(batch_idx + 1, len(train_loader)),\n",
        "                      \" Elapsed: {}...\".format(elapsed),\n",
        "                      \" Train Loss: {:.3f}...\".format(current))\n",
        "                \n",
        "                # Reset the loss for the next 1000 bathes.\n",
        "                sum_loss = 0.0\n",
        "        \n",
        "        # Save the current model's state.\n",
        "        torch.save({\n",
        "            'epoch': epoch+1,\n",
        "            'model_state_dict': model.state_dict(),\n",
        "            'optimizer_state_dict': optimizer.state_dict(),\n",
        "        }, MODEL_FILE + \"-epoch\" + str(epoch + 1))\n",
        "        \n",
        "        print(\"\")\n",
        "        print(\"   The current model's state has been saved\")\n",
        "\n",
        "        # Calculate the average loss over all of the batches.\n",
        "        avg_train_loss = total_train_loss / len(train_loader)            \n",
        "        train_epochs_loss.append(avg_train_loss)\n",
        "\n",
        "        # Measure how long this training epoch took.\n",
        "        training_time = format_time(time.time() - t0)\n",
        "\n",
        "        print(\"\")\n",
        "        print(\"   Average training loss: {:.3f}\".format(avg_train_loss))\n",
        "        print(\"   Training epcoh took: {}\".format(training_time))\n",
        "\n",
        "\n",
        "        # ========================================\n",
        "        #               Evaluation\n",
        "        # ========================================\n",
        "        \n",
        "        # After the completion of each training epoch, measure the model's performance on\n",
        "        # the dev set.\n",
        "\n",
        "        print(\"\")\n",
        "        print(\"\")\n",
        "        print(\" Evaluating...\")\n",
        "        print(\"\")\n",
        "\n",
        "        # Measure how long the evaluation takes.\n",
        "        t1 = time.time()\n",
        "\n",
        "        # Declare evaluation mode.\n",
        "        model.eval()\n",
        "\n",
        "        # Compute the loss and the model's predictions on the dev set.\n",
        "        avg_dev_loss, preds = evaluate(dev_loader, ignored_index)\n",
        "        dev_epochs_loss.append(avg_dev_loss)\n",
        "        \n",
        "        print(\"   Dev Loss: {:.3f}\".format(avg_dev_loss))\n",
        "        print(\"\")\n",
        "\n",
        "        # Report the EM and F1 score on the dev set.\n",
        "        print(\"   Dev Score:\")\n",
        "        produce_predictions_file(ids, preds, epoch + 1)\n",
        "        print(\"\")\n",
        "\n",
        "        # Report how long the evaluation took.\n",
        "        evaluation_time = format_time(time.time() - t1)\n",
        "        print(\"   Evaluation took: {}\".format(evaluation_time))\n",
        "        \n",
        "        print(\"\")\n",
        "        print(\"\")\n",
        "\n",
        "        # Measure how long the all epoch took - training and evaluating.\n",
        "        epoch_time = format_time(time.time() - t0)\n",
        "        print(\" Overall, Epoch took: {}\".format(epoch_time))\n",
        "\n",
        "        # Decay the learning rate every two epochs.\n",
        "        if (epoch + 1) % 2 == 0:\n",
        "            lr /= 2\n",
        "            for group in optimizer.param_groups:\n",
        "                group['lr'] = lr \n",
        "            print(\"\\n The learning rate updated to {}\".format(group['lr']))\n",
        "\n",
        "    print(\"\")\n",
