seq2seq.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "source": [
        "import numpy as np\n",
        "import tensorflow as tf\n",
        "import matplotlib.pyplot as plt\n",
        "\n",
        "# Read data\n",
        "lines = open(\"input.txt\", \"r\").readlines()\n",
        "input_texts, target_texts = [], []\n",
        "for line in lines:\n",
        "    input_text, target_text = line.split('*')\n",
        "    input_texts.append(input_text.strip())\n",
        "    target_texts.append(target_text.strip())\n",
        "\n",
        "# Create a vocabulary\n",
        "all_texts = ' '.join(input_texts + target_texts)\n",
        "vocab = sorted(set(all_texts))\n",
        "vocab_size = len(vocab)\n",
        "\n",
        "# Map chars to integers and vice versa\n",
        "char2idx = {char: idx for idx, char in enumerate(vocab)}\n",
        "idx2char = {idx: char for char, idx in char2idx.items()}\n",
        "\n",
        "# Convert texts to one-hot encoded\n",
        "def text_to_one_hot(text):\n",
        "    vector = np.zeros((len(text), vocab_size))\n",
        "    for i, char in enumerate(text):\n",
        "        vector[i, char2idx[char]] = 1\n",
        "    return vector\n",
        "\n",
        "input_data = [text_to_one_hot(text) for text in input_texts]\n",
        "target_data = [text_to_one_hot(text) for text in target_texts]\n",
        "\n",
        "# Pad data for consistent shape\n",
        "max_sequence_len = max(max([len(text) for text in input_texts]), max([len(text) for text in target_texts]))\n",
        "\n",
        "input_data = tf.keras.preprocessing.sequence.pad_sequences(input_data, maxlen=max_sequence_len, padding='post')\n",
        "target_data = tf.keras.preprocessing.sequence.pad_sequences(target_data, maxlen=max_sequence_len, padding='post')\n",
        "\n",
        "# Create the model\n",
        "model = tf.keras.Sequential([\n",
        "    tf.keras.layers.Flatten(input_shape=(max_sequence_len, vocab_size)),\n",
        "    tf.keras.layers.Dense(256, activation='relu'),\n",
        "    tf.keras.layers.Dense(max_sequence_len * vocab_size, activation='softmax'),\n",
        "    tf.keras.layers.Reshape((max_sequence_len, vocab_size))\n",
        "])\n",
        "\n",
        "# Compile and train the model\n",
        "model.compile(optimizer='adam', loss='categorical_crossentropy')\n",
        "history = model.fit(input_data, target_data, epochs=100)\n",
        "\n",
        "#Plot the training graph\n",
        "plt.plot(history.history['loss'])\n",
        "plt.title('Model Training Loss')\n",
        "plt.ylabel('Loss')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Train'], loc='upper right')\n",
        "plt.show()\n",
        "\n",
        "# Save model weights to CSV\n",
        "for i, weight in enumerate(model.get_weights()):\n",
        "    np.savetxt(f'weight_{i}.csv', weight, delimiter=',')\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "id": "DltLgD7BSHBT",
        "outputId": "91058e65-a774-4d5f-e899-9fd0f268bd17"
      },
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/100\n",
            "9/9 [==============================] - 1s 21ms/step - loss: 2.0369\n",
            "Epoch 2/100\n",
            "9/9 [==============================] - 0s 20ms/step - loss: 1.8302\n",
            "Epoch 3/100\n",
            "9/9 [==============================] - 0s 19ms/step - loss: 1.4704\n",
            "Epoch 4/100\n",
            "9/9 [==============================] - 0s 20ms/step - loss: 1.3340\n",
            "Epoch 5/100\n",
            "9/9 [==============================] - 0s 22ms/step - loss: 1.2934\n",
            "Epoch 6/100\n",
            "9/9 [==============================] - 0s 22ms/step - loss: 1.2496\n",
            "Epoch 7/100\n",
