Audio_DeiT_Training.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "a9031b9e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "47ecaf59b46e4a1ea86d28692a603ddc",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Resolving data files:   0%|          | 0/8733 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using custom data configuration default-056bb73702904399\n",
      "Reusing dataset image_folder (/home/ubuntu/.cache/huggingface/datasets/image_folder/default-056bb73702904399/0.0.0/48efdc62d40223daee675ca093d163bcb6cb0b7d7f93eb25aebf5edca72dc597)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "5500 2620 612\n"
     ]
    }
   ],
   "source": [
    "from datasets import load_dataset\n",
    "\n",
    "train_ds = load_dataset(\"imagefolder\", data_dir=\"audio_transformer/data/train\", split=\"train\")\n",
    "\n",
    "\n",
    "# split up training into training + validation\n",
    "splits = train_ds.train_test_split(test_size=0.3)\n",
    "train_val_ds = splits['train']\n",
    "test_ds = splits['test']\n",
    "\n",
    "splits = train_val_ds.train_test_split(test_size=0.1)\n",
    "train_ds = splits['train']\n",
    "val_ds = splits['test']\n",
    "\n",
    "\n",
    "print(len(train_ds),len(test_ds),len(val_ds))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "ce264938",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Dataset({\n",
       "    features: ['image', 'classID', 'class'],\n",
       "    num_rows: 5500\n",
       "})"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_ds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "77cced10",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{0: 'air_conditioner',\n",
       " 1: 'car_horn',\n",
       " 2: 'children_playing',\n",
       " 3: 'dog_bark',\n",
       " 4: 'drilling',\n",
       " 5: 'engine_idling',\n",
       " 6: 'gun_shot',\n",
       " 7: 'jackhammer',\n",
       " 8: 'siren',\n",
       " 9: 'street_music'}"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import json\n",
    "\n",
    "with open('audio_transformer/data/label2id.json') as f:\n",
    "    label2id = json.load(f)\n",
    "\n",
    "with open('audio_transformer/data/id2label.json') as f:\n",
    "    id2label = json.load(f)\n",
    "\n",
    "id2label = {int(key):value for key,value in id2label.items()}\n",
    "id2label"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "ccd870a8",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "https://huggingface.co/facebook/deit-base-distilled-patch16-224/resolve/main/preprocessor_config.json not found in cache or force_download set to True, downloading to /home/ubuntu/.cache/huggingface/transformers/tmpytstgn21\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "28e60e1c2c7742f08aadbafa2d6294ed",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading:   0%|          | 0.00/287 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "storing https://huggingface.co/facebook/deit-base-distilled-patch16-224/resolve/main/preprocessor_config.json in cache at /home/ubuntu/.cache/huggingface/transformers/a3c9a583827005951a2f3c81e5d6f3bda9f52700892e2acb36110a8b5a26ef1b.29240c085a2cf1c33c4279bd14ec75a73811c11a5873eab1f004bde6fe6309f7\n",
      "creating metadata file for /home/ubuntu/.cache/huggingface/transformers/a3c9a583827005951a2f3c81e5d6f3bda9f52700892e2acb36110a8b5a26ef1b.29240c085a2cf1c33c4279bd14ec75a73811c11a5873eab1f004bde6fe6309f7\n",
      "loading feature extractor configuration file https://huggingface.co/facebook/deit-base-distilled-patch16-224/resolve/main/preprocessor_config.json from cache at /home/ubuntu/.cache/huggingface/transformers/a3c9a583827005951a2f3c81e5d6f3bda9f52700892e2acb36110a8b5a26ef1b.29240c085a2cf1c33c4279bd14ec75a73811c11a5873eab1f004bde6fe6309f7\n",