        "    print(\"Training completed!\")  \n",
        "    print(\"Total training took {} (hh:mm:ss)\".format(format_time(time.time()-total_t0)))\n",
        "\n",
        "    return train_epochs_loss, dev_epochs_loss, train_steps_loss\n",
        "\n",
        "train_epochs_loss, dev_epochs_loss, train_steps_loss = train(train_loader, dev_loader, optimizer, epochs=EPOCHS, ignored_index=ignored_index)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n",
            "============================ Initial Evaluation ===========================\n",
            "\n",
            " Evaluating...\n",
            "\n",
            "   Initial Dev Loss: 6.300\n",
            "\n",
            "   Initial Dev Score:\n",
            "   {\n",
            "      \"exact\": 20.567674555714646,\n",
            "      \"f1\": 22.451588793441395,\n",
            "      \"total\": 11873,\n",
            "      \"HasAns_exact\": 0.033738191632928474,\n",
            "      \"HasAns_f1\": 3.8069692551504253,\n",
            "      \"HasAns_total\": 5928,\n",
            "      \"NoAns_exact\": 41.042893187552565,\n",
            "      \"NoAns_f1\": 41.042893187552565,\n",
            "      \"NoAns_total\": 5945\n",
            "   }\n",
            "\n",
            "   Initial Evaluation took: 0:03:50\n",
            "\n",
            "\n",
            "=============================== Epoch 1 / 3 ===============================\n",
            "\n",
            " Training...\n",
            "\n",
            "   Batch:  1,000 / 13,032...  Elapsed: 0:07:31...  Train Loss: 2.295...\n",
            "   Batch:  2,000 / 13,032...  Elapsed: 0:15:01...  Train Loss: 1.715...\n",
            "   Batch:  3,000 / 13,032...  Elapsed: 0:22:32...  Train Loss: 1.561...\n",
            "   Batch:  4,000 / 13,032...  Elapsed: 0:30:03...  Train Loss: 1.458...\n",
            "   Batch:  5,000 / 13,032...  Elapsed: 0:37:34...  Train Loss: 1.390...\n",
            "   Batch:  6,000 / 13,032...  Elapsed: 0:45:05...  Train Loss: 1.341...\n",
            "   Batch:  7,000 / 13,032...  Elapsed: 0:52:35...  Train Loss: 1.312...\n",
            "   Batch:  8,000 / 13,032...  Elapsed: 1:00:06...  Train Loss: 1.270...\n",
            "   Batch:  9,000 / 13,032...  Elapsed: 1:07:37...  Train Loss: 1.237...\n",
            "   Batch: 10,000 / 13,032...  Elapsed: 1:15:08...  Train Loss: 1.215...\n",
            "   Batch: 11,000 / 13,032...  Elapsed: 1:22:39...  Train Loss: 1.212...\n",
            "   Batch: 12,000 / 13,032...  Elapsed: 1:30:10...  Train Loss: 1.186...\n",
            "   Batch: 13,000 / 13,032...  Elapsed: 1:37:41...  Train Loss: 1.144...\n",
            "\n",
            "   The current model's state has been saved\n",
            "\n",
            "   Average training loss: 1.410\n",
            "   Training epcoh took: 1:37:59\n",
            "\n",
            "\n",
            " Evaluating...\n",
            "\n",
            "   Dev Loss: 1.136\n",
            "\n",
            "   Dev Score:\n",
            "   {\n",
            "      \"exact\": 63.42963025351638,\n",
            "      \"f1\": 67.60832839406845,\n",
            "      \"total\": 11873,\n",
            "      \"HasAns_exact\": 51.90620782726046,\n",
            "      \"HasAns_f1\": 60.2755875544489,\n",
            "      \"HasAns_total\": 5928,\n",
            "      \"NoAns_exact\": 74.92010092514718,\n",
            "      \"NoAns_f1\": 74.92010092514718,\n",
            "      \"NoAns_total\": 5945\n",
            "   }\n",
            "\n",
            "   Evaluation took: 0:03:47\n",
            "\n",
            "\n",
            " Overall, Epoch took: 1:41:45\n",
