twitter__sentiment_BERT.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "twitter _sentiment_BERT.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "toc_visible": true,
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/maxmatical/TMLS_NLP/blob/master/twitter__sentiment_BERT.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QMgx99NYbpIV",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "%%capture\n",
        "!pip install git+https://github.com/fastai/fastai.git"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YnCNplZFCBp4",
        "colab_type": "code",
        "outputId": "feee3d3c-e538-41ad-fbdb-5792672750c7",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        }
      },
      "source": [
        "# Set up environment and download course-v3\n",
        "!curl -s https://course.fast.ai/setup/colab | bash"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Updating fastai...\n",
            "Done.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zhDI7X-L7lby",
        "colab_type": "code",
        "outputId": "875034b3-8044-4886-bb4c-1f279e086d6e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 819
        }
      },
      "source": [
        "!pip install pretrainedmodels\n",
        "\n",
        "!pip install pytorch-pretrained-bert"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Collecting pretrainedmodels\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/84/0e/be6a0e58447ac16c938799d49bfb5fb7a80ac35e137547fc6cee2c08c4cf/pretrainedmodels-0.7.4.tar.gz (58kB)\n",
            "\r\u001b[K     |█████▋                          | 10kB 15.9MB/s eta 0:00:01\r\u001b[K     |███████████▏                    | 20kB 1.8MB/s eta 0:00:01\r\u001b[K     |████████████████▊               | 30kB 2.6MB/s eta 0:00:01\r\u001b[K     |██████████████████████▎         | 40kB 1.7MB/s eta 0:00:01\r\u001b[K     |███████████████████████████▉    | 51kB 2.1MB/s eta 0:00:01\r\u001b[K     |████████████████████████████████| 61kB 2.4MB/s \n",
            "\u001b[?25hRequirement already satisfied: torch in /usr/local/lib/python3.6/dist-packages (from pretrainedmodels) (1.2.0)\n",
            "Requirement already satisfied: torchvision in /usr/local/lib/python3.6/dist-packages (from pretrainedmodels) (0.4.0)\n",
            "Collecting munch (from pretrainedmodels)\n",
            "  Downloading https://files.pythonhosted.org/packages/68/f4/260ec98ea840757a0da09e0ed8135333d59b8dfebe9752a365b04857660a/munch-2.3.2.tar.gz\n",
            "Requirement already satisfied: tqdm in /usr/local/lib/python3.6/dist-packages (from pretrainedmodels) (4.28.1)\n",
            "Requirement already satisfied: numpy in /usr/local/lib/python3.6/dist-packages (from torch->pretrainedmodels) (1.16.5)\n",
            "Requirement already satisfied: pillow>=4.1.1 in /usr/local/lib/python3.6/dist-packages (from torchvision->pretrainedmodels) (4.3.0)\n",
            "Requirement already satisfied: six in /usr/local/lib/python3.6/dist-packages (from torchvision->pretrainedmodels) (1.12.0)\n",
            "Requirement already satisfied: olefile in /usr/local/lib/python3.6/dist-packages (from pillow>=4.1.1->torchvision->pretrainedmodels) (0.46)\n",
            "Building wheels for collected packages: pretrainedmodels, munch\n",
            "  Building wheel for pretrainedmodels (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Created wheel for pretrainedmodels: filename=pretrainedmodels-0.7.4-cp36-none-any.whl size=60963 sha256=a24dccfa824eba6cd4d346102f4a4c3314cf3b5f220e6f0720609fad8b09c837\n",
            "  Stored in directory: /root/.cache/pip/wheels/69/df/63/62583c096289713f22db605aa2334de5b591d59861a02c2ecd\n",
            "  Building wheel for munch (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Created wheel for munch: filename=munch-2.3.2-py2.py3-none-any.whl size=6614 sha256=45c8b7ef5aecfc48ae5462aeca926836826800e95686752a8e148e0fd689a0e6\n",
            "  Stored in directory: /root/.cache/pip/wheels/db/bf/bc/06a3e1bfe0ab27d2e720ceb3cff3159398d92644c0cec2c125\n",
            "Successfully built pretrainedmodels munch\n",
            "Installing collected packages: munch, pretrainedmodels\n",
            "Successfully installed munch-2.3.2 pretrainedmodels-0.7.4\n",
            "Collecting pytorch-pretrained-bert\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/d7/e0/c08d5553b89973d9a240605b9c12404bcf8227590de62bae27acbcfe076b/pytorch_pretrained_bert-0.6.2-py3-none-any.whl (123kB)\n",
            "\u001b[K     |████████████████████████████████| 133kB 2.8MB/s \n",
            "\u001b[?25hCollecting regex (from pytorch-pretrained-bert)\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/ff/60/d9782c56ceefa76033a00e1f84cd8c586c75e6e7fea2cd45ee8b46a386c5/regex-2019.08.19-cp36-cp36m-manylinux1_x86_64.whl (643kB)\n",
            "\u001b[K     |████████████████████████████████| 645kB 44.2MB/s \n",
            "\u001b[?25hRequirement already satisfied: requests in /usr/local/lib/python3.6/dist-packages (from pytorch-pretrained-bert) (2.21.0)\n",
