SentimentAnalysisMR.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "SentimentAnalysisMR.ipynb",
      "provenance": [],
      "collapsed_sections": [
        "FcnSO7Nk2I4N"
      ],
      "toc_visible": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5bpoxTVMiiL0"
      },
      "source": [
        "# Sentiment Analysis of Movie Reviews"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "o9GKTaNcmCXV"
      },
      "source": [
        "Pytorch implementation of Sentiment Analysis of long textual movie reviews written in Serbian language (which is underused language) using pretrained multinigual Roberta based language model [XLM-R](https://ai.facebook.com/blog/-xlm-r-state-of-the-art-cross-lingual-understanding-through-self-supervision/). Dataset is balanced and it consits of 1684 **positive** and **negative** reviews, and it can be found [here](https://github.com/vukbatanovic/SerbMR). This balanced dataset is a subset of unbalanced set of 4725 movie reviews. Reviews have scoring metric which ranges from 1 to 10. Scores from 1-4 are treated as negative, 5-6 as neutral and 7-10 as positive. So some reviews have very weak sentiment and are very hard to classify correctly even by humans. "
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tpZfasoQiGin"
      },
      "source": [
        "## Connect to google drive"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4QcZVaOvpeBw"
      },
      "source": [
        "Connect to Google Drive so that you can read the data and save checkpoints."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "iKFzRs3-NDkL",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "7d99dce9-58b2-4b94-a56a-bb94f58e9de5"
      },
      "source": [
        "from google.colab import drive\r\n",
        "drive.mount('/content/google_drive')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Mounted at /content/google_drive\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9A2AYxVhiEOA"
      },
      "source": [
        "Change to directory where the notebook is located in the drive"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YHgYX8AmNQLT"
      },
      "source": [
        "%cd /content/google_drive/MyDrive/Colab\\ Notebooks/Sentiment-Analysis-Roberta\r\n",
        "%ls"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1I_JI-p6dtGh"
      },
      "source": [
        "## Load the data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WBs1g4CUbCJR",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 204
        },
        "outputId": "15bad5ce-fb82-40c0-9b70-ea4f0335c677"
      },
      "source": [
        "import os\r\n",
        "import pandas as pd\r\n",
        "import numpy as np\r\n",
        "import matplotlib.pyplot as plt\r\n",
        "\r\n",
        "data_path = os.path.join(os.getcwd(),'data')\r\n",
        "movies_path = os.path.join(data_path,'Movie-Reviews')\r\n",
        "df = pd.read_csv(os.path.join(movies_path,'SerbMR-2C.csv'))\r\n",
        "df.rename(columns={'class-att':'Label'},inplace=True)\r\n",
        "df.head()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "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>Text</th>\n",
              "      <th>Label</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>Braća Koen (Coen brothers) iako poznati po tri...</td>\n",
              "      <td>POSITIVE</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>Često upadam u veliku raspravu kada se izjasni...</td>\n",
              "      <td>POSITIVE</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>“Ulični Psi” je jedan od onih filmova koji sve...</td>\n",
              "      <td>POSITIVE</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>Užasno potcenjen film! Predvođen dvojicom najp...</td>\n",
              "      <td>POSITIVE</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>Naprosto sam oduševljen ovim projektom, ovo je...</td>\n",
              "      <td>POSITIVE</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                                                Text     Label\n",
              "0  Braća Koen (Coen brothers) iako poznati po tri...  POSITIVE\n",
              "1  Često upadam u veliku raspravu kada se izjasni...  POSITIVE\n",
              "2  “Ulični Psi” je jedan od onih filmova koji sve...  POSITIVE\n",
              "3  Užasno potcenjen film! Predvođen dvojicom najp...  POSITIVE\n",
              "4  Naprosto sam oduševljen ovim projektom, ovo je...  POSITIVE"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 3
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "E-scUM6-5MO4"
      },
      "source": [
        "Example of a movie review"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "3cssUfUI5-bM",
        "outputId": "633af3e4-eedb-467b-80f0-dc94c8434079"
      },
      "source": [
        "EXAMPLE_NUM =   0#@param {type: 'integer'}\r\n",
        "[print(x+'.') for x in df.Text.iloc[EXAMPLE_NUM].split('.')]\r\n",
        "print('')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Braća Koen (Coen brothers) iako poznati po trilerima, oprobali su se više puta i u komedija, i postigli potpuni uspeh.\n",
            " Ovaj film, pošto je kada se pojavio bio vrlo loše prihvaćen, nije nažalost uspeo da zablista po američkim bioskopima, ali je zato bio prava senzacija kad se pojavio na DVD-u i na osnovu toga, postao jedan od glavnih naslova u kolekciji svakog pravog filmofila.\n",
            " „The Big Lebowski“ definitvno moj omiljeni film i jedino ostvarenje koje zaista uvek iznova i iznova mogu da gledam.\n",
            " On predstavlja odu životnom stilu jednog pacifiste.\n",
            " „The Big Lewbowski“ je klasična priča prevare, kriminala i spletkarenja viđena kroz oči skromnog čoveka, tačnije jednostavne individue sa vrlo malo prohteva, želja i ambicija.\n",
            " Žanr ovog filma nije lako odrediti – može se reći da je komedija zbog svog izuzetno originalnog humorističkog sadržaja.\n",