      "Feature extractor DeiTFeatureExtractor {\n",
      "  \"crop_size\": 224,\n",
      "  \"do_center_crop\": true,\n",
      "  \"do_normalize\": true,\n",
      "  \"do_resize\": true,\n",
      "  \"feature_extractor_type\": \"DeiTFeatureExtractor\",\n",
      "  \"image_mean\": [\n",
      "    0.485,\n",
      "    0.456,\n",
      "    0.406\n",
      "  ],\n",
      "  \"image_std\": [\n",
      "    0.229,\n",
      "    0.224,\n",
      "    0.225\n",
      "  ],\n",
      "  \"resample\": 3,\n",
      "  \"size\": 256\n",
      "}\n",
      "\n"
     ]
    }
   ],
   "source": [
    "from transformers import DeiTFeatureExtractor\n",
    "\n",
    "feature_extractor = DeiTFeatureExtractor.from_pretrained(\"facebook/deit-base-distilled-patch16-224\")\n",
    "#  model = SwinForImageClassification.from_pretrained(\"microsoft/swin-tiny-patch4-window7-224\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "f871b926",
   "metadata": {},
   "outputs": [],
   "source": [
    "from torchvision.transforms import (CenterCrop, \n",
    "                                    Compose, \n",
    "                                    Normalize, \n",
    "                                    RandomHorizontalFlip,\n",
    "                                    RandomResizedCrop, \n",
    "                                    Resize, \n",
    "                                    ToTensor)\n",
    "\n",
    "normalize = Normalize(mean=feature_extractor.image_mean, std=feature_extractor.image_std)\n",
    "_train_transforms = Compose(\n",
    "        [\n",
    "            RandomResizedCrop(feature_extractor.size),\n",
    "            RandomHorizontalFlip(),\n",
    "            ToTensor(),\n",
    "            normalize,\n",
    "        ]\n",
    "    )\n",
    "\n",
    "_val_transforms = Compose(\n",
    "        [\n",
    "            Resize(feature_extractor.size),\n",
    "            CenterCrop(feature_extractor.size),\n",
    "            ToTensor(),\n",
    "            normalize,\n",
    "        ]\n",
    "    )\n",
    "\n",
    "def train_transforms(examples):\n",
    "    examples['pixel_values'] = [_train_transforms(image.convert(\"RGB\")) for image in examples['image']]\n",
    "    return examples\n",
    "\n",
    "def val_transforms(examples):\n",
    "    examples['pixel_values'] = [_val_transforms(image.convert(\"RGB\")) for image in examples['image']]\n",
    "    return examples"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "e692871b",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Set the transforms\n",
    "train_ds.set_transform(train_transforms)\n",
    "val_ds.set_transform(val_transforms)\n",
    "test_ds.set_transform(val_transforms)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "e2e5a994",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'image': [<PIL.PngImagePlugin.PngImageFile image mode=RGBA size=496x369 at 0x7EFD949DF4E0>,\n",
       "  <PIL.PngImagePlugin.PngImageFile image mode=RGBA size=496x369 at 0x7EFD949DF470>],\n",
       " 'classID': [7, 3],\n",
       " 'class': ['jackhammer', 'dog_bark'],\n",
       " 'pixel_values': [tensor([[[ 1.1187,  1.1187,  1.1187,  ...,  1.0673,  0.7077,  0.7077],\n",
       "           [ 1.0844,  1.0844,  1.0844,  ...,  1.0331,  0.7248,  0.7248],\n",
       "           [ 0.9303,  0.9303,  0.9303,  ...,  0.8618,  0.8104,  0.8104],\n",
       "           ...,\n",
       "           [ 2.1119,  2.1119,  2.1119,  ...,  2.1804,  2.0948,  2.0948],\n",
       "           [ 2.1119,  2.1119,  2.1119,  ...,  2.2147,  2.1633,  2.1633],\n",
       "           [ 2.1119,  2.1119,  2.1119,  ...,  2.2147,  2.1633,  2.1633]],\n",
       "  \n",
       "          [[-1.0378, -1.0378, -1.0378,  ..., -1.0553, -1.1779, -1.1779],\n",
       "           [-1.0553, -1.0553, -1.0553,  ..., -1.0728, -1.1779, -1.1779],\n",