            "\n",
            "=============================== Epoch 2 / 3 ===============================\n",
            "\n",
            " Training...\n",
            "\n",
            "   Batch:  1,000 / 13,032...  Elapsed: 0:07:31...  Train Loss: 0.965...\n",
            "   Batch:  2,000 / 13,032...  Elapsed: 0:15:01...  Train Loss: 0.980...\n",
            "   Batch:  3,000 / 13,032...  Elapsed: 0:22:32...  Train Loss: 0.984...\n",
            "   Batch:  4,000 / 13,032...  Elapsed: 0:30:03...  Train Loss: 1.012...\n",
            "   Batch:  5,000 / 13,032...  Elapsed: 0:37:34...  Train Loss: 0.973...\n",
            "   Batch:  6,000 / 13,032...  Elapsed: 0:45:05...  Train Loss: 0.971...\n",
            "   Batch:  7,000 / 13,032...  Elapsed: 0:52:36...  Train Loss: 0.978...\n",
            "   Batch:  8,000 / 13,032...  Elapsed: 1:00:06...  Train Loss: 0.961...\n",
            "   Batch:  9,000 / 13,032...  Elapsed: 1:07:37...  Train Loss: 0.966...\n",
            "   Batch: 10,000 / 13,032...  Elapsed: 1:15:08...  Train Loss: 0.968...\n",
            "   Batch: 11,000 / 13,032...  Elapsed: 1:22:39...  Train Loss: 0.971...\n",
            "   Batch: 12,000 / 13,032...  Elapsed: 1:30:10...  Train Loss: 0.969...\n",
            "   Batch: 13,000 / 13,032...  Elapsed: 1:37:41...  Train Loss: 0.958...\n",
            "\n",
            "   The current model's state has been saved\n",
            "\n",
            "   Average training loss: 0.974\n",
            "   Training epcoh took: 1:37:59\n",
            "\n",
            "\n",
            " Evaluating...\n",
            "\n",
            "   Dev Loss: 1.223\n",
            "\n",
            "   Dev Score:\n",
            "   {\n",
            "      \"exact\": 64.1539627726775,\n",
            "      \"f1\": 68.51473381037972,\n",
            "      \"total\": 11873,\n",
            "      \"HasAns_exact\": 57.675438596491226,\n",
            "      \"HasAns_f1\": 66.40948625685506,\n",
            "      \"HasAns_total\": 5928,\n",
            "      \"NoAns_exact\": 70.61396131202692,\n",
            "      \"NoAns_f1\": 70.61396131202692,\n",
            "      \"NoAns_total\": 5945\n",
            "   }\n",
            "\n",
            "   Evaluation took: 0:03:47\n",
            "\n",
            "\n",
            " Overall, Epoch took: 1:41:46\n",
            "\n",
            " The learning rate updated to 3e-05\n",
            "\n",
            "=============================== Epoch 3 / 3 ===============================\n",
            "\n",
            " Training...\n",
            "\n",
            "   Batch:  1,000 / 13,032...  Elapsed: 0:07:31...  Train Loss: 0.719...\n",
            "   Batch:  2,000 / 13,032...  Elapsed: 0:15:02...  Train Loss: 0.709...\n",
            "   Batch:  3,000 / 13,032...  Elapsed: 0:22:33...  Train Loss: 0.672...\n",
            "   Batch:  4,000 / 13,032...  Elapsed: 0:30:03...  Train Loss: 0.702...\n",
            "   Batch:  5,000 / 13,032...  Elapsed: 0:37:34...  Train Loss: 0.717...\n",
            "   Batch:  6,000 / 13,032...  Elapsed: 0:45:05...  Train Loss: 0.687...\n",
            "   Batch:  7,000 / 13,032...  Elapsed: 0:52:36...  Train Loss: 0.708...\n",
            "   Batch:  8,000 / 13,032...  Elapsed: 1:00:07...  Train Loss: 0.702...\n",
            "   Batch:  9,000 / 13,032...  Elapsed: 1:07:38...  Train Loss: 0.701...\n",
            "   Batch: 10,000 / 13,032...  Elapsed: 1:15:10...  Train Loss: 0.713...\n",
            "   Batch: 11,000 / 13,032...  Elapsed: 1:22:41...  Train Loss: 0.692...\n",
            "   Batch: 12,000 / 13,032...  Elapsed: 1:30:12...  Train Loss: 0.692...\n",