            " Bogat fantastičnim likovima i još boljim dijalozima, koji iako su se transformisali u besmrtne citate koji se koriste u svakodnevnom životu, „The Big Lebowski“ i pored ovih sada već navedenih segmenata ima još toliko mnogo toga da ponudi.\n",
            " Gluma u ovom filmu je zaista neponovljiva.\n",
            " Definitvni vrh karijere za Džefa Bridžesa (Jeff Bridges) i Džona Gudmana (John Goodman).\n",
            " Iako su obojica vrhunska klasa glumaca, sa izvanrednim karijerama, nikada nisu uspeli da se udalje od ovih kultnih likova, a otuda i Bridžesu nadimak koji ga prati već deceniju i pratiće ga ceo život – The Dude.\n",
            " Cela glumačka ekipa je izuzetnog kvaliteta.\n",
            " Tu stoje još imena kao što su Stiv Bušemi (Steve Buscemi), Džon Torturo (John Turturro), Džulijana Mur (Julianne Moore) i Filip Sejmur Hofman (Philip Seymour Hoffman).\n",
            " Svako je zaista uradio i više što se od njih moglo očekivati, a pogotovo Torturo koje je ovde stvario jednog od najzabavnijih epizodnih likova u svim filmovima ikada.\n",
            " Po mom mišljenju, „The Big Lebowski“ je jedno od najvećih dostignuća u modernoj kinematografiji.\n",
            " Ovaj projekat će vas nasmejati, zbuniti i zadiviti, i on se sa razlogom smatra za jednu od najcenjenijih komedija od strane većine kritičara.\n",
            ".\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sI_9x0jYqOh1"
      },
      "source": [
        "##  Split into train, test and val set"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "luNZItzZ4yfF"
      },
      "source": [
        "Ratio: (80%-10%-10%) - (train - val - test)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TkrH-vN-gL7a"
      },
      "source": [
        "def split_df(df):\r\n",
        "    n_rows = len(df)\r\n",
        "    ind = np.random.permutation(n_rows)\r\n",
        "    df = df.iloc[ind,:]\r\n",
        "    df_train = df.iloc[:int(0.8*n_rows),:]\r\n",
        "    df_val = df.iloc[int(0.8*n_rows):int(0.9*n_rows),:]\r\n",
        "    df_test = df.iloc[int(0.9*n_rows):,:]\r\n",
        "\r\n",
        "    return df_train, df_val, df_test"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "56nLjaCUd13B",
        "outputId": "5e2d8567-ced8-4430-9b51-b709c547ee82"
      },
      "source": [
        "n_rows = len(df)\r\n",
        "dfs_train, dfs_val, dfs_test = [],[],[]\r\n",
        "gb = df.groupby('Label')\r\n",
        "for x in gb.groups:\r\n",
        "    group = gb.get_group(x)\r\n",
        "    df_train, df_val, df_test = split_df(group)\r\n",
        "    dfs_train.append(df_train)\r\n",
        "    dfs_val.append(df_val)\r\n",
        "    dfs_test.append(df_test) \r\n",
        "df_train = pd.concat(dfs_train, ignore_index=True)\r\n",
        "ind = np.random.permutation(len(df_train))\r\n",
        "df_train = df_train.iloc[ind,:]\r\n",
        "df_val = pd.concat(dfs_val, ignore_index=True)\r\n",
        "ind = np.random.permutation(len(df_val))\r\n",
        "df_val = df_val.iloc[ind,:]\r\n",
        "df_test = pd.concat(dfs_test, ignore_index=True)\r\n",
        "ind = np.random.permutation(len(df_test))\r\n",
        "df_test = df_test.iloc[ind,:]\r\n",
        "print(f'Number of training examples: {len(df_train)}')\r\n",
        "print(f'Number of validation examples: {len(df_val)}')\r\n",
        "print(f'Number of test examples: {len(df_test)}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Number of training examples: 1344\n",
            "Number of validation examples: 168\n",
            "Number of test examples: 170\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 245
        },
        "id": "D76T7DAqcG-2",
        "outputId": "52f5aac8-5afa-4a9f-906f-cfb926f443ec"
      },
      "source": [
        "fig = plt.figure(figsize=(18,4))\r\n",
        "titles = ['Train set','Test set','Val set']\r\n",
        "for i,dataf in enumerate([df_train,df_test,df_val]):\r\n",
        "    ax = fig.add_subplot(1,3,i+1)\r\n",
        "    Y, labels = pd.factorize(dataf['Label'])\r\n",
        "    ax.bar(labels, height=pd.Series(Y).value_counts())\r\n",
        "    ax.set_xticks(ticks=range(len(labels)))\r\n",
        "    ax.set_xticklabels(labels,fontsize=10)\r\n",
        "    ax.set_xlabel('Review')\r\n",
        "    ax.set_ylabel('Number of examples')\r\n",
        "    ax.set_title(titles[i])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1296x288 with 3 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FDb91FO6upLb"
      },
      "source": [
        "## Load XLM-R model"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9-WDMS9L8mJq"
      },
      "source": [
        "Since the Facebook researchers are the creators of the model, the model can be found in fairseq package. So we need to install dependencies for fairseq and download the model from pytorch hub. We also need to install transformer package, because we will be using advanced optimizers and learning rate schedulers."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rh7r0_SE8cXJ"
      },
      "source": [
        "%pip install hydra-core\r\n",
        "%pip install omegaconf\r\n",
        "%pip install sentencepiece\r\n",
        "%pip install transformers"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8RUKzgar_lIw"
      },
      "source": [
        "Load XLM-R model and sentencepiece tokenizer"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oRDEIlh2usDI"
      },
      "source": [
        "import torch\r\n",
        "from torch.utils.data import Dataset, DataLoader\r\n",
        "import torch.nn as nn\r\n",
        "from statistics import mean\r\n",
        "import pickle\r\n",
        "from transformers import XLMRobertaTokenizer, AdamW, get_linear_schedule_with_warmup\r\n",
        "from torch.optim import SGD"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "RTEACFRwa2yg"
      },
      "source": [
        "BASE (2 times smaller than large)\r\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0sHT_T1Ja5o1"
      },
      "source": [