            "Requirement already satisfied: torch>=0.4.1 in /usr/local/lib/python3.6/dist-packages (from pytorch-pretrained-bert) (1.2.0)\n",
            "Requirement already satisfied: tqdm in /usr/local/lib/python3.6/dist-packages (from pytorch-pretrained-bert) (4.28.1)\n",
            "Requirement already satisfied: numpy in /usr/local/lib/python3.6/dist-packages (from pytorch-pretrained-bert) (1.16.5)\n",
            "Requirement already satisfied: boto3 in /usr/local/lib/python3.6/dist-packages (from pytorch-pretrained-bert) (1.9.243)\n",
            "Requirement already satisfied: chardet<3.1.0,>=3.0.2 in /usr/local/lib/python3.6/dist-packages (from requests->pytorch-pretrained-bert) (3.0.4)\n",
            "Requirement already satisfied: urllib3<1.25,>=1.21.1 in /usr/local/lib/python3.6/dist-packages (from requests->pytorch-pretrained-bert) (1.24.3)\n",
            "Requirement already satisfied: idna<2.9,>=2.5 in /usr/local/lib/python3.6/dist-packages (from requests->pytorch-pretrained-bert) (2.8)\n",
            "Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.6/dist-packages (from requests->pytorch-pretrained-bert) (2019.9.11)\n",
            "Requirement already satisfied: jmespath<1.0.0,>=0.7.1 in /usr/local/lib/python3.6/dist-packages (from boto3->pytorch-pretrained-bert) (0.9.4)\n",
            "Requirement already satisfied: s3transfer<0.3.0,>=0.2.0 in /usr/local/lib/python3.6/dist-packages (from boto3->pytorch-pretrained-bert) (0.2.1)\n",
            "Requirement already satisfied: botocore<1.13.0,>=1.12.243 in /usr/local/lib/python3.6/dist-packages (from boto3->pytorch-pretrained-bert) (1.12.243)\n",
            "Requirement already satisfied: python-dateutil<3.0.0,>=2.1; python_version >= \"2.7\" in /usr/local/lib/python3.6/dist-packages (from botocore<1.13.0,>=1.12.243->boto3->pytorch-pretrained-bert) (2.5.3)\n",
            "Requirement already satisfied: docutils<0.16,>=0.10 in /usr/local/lib/python3.6/dist-packages (from botocore<1.13.0,>=1.12.243->boto3->pytorch-pretrained-bert) (0.15.2)\n",
            "Requirement already satisfied: six>=1.5 in /usr/local/lib/python3.6/dist-packages (from python-dateutil<3.0.0,>=2.1; python_version >= \"2.7\"->botocore<1.13.0,>=1.12.243->boto3->pytorch-pretrained-bert) (1.12.0)\n",
            "Installing collected packages: regex, pytorch-pretrained-bert\n",
            "Successfully installed pytorch-pretrained-bert-0.6.2 regex-2019.8.19\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qKq5YwwRCM5r",
        "colab_type": "code",
        "outputId": "d8d4e3c0-701a-4aa4-de9d-09d8dc7bb59b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 697
        }
      },
      "source": [
        "import numpy as np\n",
        "import pandas as pd\n",
        "from matplotlib import pyplot as plt\n",
        "\n",
        "from fastai import *\n",
        "from fastai.text import *\n",
        "from fastai.utils.show_install import *\n",
        "from fastai.callbacks import *\n",
        "\n",
        "# import torch\n",
        "show_install()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n",
            "\n",
            "```text\n",
            "=== Software === \n",
            "python        : 3.6.8\n",
            "fastai        : 1.0.59.dev0\n",
            "fastprogress  : 0.1.21\n",
            "torch         : 1.2.0\n",
            "nvidia driver : 418.67\n",
            "torch cuda    : 10.0.130 / is available\n",
            "torch cudnn   : 7602 / is enabled\n",
            "\n",
            "=== Hardware === \n",
            "nvidia gpus   : 1\n",
            "torch devices : 1\n",
            "  - gpu0      : 11441MB | Tesla K80\n",
            "\n",
            "=== Environment === \n",
            "platform      : Linux-4.14.137+-x86_64-with-Ubuntu-18.04-bionic\n",
            "distro        : #1 SMP Thu Aug 8 02:47:02 PDT 2019\n",
            "conda env     : Unknown\n",
            "python        : /usr/bin/python3\n",
            "sys.path      : \n",
            "/env/python\n",
            "/usr/lib/python36.zip\n",
            "/usr/lib/python3.6\n",
            "/usr/lib/python3.6/lib-dynload\n",
            "/usr/local/lib/python3.6/dist-packages\n",
            "/usr/lib/python3/dist-packages\n",
            "/usr/local/lib/python3.6/dist-packages/IPython/extensions\n",
            "/root/.ipython\n",
            "```\n",
            "\n",
            "Please make sure to include opening/closing ``` when you paste into forums/github to make the reports appear formatted as code sections.\n",
            "\n",
            "Optional package(s) to enhance the diagnostics can be installed with:\n",
            "pip install distro\n",
            "Once installed, re-run this utility to get the additional information\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "k2WuhH6hAHel",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# from fastai.callbacks import *\n",