            "   Batch: 13,000 / 13,032...  Elapsed: 1:37:44...  Train Loss: 0.667...\n",
            "\n",
            "   The current model's state has been saved\n",
            "\n",
            "   Average training loss: 0.699\n",
            "   Training epcoh took: 1:38:02\n",
            "\n",
            "\n",
            " Evaluating...\n",
            "\n",
            "   Dev Loss: 1.107\n",
            "\n",
            "   Dev Score:\n",
            "   {\n",
            "      \"exact\": 67.3545018108313,\n",
            "      \"f1\": 71.76107902249967,\n",
            "      \"total\": 11873,\n",
            "      \"HasAns_exact\": 57.32118758434548,\n",
            "      \"HasAns_f1\": 66.14697895312725,\n",
            "      \"HasAns_total\": 5928,\n",
            "      \"NoAns_exact\": 77.35912531539108,\n",
            "      \"NoAns_f1\": 77.35912531539108,\n",
            "      \"NoAns_total\": 5945\n",
            "   }\n",
            "\n",
            "   Evaluation took: 0:03:47\n",
            "\n",
            "\n",
            " Overall, Epoch took: 1:41:49\n",
            "\n",
            "Training completed!\n",
            "Total training took 5:09:10 (hh:mm:ss)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8lh9s6nXJoME",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 404
        },
        "outputId": "b3b3c160-551b-4873-e941-bbcd93161367"
      },
      "source": [
        "def plot_loss_graph(title):\n",
        "    \n",
        "    fig = plt.figure()\n",
        "    ax = fig.add_subplot(111)\n",
        "    ax.set_title(title)\n",
        "    ax.set_xlabel(\"Epochs\")\n",
        "    ax.set_ylabel(\"Loss\")\n",
        "\n",
        "    ticks = [1, 2, 3]\n",
        "    train_loss = [1.410, 0.974, 0.699]\n",
        "    dev_loss = [1.136, 1.223, 1.107]\n",
        "    plt.xticks(ticks)\n",
        "\n",
        "    plt.plot(ticks, train_loss, label='Train', color='indigo')\n",
        "    plt.plot(ticks, dev_loss, label='Dev', color='orange')\n",
        "\n",
        "    for idx, (a, b) in enumerate(zip(ticks, train_loss)):\n",
        "\n",
        "        if idx == 0:\n",
        "            ax.text(a+0.1, b-0.002, '1.410', fontsize=6, bbox={'facecolor': 'green', 'alpha': 0.1, 'pad': 1},\n",
        "                    verticalalignment='bottom', horizontalalignment='center')\n",
        "        else:\n",
        "            if idx == 1:\n",
        "                ax.text(a + 0.07, b + 0.012, str(b), fontsize=6, bbox={'facecolor': 'green', 'alpha': 0.1, 'pad': 1},\n",
        "                        verticalalignment='bottom', horizontalalignment='center')\n",
        "            else:\n",
        "                if idx == 2:\n",
        "                    ax.text(a+0.02, b + 0.025, str(b), fontsize=6,\n",
        "                            bbox={'facecolor': 'green', 'alpha': 0.1, 'pad': 1},\n",
        "                            verticalalignment='bottom', horizontalalignment='center')\n",
        "\n",
        "    for idx, (a, b) in enumerate(zip(ticks, dev_loss)):\n",
        "\n",
        "        if idx == 0:\n",
        "            ax.text(a, b - 0.035, str(b), fontsize=6, bbox={'facecolor': 'crimson', 'alpha': 0.1, 'pad': 1},\n",
        "                    verticalalignment='bottom', horizontalalignment='center')\n",
        "        else:\n",
        "            if idx == 1:\n",
        "                ax.text(a, b + 0.014, '1.223', fontsize=6, bbox={'facecolor': 'crimson', 'alpha': 0.1, 'pad': 1},\n",
        "                        verticalalignment='bottom', horizontalalignment='center')\n",
        "            else:\n",
        "                if idx == 2:\n",