        "xlmr = torch.hub.load('pytorch/fairseq', 'xlmr.base')\r\n",
        "tokenizer = XLMRobertaTokenizer.from_pretrained(\"xlm-roberta-base\")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "X85GNtTsQlxD"
      },
      "source": [
        "Wrap tokenizer by specifing its hyperparametrs"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qVgLkRlGNjPj"
      },
      "source": [
        "def wrap_tokenizer(tokenizer, padding=True, truncation=True, return_tensors='pt', max_length=None):\r\n",
        "    def tokenize(text):\r\n",
        "        text = list(text)\r\n",
        "        tokens = tokenizer(\r\n",
        "            text,\r\n",
        "            padding=padding, \r\n",
        "            return_attention_mask=False,\r\n",
        "            truncation=truncation,\r\n",
        "            max_length=max_length,\r\n",
        "            return_tensors=return_tensors\r\n",
        "            )['input_ids']\r\n",
        "        return tokens\r\n",
        "    return tokenize"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Rx97ed3FYCJA"
      },
      "source": [
        "## Determine the distribution of number of tokens of texts"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "EPx8eVP1rr0m"
      },
      "source": [
        "Since we are dealing with large texts which probably can not fit into the XLM-R model (which is around 512), so we will examine the distribution of number of tokens of texts to see how we can split the text to fit into our model."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2j6jsaVMXpvo",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 590
        },
        "outputId": "003ec42c-846c-41d0-cc62-117f9187d471"
      },
      "source": [
        "import matplotlib.pyplot as plt\r\n",
        "import seaborn as sns\r\n",
        "\r\n",
        "tokenize = wrap_tokenizer(tokenizer, padding=False, truncation = False, return_tensors=None)\r\n",
        "tokens = tokenize(df['Text'])\r\n",
        "num_tokens = [len(x) for x in tokens]\r\n",
        "# Plot histogram with the length. Truncate max length to 5000 tokens.\r\n",
        "plt.style.use(\"ggplot\")\r\n",
        "plt.figure(figsize=(10, 8))\r\n",
        "df['length'] = pd.Series(num_tokens)\r\n",
        "sns.distplot(df['length'])\r\n",
        "plt.title('Frequence of text of a given token length', fontsize=14)\r\n",
        "plt.xlabel('token length', fontsize=14)\r\n",
        "plt.axvline(x=512,linewidth=4,color='black')\r\n",
        "plt.axvline(x=2*512,linewidth=4,color='green')\r\n",
        "plt.axvline(x=3*512,linewidth=4,color='blue')\r\n",
        "plt.axvline(x=4*512,linewidth=4,color='yellow')\r\n",
        "plt.axvline(x=5*512,linewidth=4,color='purple')\r\n",
        "plt.legend(['freq','max_tokens','2*max_tokens','3*max_tokens','4*max_tokens','5*max_tokens'])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/seaborn/distributions.py:2557: FutureWarning: `distplot` is a deprecated function and will be removed in a future version. Please adapt your code to use either `displot` (a figure-level function with similar flexibility) or `histplot` (an axes-level function for histograms).\n",
            "  warnings.warn(msg, FutureWarning)\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.legend.Legend at 0x7ff1fe6d8d30>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 47
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 720x576 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wvymaabjY5U8",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "4a3e7f2e-4bcd-4e6d-f7d8-afb3cf152b96"
      },
      "source": [
        "max_tokens = 512 # 512\r\n",
        "for i in range(1,6):\r\n",
        "    num = sum(df['length']>i*max_tokens)\r\n",
        "    print(f\"Number of text that have more than {i}*max_tokens is {num}\")"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Number of text that have more than 1*max_tokens is 1355\n",
            "Number of text that have more than 2*max_tokens is 320\n",
            "Number of text that have more than 3*max_tokens is 62\n",
            "Number of text that have more than 4*max_tokens is 13\n",
            "Number of text that have more than 5*max_tokens is 5\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8Iad9GhpZ5CM"
      },
      "source": [
        "We can see that 1355 examples out of 1682 would not fit into our model. So we will need to split the text into multiple chunks and input each chunk separately through the model. After we get the vector representations of context of each chunk we will join them into one with LSTM layer (there are various ways to do this, for example the easiest way would be to perform sum or average pooling on the contexts). I have chosen the LSTM out of learning purposes."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ye1h5R0d86W7"
      },
      "source": [
        "def get_text_split(text, length=200, overlap=50, max_chunks=4):\r\n",
        "    l_total = []\r\n",
        "    l_parcial = []\r\n",
        "    n_words = len(text.split()) \r\n",
        "    n = n_words//(length-overlap)+1 # n is number of splits\r\n",
        "    if n_words % (length-overlap) == 0:\r\n",
        "        n = n-1\r\n",
        "    if n ==0:\r\n",
        "        n = 1\r\n",
        "    n = min(n, max_chunks)\r\n",
        "    for w in range(n):\r\n",
        "        if w == 0:\r\n",
        "            l_parcial = text.split()[:length]\r\n",
        "        else:\r\n",
        "            l_parcial = text.split()[w*(length-overlap):w*(length-overlap) + length]\r\n",
        "        l = \" \".join(l_parcial)\r\n",
        "        if w==n-1:\r\n",
        "            if len(l_parcial) < 0.75*length and n!=1: # keep the last if it has 75% of length\r\n",
        "                continue\r\n",
        "        l_total.append(l)\r\n",
        "    return l_total"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lwV6ZloGMB7_"
      },
      "source": [