        "# from fastai import *"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uYv7PnR4FpLw",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# import fastai\n",
        "# fastai.device = torch.device('cpu')\n",
        "# fastai.device"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "colab_type": "text",
        "id": "Q1PTqkp6BfNf"
      },
      "source": [
        "# EDA"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "FKhZANyfBfNZ",
        "colab": {}
      },
      "source": [
        "df = pd.read_csv('testset1.csv')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "outputId": "70cbe80f-f76e-436e-bf74-e3926f4bed36",
        "id": "uMy3VirjBfNS",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 224
        }
      },
      "source": [
        "print(len(df))\n",
        "\n",
        "df.head(5)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "5585\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>id</th>\n",
              "      <th>name</th>\n",
              "      <th>date</th>\n",
              "      <th>text</th>\n",
              "      <th>location</th>\n",
              "      <th>sentiment</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>1.130000e+18</td>\n",
              "      <td>Evolutionomics</td>\n",
              "      <td>Sun May 19 23:57:10 +0000 2019</td>\n",
              "      <td>By combining short positions on #Gold $GOLD $G...</td>\n",
              "      <td>NaN</td>\n",
              "      <td>bearish</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>1.130000e+18</td>\n",
              "      <td>macrohedged</td>\n",
              "      <td>Sun May 19 23:55:02 +0000 2019</td>\n",
              "      <td>Here are the #Volatility levels for the majors...</td>\n",
              "      <td>England</td>\n",
              "      <td>unrelated</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>1.130000e+18</td>\n",
              "      <td>Evolutionomics</td>\n",
              "      <td>Sun May 19 23:54:06 +0000 2019</td>\n",
              "      <td>We have little insight into what will happen t...</td>\n",
              "      <td>NaN</td>\n",
              "      <td>neutral</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>1.130000e+18</td>\n",
              "      <td>Evolutionomics</td>\n",
              "      <td>Sun May 19 23:50:59 +0000 2019</td>\n",
              "      <td>This brings us to #Gold miners. As a group, $G...</td>\n",
              "      <td>NaN</td>\n",
              "      <td>neutral</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>1.130000e+18</td>\n",
              "      <td>macrohedged</td>\n",
              "      <td>Sun May 19 23:50:04 +0000 2019</td>\n",
              "      <td>Good morning Asia, here are the benchmark futu...</td>\n",
              "      <td>England</td>\n",
              "      <td>unrelated</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "             id            name  ... location  sentiment\n",
              "0  1.130000e+18  Evolutionomics  ...      NaN    bearish\n",
              "1  1.130000e+18     macrohedged  ...  England  unrelated\n",
              "2  1.130000e+18  Evolutionomics  ...      NaN    neutral\n",
              "3  1.130000e+18  Evolutionomics  ...      NaN    neutral\n",
              "4  1.130000e+18     macrohedged  ...  England  unrelated\n",
              "\n",
              "[5 rows x 6 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 6
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "outputId": "31361b3d-bdea-4881-cc29-8c80c3e4d8d1",
        "id": "StOE93sZBfNM",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 324
        }
      },
      "source": [
        "df['sentiment'].value_counts().plot(kind='bar')"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.axes._subplots.AxesSubplot at 0x7f7ce33f47b8>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 7
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEiCAYAAADwEwVaAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nm/CziAAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "colab_type": "text",
        "id": "9WoDXcN_BfNK"
      },
      "source": [
        "Unrelated and neutral sentiments dominate the data currently (although this might not be the case in the future with the full dataset). It might be better to use a stratified sampling technique for classification"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "outputId": "ed8a8cb8-629f-4bc3-ad8a-dd91de7fbe97",
        "id": "uYQxJ-jfBfNE",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 52
        }
      },
      "source": [
        "from sklearn.model_selection import StratifiedShuffleSplit\n",
        "\n",
        "sss = StratifiedShuffleSplit(n_splits=5, test_size=0.2, random_state=0)\n",
        "sss.get_n_splits(df['text'], df['sentiment'])\n",
        "\n",
        "print(sss)\n",
        "\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "StratifiedShuffleSplit(n_splits=5, random_state=0, test_size=0.2,\n",
            "            train_size=None)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "outputId": "e2ab966c-ca5b-4293-8c83-a04d2e823aec",