       "           [-1.1078, -1.1078, -1.1078,  ..., -1.1253, -1.1429, -1.1429],\n",
       "           ...,\n",
       "           [-0.0399, -0.0399, -0.0399,  ...,  0.4678,  0.0476,  0.0476],\n",
       "           [-0.0749, -0.0749, -0.0749,  ...,  0.6779,  0.2402,  0.2227],\n",
       "           [-0.0749, -0.0749, -0.0749,  ...,  0.6779,  0.2402,  0.2227]],\n",
       "  \n",
       "          [[ 0.2696,  0.2696,  0.2696,  ...,  0.2871,  0.3916,  0.3916],\n",
       "           [ 0.2871,  0.2871,  0.2871,  ...,  0.3045,  0.3916,  0.3916],\n",
       "           [ 0.3393,  0.3393,  0.3393,  ...,  0.3393,  0.3568,  0.3568],\n",
       "           ...,\n",
       "           [-0.2010, -0.2010, -0.2010,  ..., -0.0092, -0.1487, -0.1487],\n",
       "           [-0.2010, -0.2010, -0.2010,  ...,  0.0605, -0.1487, -0.1487],\n",
       "           [-0.2010, -0.2010, -0.2010,  ...,  0.0605, -0.1487, -0.1487]]]),\n",
       "  tensor([[[ 0.3481,  0.3481,  0.3481,  ...,  0.4851,  0.4851,  0.1939],\n",
       "           [ 0.3481,  0.3481,  0.3481,  ...,  0.4851,  0.4851,  0.1939],\n",
       "           [ 0.3823,  0.3138,  0.3138,  ...,  0.5878,  0.5878,  0.3481],\n",
       "           ...,\n",
       "           [ 2.2318,  2.2318,  2.2318,  ...,  2.2147,  2.2147,  2.1290],\n",
       "           [ 2.2318,  2.2318,  2.2318,  ...,  2.1975,  2.1975,  2.1975],\n",
       "           [ 2.2318,  2.2318,  2.2318,  ...,  2.1975,  2.1975,  2.2147]],\n",
       "  \n",
       "          [[-1.3004, -1.3004, -1.3004,  ..., -1.2479, -1.2479, -1.3529],\n",
       "           [-1.3004, -1.3004, -1.3004,  ..., -1.2479, -1.2479, -1.3529],\n",
       "           [-1.2829, -1.3004, -1.3004,  ..., -1.2129, -1.2129, -1.3004],\n",
       "           ...,\n",
       "           [ 0.9405,  1.1506,  1.1506,  ...,  0.5903,  0.5903,  0.0301],\n",
       "           [ 0.8880,  1.1856,  1.1856,  ...,  0.4503,  0.4503,  0.4853],\n",
       "           [ 0.8704,  1.2031,  1.2031,  ...,  0.4153,  0.4153,  0.6078]],\n",
       "  \n",
       "          [[ 0.4439,  0.4439,  0.4439,  ...,  0.4265,  0.4265,  0.4439],\n",
       "           [ 0.4439,  0.4439,  0.4439,  ...,  0.4265,  0.4265,  0.4439],\n",
       "           [ 0.4439,  0.4439,  0.4439,  ...,  0.4091,  0.4091,  0.4265],\n",
       "           ...,\n",
       "           [ 0.2173,  0.3568,  0.3568,  ...,  0.0082,  0.0082, -0.1835],\n",
       "           [ 0.1825,  0.3916,  0.3916,  ..., -0.0615, -0.0615, -0.0267],\n",
       "           [ 0.1825,  0.4091,  0.4091,  ..., -0.0790, -0.0790,  0.0256]]])]}"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_ds[:2]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "65c25ad8",
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch.utils.data import DataLoader\n",
    "import torch\n",
    "\n",
    "def collate_fn(examples):\n",
    "    pixel_values = torch.stack([example[\"pixel_values\"] for example in examples])\n",
    "    labels = torch.tensor([example[\"classID\"] for example in examples])\n",
    "    return {\"pixel_values\": pixel_values, \"labels\": labels}\n",
    "\n",
    "train_dataloader = DataLoader(train_ds, collate_fn=collate_fn, batch_size=4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "4921f1c6",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "pixel_values torch.Size([4, 3, 224, 224])\n",
      "labels torch.Size([4])\n"
     ]
    }
   ],
   "source": [
    "batch = next(iter(train_dataloader))\n",
    "for k,v in batch.items():\n",
    "  if isinstance(v, torch.Tensor):\n",