            "9/9 [==============================] - 0s 24ms/step - loss: 1.2116\n",
            "Epoch 8/100\n",
            "9/9 [==============================] - 0s 23ms/step - loss: 1.1675\n",
            "Epoch 9/100\n",
            "9/9 [==============================] - 0s 26ms/step - loss: 1.1185\n",
            "Epoch 10/100\n",
            "9/9 [==============================] - 0s 26ms/step - loss: 1.0701\n",
            "Epoch 11/100\n",
            "9/9 [==============================] - 0s 26ms/step - loss: 1.0191\n",
            "Epoch 12/100\n",
            "9/9 [==============================] - 0s 21ms/step - loss: 0.9682\n",
            "Epoch 13/100\n",
            "9/9 [==============================] - 0s 22ms/step - loss: 0.9194\n",
            "Epoch 14/100\n",
            "9/9 [==============================] - 0s 21ms/step - loss: 0.8725\n",
            "Epoch 15/100\n",
            "9/9 [==============================] - 0s 21ms/step - loss: 0.8271\n",
            "Epoch 16/100\n",
            "9/9 [==============================] - 0s 20ms/step - loss: 0.7847\n",
            "Epoch 17/100\n",
            "9/9 [==============================] - 0s 26ms/step - loss: 0.7439\n",
            "Epoch 18/100\n",
            "9/9 [==============================] - 0s 29ms/step - loss: 0.7076\n",
            "Epoch 19/100\n",
            "9/9 [==============================] - 0s 30ms/step - loss: 0.6680\n",
            "Epoch 20/100\n",
            "9/9 [==============================] - 0s 24ms/step - loss: 0.6329\n",
            "Epoch 21/100\n",
            "9/9 [==============================] - 0s 27ms/step - loss: 0.5975\n",
            "Epoch 22/100\n",
            "9/9 [==============================] - 0s 27ms/step - loss: 0.5643\n",
            "Epoch 23/100\n",
            "9/9 [==============================] - 0s 27ms/step - loss: 0.5334\n",
            "Epoch 24/100\n",
            "9/9 [==============================] - 0s 25ms/step - loss: 0.5030\n",
            "Epoch 25/100\n",
            "9/9 [==============================] - 0s 24ms/step - loss: 0.4734\n",
            "Epoch 26/100\n",
            "9/9 [==============================] - 0s 24ms/step - loss: 0.4472\n",
            "Epoch 27/100\n",
            "9/9 [==============================] - 0s 26ms/step - loss: 0.4209\n",
            "Epoch 28/100\n",
            "9/9 [==============================] - 0s 27ms/step - loss: 0.3966\n",
            "Epoch 29/100\n",
            "9/9 [==============================] - 0s 27ms/step - loss: 0.3728\n",
            "Epoch 30/100\n",
            "9/9 [==============================] - 0s 26ms/step - loss: 0.3518\n",
            "Epoch 31/100\n",
            "9/9 [==============================] - 0s 25ms/step - loss: 0.3318\n",
            "Epoch 32/100\n",
            "9/9 [==============================] - 0s 35ms/step - loss: 0.3128\n",
            "Epoch 33/100\n",
            "9/9 [==============================] - 0s 30ms/step - loss: 0.2937\n",
            "Epoch 34/100\n",
            "9/9 [==============================] - 0s 28ms/step - loss: 0.2751\n",
            "Epoch 35/100\n",
            "9/9 [==============================] - 0s 28ms/step - loss: 0.2598\n",
            "Epoch 36/100\n",
            "9/9 [==============================] - 0s 30ms/step - loss: 0.2440\n",
            "Epoch 37/100\n",
            "9/9 [==============================] - 0s 27ms/step - loss: 0.2286\n",
            "Epoch 38/100\n",
            "9/9 [==============================] - 0s 31ms/step - loss: 0.2159\n",
            "Epoch 39/100\n",
            "9/9 [==============================] - 0s 30ms/step - loss: 0.2023\n",
            "Epoch 40/100\n",
            "9/9 [==============================] - 0s 29ms/step - loss: 0.1901\n",