        "id": "jht9mffiBfM-",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "train_idx, val_idx = next(sss.split(df['text'], df['sentiment']))\n",
        "\n",
        "print(len(train_idx), len(val_idx))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "4468 1117\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "colab_type": "text",
        "id": "KB8MkkEbBfM9"
      },
      "source": [
        "Verifying that the stratified split preseves the distribution of the labels"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "jhuNlUEKBfM3",
        "colab": {}
      },
      "source": [
        "df_train = df.iloc[train_idx, :]\n",
        "df_val = df.iloc[val_idx, :]"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "outputId": "fc81fd77-85fe-4847-d170-2cb651206cc7",
        "id": "RkiLk2ggBfMv",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 324
        }
      },
      "source": [
        "df_train['sentiment'].value_counts().plot(kind='bar')"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.axes._subplots.AxesSubplot at 0x7f7cde7cab38>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 11
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEiCAYAAADwEwVaAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nzwFIUqMMAElqlAEgSY0yACSpUf8PzVgX6BDadmMAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "outputId": "7f76675f-f7d5-45b2-d44b-06881c19fa8e",
        "id": "GkvIlQihBfMd",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 324
        }
      },
      "source": [
        "df_val['sentiment'].value_counts().plot(kind='bar')"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.axes._subplots.AxesSubplot at 0x7f7cde584390>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 12
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXoAAAEiCAYAAAD3fRkKAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "colab_type": "text",
        "id": "dEW7WWK7BfMI"
      },
      "source": [
        "Looks like the distribution of the labels in train and val are the same, so we can start modelling now"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lnlWeHCMCsPI",
        "colab_type": "text"
      },
      "source": [
        "# Handling Imbalanced Data: Oversampling"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LXjuHfxDD0fw",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "df_class_bull = df_train[df_train['sentiment'] == 'bullish']\n",
        "df_class_bear = df_train[df_train['sentiment'] == 'bearish']\n",
        "df_class_neutral = df_train[df_train['sentiment'] == 'neutral']\n",
        "df_class_unrelated = df_train[df_train['sentiment'] == 'unrelated']\n",
        "\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "T5yI-FewEDT-",
        "colab_type": "code",
        "outputId": "c93e2719-4f87-431e-da1b-7e1136a3527d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 36
        }
      },
      "source": [
        "print(len(df_class_unrelated))"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "2298\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SbBCJgpZEk8O",
        "colab_type": "text"
      },
      "source": [
        "Want to oversample so that every class has 2298 counts"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FcT0MDt8EoSn",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "df_class_bull_over = df_class_bull.sample(len(df_class_unrelated), replace = True)\n",
        "df_class_bear_over = df_class_bear.sample(len(df_class_unrelated), replace = True)\n",
        "df_class_neutral_over = df_class_neutral.sample(len(df_class_unrelated), replace = True)\n",
        "\n",
        "\n",
        "df_train_oversampled = pd.concat([df_class_bull_over, df_class_bear_over, df_class_neutral_over, df_class_unrelated], axis = 0)\n",
        "\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kxoV7EH5FPIY",
        "colab_type": "text"
      },
      "source": [
        "Look at distribution of the labels in the oversampled data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xA3owdg9FMWP",
        "colab_type": "code",
        "outputId": "520ea12b-6519-45af-9010-5eef60fe3439",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 326
        }
      },
      "source": [
        "df_train_oversampled['sentiment'].value_counts().plot(kind='bar')"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.axes._subplots.AxesSubplot at 0x7f2feae9b198>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 15
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAEiCAYAAADwEwVaAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nRK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Nq1WpX2hBtmH",
        "colab_type": "text"
      },
      "source": [
        "# BERT"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bLU6q8dj9bAk",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import pretrainedmodels"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "w8f_ySJcIgCC",
        "colab_type": "text"