    "    print(k, v.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "26dd9088",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "https://huggingface.co/facebook/deit-base-distilled-patch16-224/resolve/main/config.json not found in cache or force_download set to True, downloading to /home/ubuntu/.cache/huggingface/transformers/tmp66s4h8fp\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "05269f154e464c9ea4c109b730b59986",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading:   0%|          | 0.00/68.0k [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "storing https://huggingface.co/facebook/deit-base-distilled-patch16-224/resolve/main/config.json in cache at /home/ubuntu/.cache/huggingface/transformers/714ee60a09194779d2acdbf17d75f0cca867f496a85550c69b29fd6f66c8b4fa.a46ddd6a2d7a28e819ee5dfa80b7ccfbcb74b937148567c84d918781d35b3721\n",
      "creating metadata file for /home/ubuntu/.cache/huggingface/transformers/714ee60a09194779d2acdbf17d75f0cca867f496a85550c69b29fd6f66c8b4fa.a46ddd6a2d7a28e819ee5dfa80b7ccfbcb74b937148567c84d918781d35b3721\n",
      "loading configuration file https://huggingface.co/facebook/deit-base-distilled-patch16-224/resolve/main/config.json from cache at /home/ubuntu/.cache/huggingface/transformers/714ee60a09194779d2acdbf17d75f0cca867f496a85550c69b29fd6f66c8b4fa.a46ddd6a2d7a28e819ee5dfa80b7ccfbcb74b937148567c84d918781d35b3721\n",
      "Model config DeiTConfig {\n",
      "  \"architectures\": [\n",
      "    \"DeiTForImageClassificationWithTeacher\"\n",
      "  ],\n",
      "  \"attention_probs_dropout_prob\": 0.0,\n",
      "  \"encoder_stride\": 16,\n",
      "  \"hidden_act\": \"gelu\",\n",
      "  \"hidden_dropout_prob\": 0.0,\n",
      "  \"hidden_size\": 768,\n",
      "  \"id2label\": {\n",
      "    \"0\": \"air_conditioner\",\n",
      "    \"1\": \"car_horn\",\n",
      "    \"2\": \"children_playing\",\n",
      "    \"3\": \"dog_bark\",\n",
      "    \"4\": \"drilling\",\n",
      "    \"5\": \"engine_idling\",\n",
      "    \"6\": \"gun_shot\",\n",
      "    \"7\": \"jackhammer\",\n",
      "    \"8\": \"siren\",\n",
      "    \"9\": \"street_music\"\n",
      "  },\n",
      "  \"image_size\": 224,\n",
      "  \"initializer_range\": 0.02,\n",
      "  \"intermediate_size\": 3072,\n",
      "  \"label2id\": {\n",
      "    \"air_conditioner\": 0,\n",
      "    \"car_horn\": 1,\n",
      "    \"children_playing\": 2,\n",
      "    \"dog_bark\": 3,\n",
      "    \"drilling\": 4,\n",
      "    \"engine_idling\": 5,\n",
      "    \"gun_shot\": 6,\n",
      "    \"jackhammer\": 7,\n",
      "    \"siren\": 8,\n",
      "    \"street_music\": 9\n",
      "  },\n",
      "  \"layer_norm_eps\": 1e-12,\n",
      "  \"model_type\": \"deit\",\n",
      "  \"num_attention_heads\": 12,\n",
      "  \"num_channels\": 3,\n",
      "  \"num_hidden_layers\": 12,\n",
      "  \"patch_size\": 16,\n",
      "  \"qkv_bias\": true,\n",
      "  \"transformers_version\": \"4.18.0\"\n",
      "}\n",
      "\n",
      "https://huggingface.co/facebook/deit-base-distilled-patch16-224/resolve/main/pytorch_model.bin not found in cache or force_download set to True, downloading to /home/ubuntu/.cache/huggingface/transformers/tmpvv5akrzc\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "1060fa750ec7491aa11b73f7ebd786c9",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading:   0%|          | 0.00/333M [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "storing https://huggingface.co/facebook/deit-base-distilled-patch16-224/resolve/main/pytorch_model.bin in cache at /home/ubuntu/.cache/huggingface/transformers/4bd4c7588647eabacf363ecf61f5285633607d10e315048b47ca4bc5b9742f49.89cf9afdd85db0db778890939fde24ed5f60ce453af494c4f12946c6b3efda91\n",
      "creating metadata file for /home/ubuntu/.cache/huggingface/transformers/4bd4c7588647eabacf363ecf61f5285633607d10e315048b47ca4bc5b9742f49.89cf9afdd85db0db778890939fde24ed5f60ce453af494c4f12946c6b3efda91\n",