            "Epoch 41/100\n",
            "9/9 [==============================] - 0s 33ms/step - loss: 0.1783\n",
            "Epoch 42/100\n",
            "9/9 [==============================] - 0s 24ms/step - loss: 0.1675\n",
            "Epoch 43/100\n",
            "9/9 [==============================] - 0s 21ms/step - loss: 0.1570\n",
            "Epoch 44/100\n",
            "9/9 [==============================] - 0s 21ms/step - loss: 0.1480\n",
            "Epoch 45/100\n",
            "9/9 [==============================] - 0s 20ms/step - loss: 0.1400\n",
            "Epoch 46/100\n",
            "9/9 [==============================] - 0s 20ms/step - loss: 0.1319\n",
            "Epoch 47/100\n",
            "9/9 [==============================] - 0s 21ms/step - loss: 0.1242\n",
            "Epoch 48/100\n",
            "9/9 [==============================] - 0s 21ms/step - loss: 0.1169\n",
            "Epoch 49/100\n",
            "9/9 [==============================] - 0s 22ms/step - loss: 0.1113\n",
            "Epoch 50/100\n",
            "9/9 [==============================] - 0s 19ms/step - loss: 0.1044\n",
            "Epoch 51/100\n",
            "9/9 [==============================] - 0s 26ms/step - loss: 0.0988\n",
            "Epoch 52/100\n",
            "9/9 [==============================] - 0s 24ms/step - loss: 0.0938\n",
            "Epoch 53/100\n",
            "9/9 [==============================] - 0s 26ms/step - loss: 0.0883\n",
            "Epoch 54/100\n",
            "9/9 [==============================] - 0s 28ms/step - loss: 0.0836\n",
            "Epoch 55/100\n",
            "9/9 [==============================] - 0s 25ms/step - loss: 0.0798\n",
            "Epoch 56/100\n",
            "9/9 [==============================] - 0s 27ms/step - loss: 0.0753\n",
            "Epoch 57/100\n",
            "9/9 [==============================] - 0s 26ms/step - loss: 0.0714\n",
            "Epoch 58/100\n",
            "9/9 [==============================] - 0s 27ms/step - loss: 0.0701\n",
            "Epoch 59/100\n",
            "9/9 [==============================] - 0s 22ms/step - loss: 0.0663\n",
            "Epoch 60/100\n",
            "9/9 [==============================] - 0s 21ms/step - loss: 0.0624\n",
            "Epoch 61/100\n",
            "9/9 [==============================] - 0s 20ms/step - loss: 0.0583\n",
            "Epoch 62/100\n",
            "9/9 [==============================] - 0s 21ms/step - loss: 0.0555\n",
            "Epoch 63/100\n",
            "9/9 [==============================] - 0s 22ms/step - loss: 0.0529\n",
            "Epoch 64/100\n",
            "9/9 [==============================] - 0s 23ms/step - loss: 0.0520\n",
            "Epoch 65/100\n",
            "9/9 [==============================] - 0s 26ms/step - loss: 0.0488\n",
            "Epoch 66/100\n",
            "9/9 [==============================] - 0s 28ms/step - loss: 0.0459\n",
            "Epoch 67/100\n",
            "9/9 [==============================] - 0s 27ms/step - loss: 0.0435\n",
            "Epoch 68/100\n",
            "9/9 [==============================] - 0s 31ms/step - loss: 0.0418\n",
            "Epoch 69/100\n",
            "9/9 [==============================] - 0s 23ms/step - loss: 0.0409\n",
            "Epoch 70/100\n",
            "9/9 [==============================] - 0s 28ms/step - loss: 0.0396\n",
            "Epoch 71/100\n",
            "9/9 [==============================] - 0s 27ms/step - loss: 0.0374\n",
            "Epoch 72/100\n",
            "9/9 [==============================] - 0s 28ms/step - loss: 0.0360\n",
            "Epoch 73/100\n",
            "9/9 [==============================] - 0s 19ms/step - loss: 0.0343\n",
            "Epoch 74/100\n",