        "df_train.Text = df_train.Text.apply(lambda x: get_text_split(x))\r\n",
        "df_val.Text = df_val.Text.apply(lambda x: get_text_split(x))\r\n",
        "df_test.Text = df_test.Text.apply(lambda x: get_text_split(x))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NvbvLRpEPybA"
      },
      "source": [
        "df_train['n_chunks'] = df_train.Text.apply(lambda x: len(x))\r\n",
        "df_val['n_chunks'] = df_val.Text.apply(lambda x: len(x))\r\n",
        "df_test['n_chunks'] = df_test.Text.apply(lambda x: len(x))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Nr0BXEOzxe3n"
      },
      "source": [
        "Encode Labels"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pSW7fEsTe1zn"
      },
      "source": [
        "for df in [df_train, df_val, df_test]:\r\n",
        "    df.loc[df.Label=='POSITIVE','Label'] = 1\r\n",
        "    df.loc[df.Label=='NEGATIVE','Label'] = 0"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-ASeQs6fxTtb"
      },
      "source": [
        "Save train, val and test sets into files"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OANTabGMr6CI"
      },
      "source": [
        "df_train.to_csv(os.path.join(data_path,'train.csv'))\r\n",
        "df_val.to_csv(os.path.join(data_path,'val.csv'))\r\n",
        "df_test.to_csv(os.path.join(data_path,'test.csv'))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "AAKrx0q-xa5J"
      },
      "source": [
        "Examine the training dataset once more"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 202
        },
        "id": "wElGljUYo2-r",
        "outputId": "00c7f56b-e5d8-4430-d407-8bafdb2f4996"
      },
      "source": [
        "df_train.head()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "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>Text</th>\n",
              "      <th>Label</th>\n",
              "      <th>n_chunks</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>281</th>\n",
              "      <td>[Ko se posle očajnog prethodnika Terminator 3:...</td>\n",
              "      <td>0</td>\n",
              "      <td>3</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>717</th>\n",
              "      <td>[﻿Filmovi koji se bave moralnom degradacijom s...</td>\n",
              "      <td>1</td>\n",
              "      <td>3</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1274</th>\n",
              "      <td>[Hirokazu Koreeda je režirao jedan od najbolji...</td>\n",
              "      <td>1</td>\n",
              "      <td>3</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>918</th>\n",
              "      <td>[U filmu Dead Man (Jarmusch, 1995), muzika Nei...</td>\n",
              "      <td>1</td>\n",
              "      <td>3</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>745</th>\n",
              "      <td>[Postoji li ili ne žanr strip-film i koliko je...</td>\n",
              "      <td>1</td>\n",
              "      <td>3</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                                                   Text Label  n_chunks\n",
              "281   [Ko se posle očajnog prethodnika Terminator 3:...     0         3\n",
              "717   [﻿Filmovi koji se bave moralnom degradacijom s...     1         3\n",
              "1274  [Hirokazu Koreeda je režirao jedan od najbolji...     1         3\n",
              "918   [U filmu Dead Man (Jarmusch, 1995), muzika Nei...     1         3\n",
              "745   [Postoji li ili ne žanr strip-film i koliko je...     1         3"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 18
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Re_jc_pv2ZOl"
      },
      "source": [
        "## Complete Model architecture\r\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "W3fGT-MUBOWG"
      },
      "source": [
        "![model_sa_MR_complete.png](data:image/png;base64,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)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FcnSO7Nk2I4N"
      },
      "source": [
        "## Train functions\r\n",
        "\r\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xA9Dc9TeAV4g"
      },
      "source": [
        "Define custom dataset"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1o2k24F9YwCR"
      },
      "source": [
        "class MyDataset(Dataset):\r\n",
        "    def __init__(self,df):\r\n",
        "        self.n_chunks = df['n_chunks'].to_list()\r\n",
        "        self.X = df['Text'].to_list()\r\n",
        "        self.Y = df['Label']\r\n",
        "    def __len__(self):\r\n",
        "        return len(self.X)\r\n",
        "    def __getitem__(self,index):\r\n",
        "        return self.X[index], self.Y.iloc[index], self.n_chunks[index]\r\n",
        "def collate_func(batch):\r\n",
        "    X = [x[0] for x in batch]\r\n",
        "    Y = torch.Tensor([x[1] for x in batch])\r\n",
        "    c = [x[2] for x in batch]\r\n",
        "    return [X,Y,c]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jiLgVZxtXpxQ"
      },
      "source": [
        "from sklearn.metrics import confusion_matrix\r\n",
        "\r\n",
        "def train_step(train_loader, tokenize, device, optimizer,scheduler, xlmr, classifier,logs,pooled_emb_do):\r\n",
        "    xlmr.train()\r\n",
        "    classifier.train()\r\n",
        "    iteration = 0\r\n",
        "    acc_sum=0\r\n",
        "    accuracy = []\r\n",
        "    losses = []\r\n",
        "    cms = []\r\n",
        "    for text, target, n_chunks in train_loader:\r\n",
        "        logs['lr'].append(get_lr(optimizer))\r\n",
        "        target = target.to(device)\r\n",
        "        flat_text = [item for sublist in text for item in sublist]\r\n",
        "        tokens = tokenize(flat_text)\r\n",
        "        tokens = tokens.to(device)\r\n",
        "        embeddings = xlmr.extract_features(tokens)\r\n",
        "        pooled_emb = torch.mean(embeddings, axis=1)\r\n",
        "        pooled_emb = nn.Dropout(pooled_emb_do)(pooled_emb)\r\n",