      "loading weights file https://huggingface.co/facebook/deit-base-distilled-patch16-224/resolve/main/pytorch_model.bin from cache at /home/ubuntu/.cache/huggingface/transformers/4bd4c7588647eabacf363ecf61f5285633607d10e315048b47ca4bc5b9742f49.89cf9afdd85db0db778890939fde24ed5f60ce453af494c4f12946c6b3efda91\n",
      "Some weights of the model checkpoint at facebook/deit-base-distilled-patch16-224 were not used when initializing DeiTForImageClassification: ['cls_classifier.bias', 'distillation_classifier.weight', 'cls_classifier.weight', 'distillation_classifier.bias']\n",
      "- This IS expected if you are initializing DeiTForImageClassification from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).\n",
      "- This IS NOT expected if you are initializing DeiTForImageClassification from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n",
      "Some weights of DeiTForImageClassification were not initialized from the model checkpoint at facebook/deit-base-distilled-patch16-224 and are newly initialized: ['classifier.weight', 'classifier.bias']\n",
      "You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.\n"
     ]
    }
   ],
   "source": [
    "from transformers import DeiTForImageClassification\n",
    "model = DeiTForImageClassification.from_pretrained(\"facebook/deit-base-distilled-patch16-224\",\n",
    "                                                  num_labels=10,\n",
    "                                                  id2label=id2label,\n",
    "                                                  label2id=label2id,\n",
    "                                                  ignore_mismatched_sizes=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "f76924c6",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "PyTorch: setting up devices\n",
      "The default value for the training argument `--report_to` will change in v5 (from all installed integrations to none). In v5, you will need to use `--report_to all` to get the same behavior as now. You should start updating your code and make this info disappear :-).\n"
     ]
    }
   ],
   "source": [
    "from transformers import TrainingArguments, Trainer\n",
    "\n",
    "metric_name = \"accuracy\"\n",
    "\n",
    "args = TrainingArguments(\n",
    "    f\"test-cifar-10\",\n",
    "    save_strategy=\"epoch\",\n",
    "    evaluation_strategy=\"epoch\",\n",
    "    learning_rate=2e-5,\n",
    "    per_device_train_batch_size=10,\n",
    "    per_device_eval_batch_size=4,\n",
    "    num_train_epochs=5,\n",
    "    weight_decay=0.01,\n",
    "    load_best_model_at_end=True,\n",
    "    metric_for_best_model=metric_name,\n",
    "    logging_dir='logs',\n",
    "    remove_unused_columns=False,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "029166e3",
   "metadata": {},
   "outputs": [],
   "source": [
    "from datasets import load_metric\n",
    "import numpy as np\n",
    "\n",
    "metric = load_metric(\"accuracy\")\n",
    "\n",
    "def compute_metrics(eval_pred):\n",
    "    predictions, labels = eval_pred\n",
    "    predictions = np.argmax(predictions, axis=1)\n",
    "    return metric.compute(predictions=predictions, references=labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "8b9be66f",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "\n",
    "trainer = Trainer(\n",
    "    model,\n",
    "    args,\n",
    "    train_dataset=train_ds,\n",
    "    eval_dataset=val_ds,\n",
    "    data_collator=collate_fn,\n",
    "    compute_metrics=compute_metrics,\n",
    "    tokenizer=feature_extractor,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "29623196",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "      <iframe id=\"tensorboard-frame-1c80317fa3b1799d\" width=\"100%\" height=\"800\" frameborder=\"0\">\n",
       "      </iframe>\n",
