F# Keras.NET Text Sentiment Analysis

fsharp_keras_net.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "fsharp_keras_colab_v5.ipynb",
      "provenance": []
    },
    "kernelspec": {
      "name": ".net-fsharp",
      "display_name": "F#"
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        },
        "id": "ljaRA9itH5ew",
        "outputId": "fc67caff-352b-47b8-e070-152dcb966169"
      },
      "source": [
        "#r \"nuget: Keras.NET.Mono, 3.7.5\"\r\n",
        "#r \"nuget: Keras.NET, 3.7.5\"\r\n",
        "  \r\n",
        "open System\r\n",
        "open Keras\r\n",
        "open Keras.Layers\r\n",
        "open Keras.Models\r\n",
        "open Keras.Datasets\r\n",
        "open Keras.PreProcessing.sequence\r\n",
        "open Numpy\r\n",
        "open System.Linq\r\n",
        "\r\n",
        "// Load the dataset but only keep the top n words, zero the rest\r\n",
        "let top_words = Nullable 20000\r\n",
        "let max_words = Nullable 200\r\n",
        "\r\n",
        "let struct ( struct (x_train, y_train), struct (x_test, y_test) ) = IMDB.LoadData(num_words=top_words)\r\n",
        "\r\n",
        "let x_train_pad = SequenceUtil.PadSequences (x_train, maxlen=max_words)\r\n",
        "let x_test_pad = SequenceUtil.PadSequences(x_test, maxlen=max_words)\r\n",
        "\r\n",
        "let model = new Sequential()\r\n",
        "new Embedding(top_words.Value, 32, input_length=max_words) |> model.Add\r\n",
        "new Conv1D(filters=32, kernel_size=3, padding=\"same\", activation=\"relu\") |> model.Add\r\n",
        "new MaxPooling1D(pool_size=2) |> model.Add\r\n",
        "new Flatten() |> model.Add\r\n",
        "new Dense(250, activation=\"relu\") |> model.Add\r\n",
        "new Dense(1, activation=\"sigmoid\") |> model.Add\r\n",
        "model.Compile(loss=\"binary_crossentropy\", \r\n",
        "    optimizer=StringOrInstance.op_Implicit(\"adam\"), \r\n",
        "    metrics=[| \"accuracy\" |])\r\n",
        "model.Summary()"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "Installed package Keras.NET version 3.7.5"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "Installed package Keras.NET.Mono version 3.7.5"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "16LBQtk2Xb9O"
      },
      "source": [
        "// Fit the model\r\n",
        "model.Fit(x_train_pad, y_train, validation_data=[| x_test_pad; y_test |], \r\n",
        "    epochs=10, batch_size=Nullable 128, verbose=2) "
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "cH8EU2lzA6Vt",
        "outputId": "0017e9aa-c810-4b35-86f4-efc74206b151"
      },
      "source": [
        "// Final evaluation of the model\r\n",
        "let scores = model.Evaluate(x_test_pad, y_test, verbose=0)\r\n",
        "scores.[1] * 100. |> printfn \"Accuracy: %A\""
      ],
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Accuracy: 87.23999858\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ErWaelIFU8__"
      },
      "source": [
        "let json = model.ToJson()\r\n",
        "System.IO.File.WriteAllText(\"/model.json\", json)\r\n",
        "model.SaveWeight(\"/model.h5\")"
      ],
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "m31bULRz7t3N"
      },
      "source": [
        "let loaded_model = BaseModel.ModelFromJson(System.IO.File.ReadAllText(\"/model.json\"))\r\n",
        "loaded_model.LoadWeight(\"/model.h5\")\r\n",
        "\r\n",
        "let testText = \"The movie was very bad\"\r\n",
        "let indexes = IMDB.GetWordIndex()\r\n",
        "\r\n",
        "let words = testText.Split()\r\n",
        "let mutable review = [||]\r\n",
        "for word in words do\r\n",
        "  match indexes.TryGetValue word with\r\n",
        "  | true, value -> value+3 |> review.Append\r\n",
        "  | _           -> review.Append 2\r\n",
        "\r\n",
        "let x = np.array(review) \r\n",
        "let x_shaped = x.reshape(1, x.shape.[0])\r\n",
        "\r\n",
        "let x_pad = SequenceUtil.PadSequences(x_shaped, maxlen=max_words)\r\n",
        "let y = loaded_model.Predict x_pad\r\n",
        "printfn \"y = %A\" y\r\n",
        "let binary = Math.Round(y.[0].asscalar<float>())\r\n",
        "let result = if binary = 0. then \"Negative\" else \"Positive\"\r\n",
        "printfn \"Sentiment for \\\"%s\\\": %s\" testText result"
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}