        "                    ax.text(a, b - 0.035, str(b), fontsize=6,\n",
        "                            bbox={'facecolor': 'crimson', 'alpha': 0.1, 'pad': 1},\n",
        "                            verticalalignment='bottom', horizontalalignment='center')\n",
        "\n",
        "    plt.legend(frameon=False)\n",
        "    plt.savefig(\"Model2 Loss.png\", dpi=192)\n",
        "    fig.set_dpi(100)\n",
        "    plt.show()\n",
        "\n",
        "plot_loss_graph(\"Model 2: Loss over Epochs\")"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 600x400 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Xc6kPPi3SLD4",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 404
        },
        "outputId": "b0aae64b-9c10-4b9d-f469-9ec6b0b3f24b"
      },
      "source": [
        "def plot_score_graph(title):\n",
        "    \n",
        "    fig = plt.figure()\n",
        "    ax = fig.add_subplot(111)\n",
        "    ax.set_title(title)\n",
        "    ax.set_xlabel(\"Epochs\")\n",
        "    ax.set_ylabel(\"Loss\")\n",
        "\n",
        "    ticks = [1, 2, 3]\n",
        "    EM = [63.429,64.153,67.354]\n",
        "    F1 = [67.608,68.514, 71.761]\n",
        "\n",
        "    plt.xticks(ticks)\n",
        "\n",
        "    plt.plot(ticks, EM, label='EM', color='brown')\n",
        "    plt.plot(ticks, F1, label='F1', color='salmon')\n",
        "\n",
        "    for idx, (a, b) in enumerate(zip(ticks, EM)):\n",
        "\n",
        "        if idx == 0:\n",
        "            ax.text(a, b + 0.2, str(b), fontsize=6, bbox={'facecolor': 'teal', 'alpha': 0.1, 'pad': 1},\n",
        "                    verticalalignment='bottom', horizontalalignment='center')\n",
        "        else:\n",
        "            if idx == 1:\n",
        "                ax.text(a , b - 0.45, str(b), fontsize=6, bbox={'facecolor': 'teal', 'alpha': 0.1, 'pad': 1},\n",
        "                        verticalalignment='bottom', horizontalalignment='center')\n",
        "            else:\n",
        "                if idx == 2:\n",
        "                    ax.text(a, b + 0.15, str(b), fontsize=6,\n",
        "                            bbox={'facecolor': 'teal', 'alpha': 0.1, 'pad': 1},\n",
        "                            verticalalignment='bottom', horizontalalignment='center')\n",
        "\n",
        "    for idx, (a, b) in enumerate(zip(ticks, F1)):\n",
        "\n",
        "        if idx == 0:\n",
        "            ax.text(a - 0.005, b - 0.4, str(b), fontsize=6, bbox={'facecolor': 'orange', 'alpha': 0.1, 'pad': 1},\n",
        "                    verticalalignment='bottom', horizontalalignment='center')\n",
        "        else:\n",
        "            if idx == 1:\n",
        "                ax.text(a, b - 0.45, str(b), fontsize=6, bbox={'facecolor': 'orange', 'alpha': 0.1, 'pad': 1},\n",
        "                        verticalalignment='bottom', horizontalalignment='center')\n",
        "            else:\n",
        "                if idx == 2:\n",
        "                    ax.text(a-0.11, b+0.05, str(b), fontsize=6,\n",
        "                            bbox={'facecolor': 'orange', 'alpha': 0.1, 'pad': 1},\n",
        "                            verticalalignment='bottom', horizontalalignment='center')\n",
        "\n",
        "    plt.legend(frameon=False)\n",
        "    plt.savefig(\"Model2 Score.png\", dpi=192)\n",
        "    fig.set_dpi(100)\n",
        "    plt.show()\n",
        "\n",
        "plot_score_graph(\"Model 2: EM and F1 Score over Epochs\")"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 600x400 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    }
  ]
}