        "        # perform sum pooling\r\n",
        "        x = [s for s in torch.split(pooled_emb, n_chunks,dim=0)]\r\n",
        "        logit = classifier(x, n_chunks)\r\n",
        "        prob = torch.sigmoid(logit)\r\n",
        "        target = torch.reshape(target,shape=(-1,1))\r\n",
        "        loss = nn.BCELoss()(input=prob, target=target.float())\r\n",
        "        loss.backward()\r\n",
        "        optimizer.step()\r\n",
        "        if scheduler is not None:\r\n",
        "            scheduler.step()\r\n",
        "        optimizer.zero_grad()\r\n",
        "        # calculate accuracy\r\n",
        "        prob = torch.squeeze(prob)\r\n",
        "        prediction = torch.clone(prob)\r\n",
        "        prediction[prediction >= 0.5] = 1\r\n",
        "        prediction[prediction < 0.5] = 0\r\n",
        "        target = torch.squeeze(target)\r\n",
        "        acc = torch.sum(target==prediction)/float(len(target))\r\n",
        "        accuracy.append(float(acc))\r\n",
        "        losses.append(float(loss.detach().cpu().numpy()))\r\n",
        "        # calculate confusion matrix\r\n",
        "        cm = confusion_matrix(target.detach().cpu().numpy(),\r\n",
        "                              prediction.detach().cpu().numpy(),\r\n",
        "                              labels = np.array([0,1])\r\n",
        "                              )\r\n",
        "        cms.append(cm)\r\n",
        "        iteration+=1\r\n",
        "        print(f\"\\r iter: {iteration}/{len(train_loader)}\",end='')\r\n",
        "    return accuracy, losses, cms\r\n",
        "def val_step(val_loader, tokenize, device, xlmr, classifier, is_test=False):\r\n",
        "    print('')\r\n",
        "    print('Validating...')\r\n",
        "    xlmr.eval()\r\n",
        "    classifier.eval()\r\n",
        "    iteration = 0\r\n",
        "    accuracy = []\r\n",
        "    losses = []\r\n",
        "    cms = []\r\n",
        "    if is_test:\r\n",
        "        predictions,targets = [],[]\r\n",
        "    for text, target, n_chunks in val_loader:\r\n",
        "        target = target.to(device)\r\n",
        "        flat_text = [item for sublist in text for item in sublist]\r\n",
        "        tokens = tokenize(flat_text)\r\n",
        "        tokens = tokens.to(device)\r\n",
        "        with torch.no_grad():\r\n",
        "            embeddings = xlmr.extract_features(tokens)\r\n",
        "            pooled_emb = torch.mean(embeddings, axis=1)\r\n",
        "            # perform sum pooling\r\n",
        "            x = [s for s in torch.split(pooled_emb, n_chunks,dim=0)]\r\n",
        "            logit = classifier(x, n_chunks)\r\n",
        "            prob = torch.sigmoid(logit)\r\n",
        "        target = torch.reshape(target,shape=(-1,1))\r\n",
        "        loss = nn.BCELoss()(input=prob, target=target.float())\r\n",
        "        # calculate accuracy\r\n",
        "        prob = torch.squeeze(prob)\r\n",
        "        prediction = torch.clone(prob)\r\n",
        "        prediction[prediction >= 0.5] = 1\r\n",
        "        prediction[prediction < 0.5] = 0\r\n",
        "        target = torch.squeeze(target)\r\n",
        "        acc = torch.sum(target==prediction)/float(len(target))\r\n",
        "        accuracy.append(float(acc))\r\n",
        "        losses.append(float(loss.cpu().numpy()))\r\n",
        "        # calculate confusion matrix\r\n",
        "        cm = confusion_matrix(target.cpu().numpy(),\r\n",
        "                              prediction.cpu().numpy(),\r\n",
        "                              labels = np.array([0,1])\r\n",
        "                              )\r\n",
        "        cms.append(cm)\r\n",
        "        iteration+=1\r\n",
        "        print(f\"\\r iter: {iteration}/{len(val_loader)}\",end='')\r\n",
        "        if is_test:\r\n",
        "            predictions.append(prediction.cpu().numpy())\r\n",
        "            targets.append(target.cpu().numpy())\r\n",
        "    if is_test:\r\n",
        "        return accuracy, losses, cms, predictions, targets\r\n",
        "    return accuracy, losses, cms"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "G2iqGdw2xdh6"
      },
      "source": [
        "def save_to_disk(txt_path, values):\r\n",
        "    if os.path.isfile(txt_path):\r\n",
        "        os.remove(txt_path)\r\n",
        "    with open(txt_path, \"wb\") as fp:   \r\n",
        "        pickle.dump(values, fp)\r\n",
        "    return\r\n",
        "def load_from_disk(txt_path):\r\n",
        "    with open(txt_path, \"rb\") as f:\r\n",
        "        values =  pickle.load(f)\r\n",
        "    return values\r\n",
        "def save_checkpoint(xlmr, classifier, optimizer, logs, checkpoint_dir, epoch):\r\n",
        "    print('')\r\n",
        "    print('Saving checkpoint...')\r\n",
        "    state_dict = {\r\n",
        "        'classifier':classifier.state_dict(),\r\n",
        "        'optimizer': optimizer.state_dict(),\r\n",
        "        'xlmr': xlmr.state_dict()\r\n",
        "    }\r\n",
        "    torch.save(state_dict, os.path.join(checkpoint_dir, 'checkpoint_{}.pt'.format(epoch)))\r\n",
        "    save_to_disk(os.path.join(checkpoint_dir, 'logs.txt'),logs)\r\n",
        "    print(f'Checkpoint saved!')\r\n",
        "def load_checkpoint(checkpoint_dir, epoch, xlmr, classifier, device, optimizer=None):\r\n",
        "    pretrained_dict = torch.load(os.path.join(checkpoint_dir,'checkpoint_{}.pt'.format(epoch)),map_location=torch.device(device))\r\n",
        "    classifier.load_state_dict(pretrained_dict['classifier'])\r\n",
        "    xlmr.load_state_dict(pretrained_dict['xlmr'])\r\n",
        "    if optimizer is not None:\r\n",
        "        optimizer.load_state_dict(pretrained_dict['optimizer'])\r\n",
        "        return xlmr, classifier, optimizer\r\n",