       "      <script>\n",
       "        (function() {\n",
       "          const frame = document.getElementById(\"tensorboard-frame-1c80317fa3b1799d\");\n",
       "          const url = new URL(\"/\", window.location);\n",
       "          const port = 6006;\n",
       "          if (port) {\n",
       "            url.port = port;\n",
       "          }\n",
       "          frame.src = url;\n",
       "        })();\n",
       "      </script>\n",
       "    "
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Start tensorboard.\n",
    "%load_ext tensorboard\n",
    "%tensorboard --logdir logs/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "f8ad3fc4",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/ubuntu/anaconda3/envs/aws_neuron_pytorch_p36/lib/python3.6/site-packages/transformers/optimization.py:309: FutureWarning: This implementation of AdamW is deprecated and will be removed in a future version. Use the PyTorch implementation torch.optim.AdamW instead, or set `no_deprecation_warning=True` to disable this warning\n",
      "  FutureWarning,\n",
      "***** Running training *****\n",
      "  Num examples = 5500\n",
      "  Num Epochs = 5\n",
      "  Instantaneous batch size per device = 10\n",
      "  Total train batch size (w. parallel, distributed & accumulation) = 10\n",
      "  Gradient Accumulation steps = 1\n",
      "  Total optimization steps = 2750\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "\n",
       "    <div>\n",
       "      \n",
       "      <progress value='2750' max='2750' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      [2750/2750 1:12:34, Epoch 5/5]\n",
       "    </div>\n",
       "    <table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       " <tr style=\"text-align: left;\">\n",
       "      <th>Epoch</th>\n",
       "      <th>Training Loss</th>\n",
       "      <th>Validation Loss</th>\n",
       "      <th>Accuracy</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.954500</td>\n",
       "      <td>0.466000</td>\n",
       "      <td>0.841503</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.569900</td>\n",
       "      <td>0.275393</td>\n",
       "      <td>0.908497</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.452800</td>\n",
       "      <td>0.233814</td>\n",
       "      <td>0.919935</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.378800</td>\n",
       "      <td>0.164148</td>\n",
       "      <td>0.947712</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>0.320400</td>\n",
       "      <td>0.150808</td>\n",
       "      <td>0.950980</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table><p>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "***** Running Evaluation *****\n",
      "  Num examples = 612\n",
      "  Batch size = 4\n",
      "Saving model checkpoint to test-cifar-10/checkpoint-550\n",
      "Configuration saved in test-cifar-10/checkpoint-550/config.json\n",
      "Model weights saved in test-cifar-10/checkpoint-550/pytorch_model.bin\n",
      "Feature extractor saved in test-cifar-10/checkpoint-550/preprocessor_config.json\n",
      "***** Running Evaluation *****\n",
      "  Num examples = 612\n",
      "  Batch size = 4\n",
      "Saving model checkpoint to test-cifar-10/checkpoint-1100\n",
      "Configuration saved in test-cifar-10/checkpoint-1100/config.json\n",
      "Model weights saved in test-cifar-10/checkpoint-1100/pytorch_model.bin\n",
      "Feature extractor saved in test-cifar-10/checkpoint-1100/preprocessor_config.json\n",
      "***** Running Evaluation *****\n",
      "  Num examples = 612\n",
      "  Batch size = 4\n",
      "Saving model checkpoint to test-cifar-10/checkpoint-1650\n",
      "Configuration saved in test-cifar-10/checkpoint-1650/config.json\n",