      },
      "source": [
        "## Tokenizing and setting up databunch"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "On3FgZSK8DH1",
        "colab_type": "code",
        "outputId": "5c01985c-48bf-4fa7-978d-c08ad4bd604a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "from pytorch_pretrained_bert import BertTokenizer\n",
        "bert_tok = BertTokenizer.from_pretrained(\n",
        "    \"bert-base-uncased\",\n",
        ")"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "100%|██████████| 231508/231508 [00:00<00:00, 5777684.14B/s]\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7Cr62yZ3CzS-",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "class FastAiBertTokenizer(BaseTokenizer):\n",
        "    \"\"\"Wrapper around BertTokenizer to be compatible with fast.ai\"\"\"\n",
        "    def __init__(self, tokenizer: BertTokenizer, max_seq_len: int=128, **kwargs):\n",
        "        self._pretrained_tokenizer = tokenizer\n",
        "        self.max_seq_len = max_seq_len\n",
        "\n",
        "    def __call__(self, *args, **kwargs):\n",
        "        return self\n",
        "\n",
        "    def tokenizer(self, t:str) -> List[str]:\n",
        "        \"\"\"Limits the maximum sequence length\"\"\"\n",
        "        return [\"[CLS]\"] + self._pretrained_tokenizer.tokenize(t)[:self.max_seq_len - 2] + [\"[SEP]\"]"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0sFEtuTE-UYr",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "fastai_bert_vocab = Vocab(list(bert_tok.vocab.keys()))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "r_N18RBn-V_J",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "fastai_tokenizer = Tokenizer(tok_func=FastAiBertTokenizer(bert_tok, max_seq_len=256), pre_rules=[], post_rules=[])\n",
        "\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8iW0c2Il-X4W",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "data_clas = TextDataBunch.from_df(\".\", df_train, df_val, \n",
        "                  tokenizer=fastai_tokenizer,\n",
        "                  vocab=fastai_bert_vocab,\n",
        "                  include_bos=False,\n",
        "                  include_eos=False,\n",
        "                  text_cols=\"text\",\n",
        "                  label_cols='sentiment',\n",
        "                  bs=16,\n",
        "                  collate_fn=partial(pad_collate, pad_first=False, pad_idx=0))\n",
        "\n",
        "\"\"\"databunch with oversampled df to correct for class imbalance\"\"\"\n",
        "# data_clas = TextDataBunch.from_df(\".\", df_train_oversampled, df_val, \n",
        "#                   tokenizer=fastai_tokenizer,\n",
        "#                   vocab=fastai_bert_vocab,\n",
        "#                   include_bos=False,\n",
        "#                   include_eos=False,\n",
        "#                   text_cols=\"text\",\n",
        "#                   label_cols='sentiment',\n",
        "#                   bs=16,\n",
        "#                   collate_fn=partial(pad_collate, pad_first=False, pad_idx=0))\n",
        "\n",
        "data_clas.save('data_clas')"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lRz82F8e-gXr",
        "colab_type": "code",
        "outputId": "14e6cf08-9de8-45d3-cea0-d11203baa50d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 293
        }
      },
      "source": [
        "data_clas.show_batch()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th>text</th>\n",
              "      <th>target</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <td>[CLS] [UNK] - 0 . 95 % [UNK] : [UNK] &amp; amp ; [UNK] # et ##fs # daily # market # change tu ##e , may 28 [UNK] # fear - 0 . 29 % [UNK] $ xl ##k - 0 . 38 % ♣ $ q ##q ##q - 0 . 39 % [UNK] $ g ##ld - 0 . 66 % [UNK] $ xl ##y - 0</td>\n",
              "      <td>neutral</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>[CLS] buy / add long - term : $ bt ##c - $ usd $ 7 , 83 ##7 . 86 - 149 . 98 ( - 1 . 88 % ) # bit ##co ##in # usd # crypt ##oc ##ur ##ren ##cy 10 : 00 am 5 / 20 / 19 $ usd ##u $ u ##up $ cy ##b $ yin ##n $ yang $ cn $</td>\n",
              "      <td>unrelated</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>[CLS] # bit ##co ##in i ' m long - term in $ bt ##c - $ usd 8 , 125 . 97 + 249 . 43 ( + 3 . 17 % ) as of 4 : 19 ##pm bs ##t . market open . $ usd ##u $ u ##up $ cy ##b $ yin ##n $ yang $ cn $ ch ##n $ fx ##ch $ cn</td>\n",
              "      <td>unrelated</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>[CLS] [UNK] - 0 . 67 % [UNK] : [UNK] &amp; amp ; [UNK] # et ##fs # daily # market # change wed , may 29 [UNK] 0 . 03 % [UNK] $ g ##ld 0 . 02 % [UNK] $ xl ##b [UNK] # down - 0 . 19 % [UNK] $ xl ##f - 0 . 45 % [UNK] $ xl ##i - 0 . 59 %</td>\n",
              "      <td>neutral</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>[CLS] [UNK] - 0 . 93 % [UNK] : [UNK] &amp; amp ; [UNK] # et ##fs # daily # market # change fr ##i , may 31 [UNK] 1 . 35 % [UNK] $ g ##ld 0 . 83 % [UNK] $ xl ##re 0 . 47 % [UNK] $ xl ##u [UNK] # down - 0 . 72 % [UNK] $ xl ##v - 1 . 12 %</td>\n",