        "    return xlmr,classifier"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EFr8kWC3M8f8"
      },
      "source": [
        "def get_lr(optimizer):\r\n",
        "    for param_group in optimizer.param_groups:\r\n",
        "        return param_group['lr']\r\n",
        "def set_lr(optimizer, lr):\r\n",
        "    for param_group in optimizer.param_groups:\r\n",
        "        param_group['lr'] = lr"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ZtFcETs2sPe6"
      },
      "source": [
        "from torch.nn.utils.rnn import pad_sequence, pack_padded_sequence, pad_packed_sequence\r\n",
        "class Classifier(nn.Module):\r\n",
        "    def __init__(self, lstm_size, emb_dim, out_dim_lin, lstm_do):\r\n",
        "        super().__init__()\r\n",
        "        self.lstm = nn.LSTM(input_size=emb_dim, hidden_size=lstm_size, batch_first=True)\r\n",
        "        self.lstm_do = nn.Dropout(lstm_do)\r\n",
        "        self.linear = nn.Linear(in_features=lstm_size, out_features=out_dim_lin)\r\n",
        "    def forward(self, x, n_chunks):\r\n",
        "        x = pad_sequence(x, batch_first=True, padding_value=0)\r\n",
        "        x = pack_padded_sequence(input=x, lengths=n_chunks, batch_first=True, enforce_sorted=False)\r\n",
        "        x, _ = self.lstm(x)\r\n",
        "        x, _ = pad_packed_sequence(x, batch_first=True)\r\n",
        "        x = x[:,-1,:]\r\n",
        "        x = self.lstm_do(x)\r\n",
        "        logit = self.linear(x)\r\n",
        "        return logit"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "u9AjISm0wNgU"
      },
      "source": [
        "## Hyperparameters"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VU6vlF2sXhVy"
      },
      "source": [
        "BATCH_SIZE = 2\r\n",
        "#TOKEN_MAX_LEN = 512 or 354\r\n",
        "EPOCHS = 100 # total number of epochs\r\n",
        "LR = 1e-7\r\n",
        "EMBEDDING_DIM = 768 # 768 for base and 1024 for large\r\n",
        "LSTM_SIZE = 128\r\n",
        "LSTM_DO = 0.2\r\n",
        "POOLED_EMB_DO = 0.3\r\n",
        "OUT_DIM_LIN = 1\r\n",
        "WEIGHT_DECAY = 1e-3\r\n",
        "#=============================\r\n",
        "# freeze xlmr hyperparams \r\n",
        "EPOCHS_FREEZE = 2\r\n",
        "NUM_WARMUP_EPOCHS_FREEZE = 1\r\n",
        "LR_FREEZE = 6e-5\r\n",
        "#=============================\r\n",
        "LR_LIN = 3e-6\r\n",
        "EPOCHS_LIN = 50"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "D5qG2GaMdIG3"
      },
      "source": [
        "# define Data Loaders\r\n",
        "train_dataset = MyDataset(df_train)\r\n",
        "train_loader = DataLoader(dataset=train_dataset, batch_size=BATCH_SIZE, shuffle=True, collate_fn=collate_func)\r\n",
        "val_dataset = MyDataset(df_val)\r\n",
        "val_loader = DataLoader(dataset=val_dataset, batch_size=BATCH_SIZE, shuffle=False, collate_fn=collate_func)\r\n",
        "test_dataset = MyDataset(df_test)\r\n",
        "test_loader = DataLoader(dataset=test_dataset, batch_size=BATCH_SIZE, shuffle=False, collate_fn=collate_func)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "RjBy_NZawR-o"
      },
      "source": [
        "## Train loop"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ks9WeFB5LDVY"
      },
      "source": [
        "checkpoint_dir = os.path.join(os.getcwd(),'checkpoints')\r\n",
        "os.makedirs(checkpoint_dir, exist_ok=True)\r\n",
        "\r\n",
        "device = 'cuda' if torch.cuda.is_available() else 'cpu'\r\n",
        "print('Selected device is {}'.format(device))\r\n",
        "xlmr = xlmr.to(device)\r\n",
        "classifier = Classifier(lstm_size=LSTM_SIZE,\r\n",
        "                        emb_dim=EMBEDDING_DIM,\r\n",
        "                        out_dim_lin=OUT_DIM_LIN,\r\n",
        "                        lstm_do = LSTM_DO\r\n",
        "                        ).to(device)\r\n",
        "print('Number of trainable params: ',sum(p.numel() for p in xlmr.parameters())+\\\r\n",
        "      sum(p.numel() for p in classifier.parameters()))\r\n",
        "# define optimizer\r\n",
        "params = list(xlmr.parameters()) + list(classifier.parameters())\r\n",
        "# start with frozen XLMR model (it should be left in train mode)\r\n",
        "for p in xlmr.parameters():\r\n",
        "    p.requires_grad=False\r\n",
        "optimizer = SGD(params, lr=LR_FREEZE, weight_decay=WEIGHT_DECAY)\r\n",
        "# define learning rate scheduler\r\n",
        "scheduler = get_linear_schedule_with_warmup(\r\n",
        "    optimizer,\r\n",
        "    num_warmup_steps=NUM_WARMUP_EPOCHS_FREEZE*len(train_loader),\r\n",
        "    num_training_steps=EPOCHS_FREEZE*len(train_loader)\r\n",
        "    )\r\n",
        "# wrap tokenizer\r\n",
        "tokenize = wrap_tokenizer(tokenizer)\r\n",
        "logs = {'train_acc':[],'train_loss':[],\r\n",
        "        'val_acc':[],'val_loss':[], 'lr':[],\r\n",
        "        'train_cm':[],'val_cm':[]}\r\n",
        "for epoch in range(EPOCHS):\r\n",
        "    if epoch == EPOCHS_FREEZE:\r\n",
        "        print('XLM-R model is unfrozen!')\r\n",
        "        print(f'Learning rate is set to {LR_LIN}!')\r\n",
        "        for p in xlmr.parameters():\r\n",
        "             p.requires_grad=True\r\n",
        "        optimizer = AdamW(params, lr=LR_LIN, weight_decay=WEIGHT_DECAY)\r\n",
        "        scheduler = None\r\n",
        "    if epoch == EPOCHS_FREEZE + EPOCHS_LIN:\r\n",
        "        print(f'Learning rate is set to {LR}!')\r\n",
        "        set_lr(optimizer, LR)\r\n",
        "        scheduler = None\r\n",
        "    train_acc, train_loss, train_cm = train_step(\r\n",
        "        train_loader=train_loader,\r\n",
        "        tokenize=tokenize, \r\n",
        "        device=device, \r\n",
        "        optimizer=optimizer, \r\n",
        "        scheduler=scheduler,\r\n",
        "        xlmr=xlmr,\r\n",
        "        classifier=classifier,\r\n",