      "Model weights saved in test-cifar-10/checkpoint-1650/pytorch_model.bin\n",
      "Feature extractor saved in test-cifar-10/checkpoint-1650/preprocessor_config.json\n",
      "***** Running Evaluation *****\n",
      "  Num examples = 612\n",
      "  Batch size = 4\n",
      "Saving model checkpoint to test-cifar-10/checkpoint-2200\n",
      "Configuration saved in test-cifar-10/checkpoint-2200/config.json\n",
      "Model weights saved in test-cifar-10/checkpoint-2200/pytorch_model.bin\n",
      "Feature extractor saved in test-cifar-10/checkpoint-2200/preprocessor_config.json\n",
      "***** Running Evaluation *****\n",
      "  Num examples = 612\n",
      "  Batch size = 4\n",
      "Saving model checkpoint to test-cifar-10/checkpoint-2750\n",
      "Configuration saved in test-cifar-10/checkpoint-2750/config.json\n",
      "Model weights saved in test-cifar-10/checkpoint-2750/pytorch_model.bin\n",
      "Feature extractor saved in test-cifar-10/checkpoint-2750/preprocessor_config.json\n",
      "\n",
      "\n",
      "Training completed. Do not forget to share your model on huggingface.co/models =)\n",
      "\n",
      "\n",
      "Loading best model from test-cifar-10/checkpoint-2750 (score: 0.9509803921568627).\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "TrainOutput(global_step=2750, training_loss=0.5126709400523792, metrics={'train_runtime': 4356.6678, 'train_samples_per_second': 6.312, 'train_steps_per_second': 0.631, 'total_flos': 2.13122066217984e+18, 'train_loss': 0.5126709400523792, 'epoch': 5.0})"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "trainer.train()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "3d64a815",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "***** Running Prediction *****\n",
      "  Num examples = 2620\n",
      "  Batch size = 4\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "\n",
       "    <div>\n",
       "      \n",
       "      <progress value='655' max='655' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      [655/655 02:24]\n",
       "    </div>\n",
       "    "
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "outputs = trainer.predict(test_ds)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "b2a0661b",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'test_loss': 0.1805463284254074, 'test_accuracy': 0.9427480916030534, 'test_runtime': 144.4759, 'test_samples_per_second': 18.135, 'test_steps_per_second': 4.534}\n"
     ]
    }
   ],
   "source": [
    "print(outputs.metrics)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "ed6d78a1",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<sklearn.metrics._plot.confusion_matrix.ConfusionMatrixDisplay at 0x7efe778b2940>"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.metrics import confusion_matrix, ConfusionMatrixDisplay\n",
    "\n",
    "y_true = outputs.label_ids\n",
    "y_pred = outputs.predictions.argmax(1)\n",
    "\n",
    "labels = id2label.values()\n",
    "cm = confusion_matrix(y_true, y_pred)\n",
    "disp = ConfusionMatrixDisplay(confusion_matrix=cm, display_labels=labels)\n",
    "disp.plot(xticks_rotation=45)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "6a2a2b41",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Configuration saved in /home/ubuntu/models/AudioDeiT/config.json\n",
      "Model weights saved in /home/ubuntu/models/AudioDeiT/pytorch_model.bin\n"
     ]
    }
   ],
   "source": [
    "model.save_pretrained(\"/home/ubuntu/models/AudioDeiT\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "768518f3",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Environment (conda_aws_neuron_pytorch_p36)",
   "language": "python",
   "name": "conda_aws_neuron_pytorch_p36"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}