              "      <td>neutral</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "R3wnJTeZDNzL",
        "colab_type": "text"
      },
      "source": [
        "## Using BERT"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VBrRG4bABuOJ",
        "colab_type": "code",
        "outputId": "ff238e70-ba2e-42ca-c6a3-f01fe619b228",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "from pytorch_pretrained_bert.modeling import BertConfig, BertForSequenceClassification, BertForNextSentencePrediction, BertForMaskedLM\n",
        "bert_model_class = BertForSequenceClassification.from_pretrained('bert-base-uncased', num_labels=4)\n",
        "\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "100%|██████████| 407873900/407873900 [00:09<00:00, 44846989.79B/s]\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "C_BjNSYtDQNu",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "model = bert_model_class"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2obJo9MgTjGu",
        "colab_type": "text"
      },
      "source": [
        "### Classification"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jCcmYEihCcPf",
        "colab_type": "text"
      },
      "source": [
        "Not using mixed precision yet, do that after lr_find()"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "urtYAR7d_K-r",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "\n",
        "learn = Learner(data_clas,\n",
        "               model,\n",
        "               loss_func = FlattenedLoss(LabelSmoothingCrossEntropy, axis=-1),\n",
        "               metrics = [accuracy])\n",
        "\n",
        "\n",
        "# learn = Learner(databunch_1,\n",
        "#                model,\n",
        "#                loss_func = nn.BCEWithLogitsLoss(),\n",
        "#                metrics = [accuracy])\n",
        "# # learn.to_fp16()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JDwVtfDRDi8J",
        "colab_type": "text"
      },
      "source": [
        "Splitting bert for discriminative learning"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hx8XSwUZ_op2",
        "colab_type": "code",
        "outputId": "a354f2a2-b4ec-47dc-a33c-98931fc61a00",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "def bert_clas_split(self) -> List[nn.Module]:\n",
        "    \n",
        "    bert = model.bert\n",
        "    embedder = bert.embeddings\n",
        "    pooler = bert.pooler\n",
        "    encoder = bert.encoder\n",
        "    classifier = [model.dropout, model.classifier]\n",
        "    n = len(encoder.layer)//3\n",
        "    print(n)\n",
        "    groups = [[embedder], list(encoder.layer[:n]), list(encoder.layer[n+1:2*n]), list(encoder.layer[(2*n)+1:]), [pooler], classifier]\n",
        "    return groups\n",
        "\n",
        "x = bert_clas_split(model)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "4\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NStGu-43_9_4",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "learn.split([x[0], x[1], x[2], x[3], x[5]]); # should this be x[4]?"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pOzW-rmNDSbM",
        "colab_type": "code",
        "outputId": "17e63435-8b35-4238-94be-a3c8cad8903e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 335
        }
      },
      "source": [
        "learn.lr_find()\n",
        "learn.recorder.plot(suggestion = True, skip_end=15)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              ""
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "LR Finder is complete, type {learner_name}.recorder.plot() to see the graph.\n",
            "Min numerical gradient: 1.32E-04\n",
            "Min loss divided by 10: 6.92E-05\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYsAAAEKCAYAAADjDHn2AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NwWZE2adDq6p",
        "colab_type": "code",
        "outputId": "70686aee-fbf9-4f27-e56b-6e6b290bb535",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 213
        }
      },
      "source": [
        "learn.to_fp16()\n",
        "\n",
        "lr = 1e-4\n",
        "\n",
        "\"\"\"Training without adjusting the momentum\"\"\"\n",
        "learn.fit_one_cycle(3, \n",
        "                    lr, \n",
        "                    callbacks=[SaveModelCallback(learn, every='improvement', monitor='accuracy', \n",
        "                                                 name='classifier_stage1')])\n",
        "\"\"\"\n",
        "Clipping momentum\n",
        "\"\"\"\n",
        "\n",
        "\n",
        "# learn.fit_one_cycle(10, \n",
        "#                     lr, \n",
        "#                     moms=(0.8,0.7),\n",
        "#                     callbacks=[SaveModelCallback(learn, every='improvement', monitor='accuracy', \n",
        "#                                                  name='classifier_stage1')])"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: left;\">\n",