        "        logs=logs,\r\n",
        "        pooled_emb_do = POOLED_EMB_DO\r\n",
        "        )\r\n",
        "    val_acc, val_loss, val_cm = val_step(\r\n",
        "        val_loader=val_loader,\r\n",
        "        tokenize=tokenize, \r\n",
        "        device=device, \r\n",
        "        xlmr=xlmr,\r\n",
        "        classifier=classifier\r\n",
        "        )\r\n",
        "    logs['train_acc'] += train_acc\r\n",
        "    logs['train_loss'] += train_loss\r\n",
        "    logs['val_acc'] += val_acc\r\n",
        "    logs['val_loss'] += val_loss\r\n",
        "    logs['train_cm'] += train_cm\r\n",
        "    logs['val_cm'] += val_cm \r\n",
        "    if epoch % 2 == 0 and epoch != 0 :\r\n",
        "        save_checkpoint(\r\n",
        "            xlmr=xlmr,\r\n",
        "            classifier=classifier, \r\n",
        "            optimizer=optimizer, \r\n",
        "            logs=logs, \r\n",
        "            checkpoint_dir=checkpoint_dir, \r\n",
        "            epoch=epoch\r\n",
        "            )\r\n",
        "    print('')\r\n",
        "    print(f\"Epoch {epoch} --> loss:{mean(train_loss):.4f},\\\r\n",
        "        acc:{mean(train_acc):.2f}%,\\\r\n",
        "        val_loss:{mean(val_loss):.4f},\\\r\n",
        "        val_acc:{mean(val_acc):.2f}%\\\r\n",
        "        \\ntrain_cm:\\n {np.mean(train_cm, axis=0)}\\\r\n",
        "        \\nval_cm:\\n {np.mean(val_cm,axis=0)}\\\r\n",
        "        \")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zlWyquDWiBdn"
      },
      "source": [
        "torch.cuda.empty_cache()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DkuR7bsb06pA"
      },
      "source": [
        "## Test"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-zG2R5K7CQdl"
      },
      "source": [
        "### Load pretrained weights"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5eLsBC-X09Yt"
      },
      "source": [
        "import os\r\n",
        "import pandas as pd\r\n",
        "import numpy as np \r\n",
        "from ast import literal_eval\r\n",
        "from sklearn.metrics import confusion_matrix\r\n",
        "import seaborn as sn\r\n",
        "import matplotlib.pyplot as plt\r\n",
        "\r\n",
        "# load pretrained weights\r\n",
        "device = 'cuda' if torch.cuda.is_available() else 'cpu'\r\n",
        "checkpoint_dir = os.path.join(os.getcwd(),'checkpoints')\r\n",
        "best_dir = os.path.join(checkpoint_dir,'best-MR')\r\n",
        "classifier = Classifier(LSTM_SIZE,EMBEDDING_DIM, OUT_DIM_LIN, LSTM_DO)\r\n",
        "xlmr, classifier = load_checkpoint(best_dir,16,xlmr,classifier,device)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ePfSInX1CUd-"
      },
      "source": [
        "### Load test set and pretrained weights"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wwpd7oxi74nE"
      },
      "source": [
        "# load test set\r\n",
        "df_test = pd.read_csv(os.path.join(best_dir,'test.csv'))\r\n",
        "df_test.drop(columns='Unnamed: 0', inplace=True)\r\n",
        "df_test.Text = df_test.Text.apply(lambda x: literal_eval(x))\r\n",
        "test_dataset = MyDataset(df_test)\r\n",
        "test_loader = DataLoader(dataset=test_dataset, batch_size=BATCH_SIZE, shuffle=False, collate_fn=collate_func)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nCGdLLQa4qNW"
      },
      "source": [
        "### Test accuracy"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3ksxT0i45c8n",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "50072ad6-f68e-411d-e1c9-bded4c9bb1c9"
      },
      "source": [
        "tokenize = wrap_tokenizer(tokenizer)\r\n",
        "classifier = classifier.to(device)\r\n",
        "xlmr = xlmr.to(device)\r\n",
        "test_acc, test_loss,cms, predictions, targets = val_step(\r\n",
        "        val_loader=test_loader,\r\n",
        "        tokenize=tokenize, \r\n",
        "        device=device, \r\n",
        "        xlmr=xlmr,\r\n",
        "        classifier=classifier,\r\n",
        "        is_test=True\r\n",
        "        )\r\n",
        "print('')\r\n",
        "print(f\"test loss:{mean(test_loss):.4f}\\n test acc:{mean(test_acc):.2f}%\")"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n",
            "Validating...\n",
            " iter: 84/84\n",
            "test loss:0.4466\n",
            " test acc:0.79%\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "GS-LKm72sGbj"
      },
      "source": [
        "State of the art result (86.11% cross-validation accuracy) for this task on the sam dataset can be found [here](https://scindeks-clanci.ceon.rs/data/pdf/1821-3251/2017/1821-32511702104B.pdf). I believe that this score can be achieved with better tweaking of hyperparameters (in which I have not invested much time) and XLM-R large model. Also you have to keep in mind that the test score depends on sentiment intensity of examples in the test set. This is shown below."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vsbR6z19zWkQ"
      },
      "source": [
        "### Plot confusion matrix"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 273
        },
        "id": "IguKoV6yFVPx",
        "outputId": "fa58ba67-20e2-4a85-e8c3-8a9fbada0871"
      },
      "source": [
        "preds = np.concatenate(predictions)\r\n",
        "labels = np.concatenate(targets)\r\n",
        "\r\n",
        "from sklearn.metrics import confusion_matrix\r\n",
        "import seaborn as sn\r\n",
        "\r\n",
        "cm = confusion_matrix(labels, preds)\r\n",
        "names = ['negative', 'positive']\r\n",
        "df_cm = pd.DataFrame(cm, index=names, columns=names)\r\n",
        "sn.set(font_scale=1.4) # for label size\r\n",
        "sn.heatmap(df_cm, annot=True, annot_kws={\"size\": 16}) # font size\r\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "71zYI8s0SDnp"
      },
      "source": [
        "### Examine the influence of weak sentiment on wrong classification."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gs6j7ZL68BrI"