            "9/9 [==============================] - 0s 19ms/step - loss: 0.0335\n",
            "Epoch 75/100\n",
            "9/9 [==============================] - 0s 18ms/step - loss: 0.0318\n",
            "Epoch 76/100\n",
            "9/9 [==============================] - 0s 20ms/step - loss: 0.0307\n",
            "Epoch 77/100\n",
            "9/9 [==============================] - 0s 20ms/step - loss: 0.0294\n",
            "Epoch 78/100\n",
            "9/9 [==============================] - 0s 25ms/step - loss: 0.0282\n",
            "Epoch 79/100\n",
            "9/9 [==============================] - 0s 23ms/step - loss: 0.0271\n",
            "Epoch 80/100\n",
            "9/9 [==============================] - 0s 26ms/step - loss: 0.0262\n",
            "Epoch 81/100\n",
            "9/9 [==============================] - 0s 26ms/step - loss: 0.0255\n",
            "Epoch 82/100\n",
            "9/9 [==============================] - 0s 25ms/step - loss: 0.0247\n",
            "Epoch 83/100\n",
            "9/9 [==============================] - 0s 24ms/step - loss: 0.0239\n",
            "Epoch 84/100\n",
            "9/9 [==============================] - 0s 25ms/step - loss: 0.0230\n",
            "Epoch 85/100\n",
            "9/9 [==============================] - 0s 26ms/step - loss: 0.0222\n",
            "Epoch 86/100\n",
            "9/9 [==============================] - 0s 27ms/step - loss: 0.0214\n",
            "Epoch 87/100\n",
            "9/9 [==============================] - 0s 21ms/step - loss: 0.0208\n",
            "Epoch 88/100\n",
            "9/9 [==============================] - 0s 28ms/step - loss: 0.0203\n",
            "Epoch 89/100\n",
            "9/9 [==============================] - 0s 29ms/step - loss: 0.0197\n",
            "Epoch 90/100\n",
            "9/9 [==============================] - 0s 29ms/step - loss: 0.0190\n",
            "Epoch 91/100\n",
            "9/9 [==============================] - 0s 33ms/step - loss: 0.0185\n",
            "Epoch 92/100\n",
            "9/9 [==============================] - 0s 36ms/step - loss: 0.0179\n",
            "Epoch 93/100\n",
            "9/9 [==============================] - 0s 33ms/step - loss: 0.0175\n",
            "Epoch 94/100\n",
            "9/9 [==============================] - 0s 35ms/step - loss: 0.0172\n",
            "Epoch 95/100\n",
            "9/9 [==============================] - 0s 35ms/step - loss: 0.0167\n",
            "Epoch 96/100\n",
            "9/9 [==============================] - 0s 35ms/step - loss: 0.0162\n",
            "Epoch 97/100\n",
            "9/9 [==============================] - 0s 36ms/step - loss: 0.0157\n",
            "Epoch 98/100\n",
            "9/9 [==============================] - 0s 26ms/step - loss: 0.0153\n",
            "Epoch 99/100\n",
            "9/9 [==============================] - 0s 23ms/step - loss: 0.0151\n",
            "Epoch 100/100\n",
            "9/9 [==============================] - 0s 27ms/step - loss: 0.0146\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import numpy as np\n",
        "\n",
        "# Load data for vocab and other parameters\n",
        "lines = open(\"input.txt\", \"r\").readlines()\n",
        "input_texts, target_texts = [], []\n",
        "for line in lines:\n",
        "    input_text, target_text = line.split('*')\n",
        "    input_texts.append(input_text.strip())\n",
        "    target_texts.append(target_text.strip())\n",
        "\n",
        "# Create a vocabulary\n",
        "all_texts = ' '.join(input_texts + target_texts)\n",
        "vocab = sorted(set(all_texts))\n",
        "vocab_size = len(vocab)\n",
        "\n",
        "# Map chars to integers and vice versa\n",
        "char2idx = {char: idx for idx, char in enumerate(vocab)}\n",