              "      <th>epoch</th>\n",
              "      <th>train_loss</th>\n",
              "      <th>valid_loss</th>\n",
              "      <th>accuracy</th>\n",
              "      <th>time</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <td>0</td>\n",
              "      <td>1.044614</td>\n",
              "      <td>0.869819</td>\n",
              "      <td>0.702775</td>\n",
              "      <td>05:32</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>1</td>\n",
              "      <td>0.828451</td>\n",
              "      <td>0.745140</td>\n",
              "      <td>0.784244</td>\n",
              "      <td>06:04</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>2</td>\n",
              "      <td>0.676445</td>\n",
              "      <td>0.663156</td>\n",
              "      <td>0.837959</td>\n",
              "      <td>05:54</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Better model found at epoch 0 with accuracy value: 0.7027752995491028.\n",
            "Better model found at epoch 1 with accuracy value: 0.7842435240745544.\n",
            "Better model found at epoch 2 with accuracy value: 0.837958812713623.\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "'\\nClipping momentum\\n'"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 26
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ema-ZuVIjk1E",
        "colab_type": "text"
      },
      "source": [
        "Additional training by unfreezing the network and applying discriminative learning rates"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1LM4MfBPOB5R",
        "colab_type": "code",
        "outputId": "1667fb53-a06e-42d6-ca41-3f520ec9197b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 462
        }
      },
      "source": [
        "learn.load('classifier_stage1');\n",
        "learn.unfreeze()\n",
        "\n",
        "\n",
        "\"\"\"Training without adjusting the momentum\"\"\"\n",
        "learn.fit_one_cycle(3, \n",
        "                    slice(lr/100, lr), \n",
        "                    callbacks=[SaveModelCallback(learn, every='improvement', monitor='accuracy', \n",
        "                                                 name='classifier_final')])\n",
        "\"\"\"\n",
        "Clipping momentum\n",
        "\"\"\"\n",
        "# learn.fit_one_cycle(10, \n",
        "#                    slice(lr/(2.6**4),lr), \n",
        "#                    moms=(0.8,0.7),\n",
        "#                    callbacks=[SaveModelCallback(learn, every='improvement', monitor='accuracy', \n",
        "#                                                  name='classifier_stage2')])\n",
        "\n",
        "learn.recorder.plot_losses()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: left;\">\n",
              "      <th>epoch</th>\n",
              "      <th>train_loss</th>\n",
              "      <th>valid_loss</th>\n",
              "      <th>accuracy</th>\n",
              "      <th>time</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <td>0</td>\n",
              "      <td>0.638021</td>\n",
              "      <td>0.649720</td>\n",
              "      <td>0.842435</td>\n",
              "      <td>05:48</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>1</td>\n",
              "      <td>0.536543</td>\n",
              "      <td>0.619438</td>\n",
              "      <td>0.866607</td>\n",
              "      <td>05:46</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <td>2</td>\n",
              "      <td>0.531737</td>\n",
              "      <td>0.609123</td>\n",
              "      <td>0.871083</td>\n",
              "      <td>05:42</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Better model found at epoch 0 with accuracy value: 0.8424351215362549.\n",
            "Better model found at epoch 1 with accuracy value: 0.8666070103645325.\n",
            "Better model found at epoch 2 with accuracy value: 0.8710832595825195.\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYsAAAEKCAYAAADjDHn2AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FTlJCSFQOgbD",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JwOMveMMrbqS",
        "colab_type": "text"
      },
      "source": [
        "# Model Inference"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rWkj_kRYremX",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "learn = Learner(data_clas, \n",
        "                model)\n",
        "\n",
        "# learn = Learner(data_clas,\n",
        "#                model,\n",
        "#                loss_func = FlattenedLoss(LabelSmoothingCrossEntropy, axis=-1),\n",
        "#                metrics = [accuracy])\n",
        "\n",
        "\n",
        "\n",
        "learn.load('classifier_final', with_opt = False); # makesure learner  is a text_classifier_learner\n",
        "\n",
        "learn.to_fp32() # bring back to fp32\n",
        "\n",
        "\n",
        "\n",
        "learn.export(file = 'export_clas.pkl')\n",
        "\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lCzZE3Qor5Zj",
        "colab_type": "code",
        "outputId": "43f28535-f81b-428d-d900-99161e2d8ca8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "phrase = \"$GLD will lose value over the next quarter\"\n",
        "learn.predict(phrase)\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(Category bearish, tensor(0), tensor([0.6633, 0.1695, 0.1286, 0.0385]))"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 52
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hxfceH0moiWG",
        "colab_type": "text"
      },
      "source": [
        "## Making batch predictions on new dataset\n",
        "\n",
        "test_df can be loaded in directly from a csv file i.e. 1 days worth of collected tweets"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ECk16RpRokCl",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# test_df = pd.read_csv('testset1.csv')\n",
        "\n",
        "test_df = df_val\n",
        "learn.data.add_test(test_df['text'])\n",
        "preds, y, losses = learn.get_preds(ds_type=DatasetType.Test, with_loss = True)\n",
        "# preds, y, losses = learn.get_preds(ds_type=DatasetType.Test, ordered=True, with_loss = True)\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YdGaxrkzvX2F",
        "colab_type": "code",
        "outputId": "1677946d-5cce-4273-a9c1-60f53e8bce97",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "\n",
        "y_pred = torch.argmax(preds, dim= 1).numpy()\n",
        "\n",
        "\n",
        "print(y_pred, len(y_pred))\n",
        "\n",
        "pred_labels = []\n",
        "for pred in y_pred:\n",
        "    if pred == 0: \n",
        "        pred_labels.append('bearish')\n",
        "    elif pred == 1:\n",
        "        pred_labels.append('bullish')    \n",
        "    elif pred == 2:\n",
        "        pred_labels.append('neutral')\n",
        "    elif pred == 3:\n",
        "        pred_labels.append('unrelated')"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[2 3 3 3 ... 3 3 2 3] 1117\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9JswniVzw7Mi",
        "colab_type": "text"
      },
      "source": [
        "## Visualizing predictions\n",
        "\n",
        "visualize the count of predictions\n",
        "\n",
        "1 = Bearish\n",
        "2 = Bullish\n",
        "3 = Neutral\n",
        "4 = Unrelated"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gTeMOlI5vUln",
        "colab_type": "code",
        "outputId": "8ee99415-e3f0-41bd-82fb-328fbe1fb2e2",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 321
        }
      },
      "source": [
        "plt.hist(pred_labels)\n",
        "\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(array([326.,   0.,   0., 544.,   0.,   0.,  61.,   0.,   0., 186.]),\n",
              " array([0. , 0.3, 0.6, 0.9, 1.2, 1.5, 1.8, 2.1, 2.4, 2.7, 3. ]),\n",
              " <a list of 10 Patch objects>)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 45
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXoAAAD8CAYAAAB5Pm/hAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nxlYAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sHwS8YclxFn4",
        "colab_type": "text"
      },
      "source": [
        "The distribution of the predictions seems to be inline with the distribution of the val labels"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bHMHurnLzN88",
        "colab_type": "text"
      },
      "source": [
        "View a confusion matrix of the predictions\n",
        "\n",
        "y-axis is the true labels"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "iZGOBv7Jwo89",
        "colab_type": "code",
        "outputId": "7d72596c-afb3-4337-97e2-6051ffb128eb",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 286
        }
      },
      "source": [
        "from sklearn.metrics import confusion_matrix\n",
        "from sklearn.utils.multiclass import unique_labels\n",
        "import seaborn as sn\n",
        "cm = confusion_matrix(df_val['sentiment'], pred_labels)\n",
        "\n",
        "df_cm = pd.DataFrame(cm, index = ['bearish', 'bullsih', 'neutral', 'unrelated'],\n",
        "                    columns = ['bearish', 'bullsih', 'neutral', 'unrelated'])\n",
        "\n",
        "sn.heatmap(df_cm, annot=True)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.axes._subplots.AxesSubplot at 0x7f7cdad6b278>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 46
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAWAAAAD8CAYAAABJsn7AAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nrkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "k8YzBtFVieBD",
        "colab_type": "text"
      },
      "source": [
        "Observations\n",
        "\n",
        "Performance seems to match what AWD-LSTM is getting, slightly more bullish tweets categorized as bearish, but that could be due to the data split"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LW4ffxmUtR72",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}