        "idx2char = {idx: char for char, idx in char2idx.items()}\n",
        "\n",
        "max_sequence_len = max(max([len(text) for text in input_texts]), max([len(text) for text in target_texts]))\n",
        "\n",
        "# Convert texts to one-hot encoded\n",
        "def text_to_one_hot(text):\n",
        "    vector = np.zeros((len(text), vocab_size))\n",
        "    for i, char in enumerate(text):\n",
        "        vector[i, char2idx[char]] = 1\n",
        "    return vector\n",
        "\n",
        "def softmax(x):\n",
        "    e_x = np.exp(x - np.max(x))\n",
        "    return e_x / e_x.sum(axis=-1, keepdims=True)\n",
        "\n",
        "def relu(x):\n",
        "    return np.maximum(0, x)\n",
        "\n",
        "def matrix_multiply(A, B):\n",
        "    return np.matmul(A, B)\n",
        "\n",
        "def text_to_one_hot(text):\n",
        "    vector = np.zeros((len(text), vocab_size))\n",
        "    for i, char in enumerate(text):\n",
        "        vector[i, char2idx[char]] = 1\n",
        "    return vector\n",
        "\n",
        "# Load weights\n",
        "weights = [np.loadtxt(f'weight_{i}.csv', delimiter=',') for i in range(4)]\n",
        "\n",
        "input_text = \"hello world.\"\n",
        "input_data = text_to_one_hot(input_text)\n",
        "input_data = np.pad(input_data, ((0, max_sequence_len - len(input_text)), (0, 0)))\n",
        "\n",
        "one_hot_encoded = text_to_one_hot(input_text)\n",
        "flattened_one_hot = one_hot_encoded.reshape(-1)\n",
        "\n",
        "# Save to CSV\n",
        "np.savetxt(\"input_one_hot_flattened.csv\", flattened_one_hot, delimiter=\",\")\n",
        "\n",
        "# For your reference:\n",
        "print(f\"Vocab Size: {vocab_size}\")\n",
        "print(f\"Max Text Length: {max_sequence_len}\")\n",
        "\n",
        "# Inference\n",
        "flattened = input_data.reshape(1, -1)\n",
        "dense1 = relu(matrix_multiply(flattened, weights[0]) + weights[1])\n",
        "output = softmax(matrix_multiply(dense1, weights[2]) + weights[3])\n",
        "output = output.reshape(max_sequence_len, vocab_size)\n",
        "\n",
        "translated_chars = [idx2char[np.argmax(vec)] for vec in output]\n",
        "translated_text = ''.join(translated_chars).strip()\n",
        "print(translated_text)\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "SFdZLBrASlME",
        "outputId": "a24ef3cf-e626-4649-96da-3b1523d27fe6"
      },
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Vocab Size: 56\n",
            "Max Text Length: 52\n",
            "Nous ailonrens  usique.aes..nt.me dn r.nu.nnée.gne..\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "input_text = \"her cat had work\"\n",
        "input_data = text_to_one_hot(input_text)\n",
        "input_data = np.pad(input_data, ((0, max_sequence_len - len(input_text)), (0, 0)))\n",
        "\n",
        "# Inference\n",
        "flattened = input_data.reshape(1, -1)\n",
        "dense1 = relu(matrix_multiply(flattened, weights[0]) + weights[1])\n",
        "output = softmax(matrix_multiply(dense1, weights[2]) + weights[3])\n",
        "output = output.reshape(max_sequence_len, vocab_size)\n",
        "\n",
        "translated_chars = [idx2char[np.argmax(vec)] for vec in output]\n",
        "translated_text = ''.join(translated_chars).strip()\n",
        "print(translated_text)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "G79xPpm_X1XV",
        "outputId": "ba99c7a2-1cf6-44ce-ca09-809475879bf9"
      },
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Le sett no eounelr deuo..rne ...e enus.eu.e.s..vae..\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "SkKHl2UoX66R"
      },
      "execution_count": null,
      "outputs": []
    }
  ]
}