      },
      "source": [
        "# find the Scores in the original .csv file\r\n",
        "df_orig = pd.read_csv(os.path.join(movies_path,'MR.csv'))\r\n",
        "df_test['Mark'] = None\r\n",
        "for i in range(len(df_test)):\r\n",
        "    text = df_test.Text.iloc[i][0][:50]\r\n",
        "    found = False\r\n",
        "    for j in range(len(df_orig)):\r\n",
        "        if text in df_orig.Text.iloc[j]:\r\n",
        "            mark_id = df_test.columns.get_loc('Mark') \r\n",
        "            df_test.iloc[i,mark_id] = df_orig.Mark.iloc[j]\r\n",
        "            found = True\r\n",
        "            break  \r\n",
        "    if not found:\r\n",
        "        mark_id = df_test.columns.get_loc('Mark') \r\n",
        "        df_test.iloc[i,mark_id] = 0"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 333
        },
        "id": "KHtIN6q_Nrrn",
        "outputId": "8bc8074c-9524-4d4f-ae33-6861f5c7b225"
      },
      "source": [
        "height = df_test.loc[preds!=labels,'Mark'].value_counts()\r\n",
        "height = height.to_dict()\r\n",
        "values = np.arange(1,11)\r\n",
        "for value in values:\r\n",
        "    if value not in height.keys():\r\n",
        "        height[value] = 0\r\n",
        "del height[0]\r\n",
        "plt.xlabel('Score')\r\n",
        "plt.ylabel('Number of examples')\r\n",
        "plt.xticks(values)\r\n",
        "plt.title('Number of wrongly classified examples per Score')\r\n",
        "plt.bar(height.keys(), height=height.values())\r\n",
        "plt.axvline(x=4.5, linewidth=4, color='black')\r\n",
        "plt.axvline(x=6.5, linewidth=4, color='black')\r\n",
        "plt.text(8, 7.5, 'Positive', color='blue', \r\n",
        "        bbox=dict(facecolor='none', edgecolor='black', boxstyle='round,pad=1'),\r\n",
        "        fontsize=10)\r\n",
        "plt.text(4.875, 7.5, 'Neutral', color='black', \r\n",
        "        bbox=dict(facecolor='none', edgecolor='black', boxstyle='round,pad=1'),\r\n",
        "        fontsize=10)\r\n",
        "plt.text(1, 7.5, 'Negative', color='red', \r\n",
        "        bbox=dict(facecolor='none', edgecolor='black', boxstyle='round,pad=1'),\r\n",
        "        fontsize=10)\r\n",
        "print('')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1N7WfTeORp18"
      },
      "source": [
        "We can see that most of the wrongly classified examples have weak sentiment (score close to 5 and 6)."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ekq8jXQPCdzu"
      },
      "source": [
        "### Show training curves"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "mRq5EhA4CmM9"
      },
      "source": [
        "SHOW_EPOCHS = 17\r\n",
        "\r\n",
        "logs = load_from_disk(os.path.join(best_dir,'logs.txt'))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DDg-trlL8Fat"
      },
      "source": [
        "def average(data, frequency, start_avg_point=0):\r\n",
        "    average_data = []\r\n",
        "    if (len(data) > start_avg_point):\r\n",
        "        round_num = -(len(str(frequency))-1)\r\n",
        "        for i in range(0,round(len(data),round_num)-frequency,frequency):\r\n",
        "            avg_data=mean(data[i:i+frequency])\r\n",
        "            average_data.append(avg_data)\r\n",
        "    return average_data"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fusyUA0hENhr"
      },
      "source": [
        "train_acc = average(logs['train_acc'],677)[:SHOW_EPOCHS]\r\n",
        "val_acc = average(logs['val_acc'],84)[:SHOW_EPOCHS]\r\n",
        "lr = logs['lr'][:677*(SHOW_EPOCHS-1)]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_Mrq4rSqzazW"
      },
      "source": [
        "Show accuracy through epochs"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 333
        },
        "id": "p7QTLTMz8FlB",
        "outputId": "c9b2ad4b-442c-4c8e-93a9-08c9343ac9b5"
      },
      "source": [
        "plt.plot(train_acc)\r\n",
        "plt.plot(val_acc)\r\n",
        "plt.title('Training curves')\r\n",
        "plt.xlabel('epoch')\r\n",
        "plt.legend(['train_acc','val_acc'])\r\n",
        "print('')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 321
        },
        "id": "DJUhJSNw_BRq",
        "outputId": "e791585b-afcf-4ba6-a31b-4c9aadf2e185"
      },
      "source": [
        "fig = plt.figure()\r\n",
        "ax = fig.add_subplot(1, 1, 1)\r\n",
        "ax.plot(lr)\r\n",
        "ax.set_title('Learning rate',fontsize=15)\r\n",
        "minor_ticks = np.arange(-1, 677*SHOW_EPOCHS, 677)\r\n",
        "ax.set_xticks(minor_ticks, minor=True)\r\n",
        "ax.set_xticklabels(labels=range(0,17),minor=True)\r\n",
        "ax.grid(which='minor', alpha=0.7)\r\n",
        "xax = ax.get_xaxis()\r\n",
        "ax.set_xlabel('global step\\nepoch',labelpad=-25,horizontalalignment='right', x=-0.02, fontsize=12)\r\n",
        "xax.set_tick_params(which='minor', pad=17,labelsize=12)\r\n",
        "xax.set_tick_params(which='major',labelsize=12)\r\n",
        "ax.get_yaxis().set_tick_params(labelsize=12)\r\n",
        "ax.axvline(x=2*677, linewidth=4, color='black')\r\n",
        "ax.text(0, 4e-5, 'Frozen', color='black', \r\n",
        "        bbox=dict(facecolor='none', edgecolor='black', boxstyle='round,pad=1'),\r\n",
        "        fontsize=12)\r\n",
        "ax.text(6000, 4e-5, 'Fine-Tuning', color='black', \r\n",
        "        bbox=dict(facecolor='none', edgecolor='black', boxstyle='round,pad=1'),\r\n",
        "        fontsize=12)\r\n",
        "print('')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 323
        },
        "id": "VLz_Uz_ADCof",
        "outputId": "69d7279b-6d31-4f1a-fd05-25e116d2f3e1"
      },
      "source": [
        "fig = plt.figure()\r\n",
        "ax1=fig.add_subplot(111, label=\"1\")\r\n",
        "l1,=ax1.plot(train_acc,color='blue')\r\n",
        "l2,=ax1.plot(val_acc,color='orange')\r\n",
        "ax1.legend(['train acc','val_acc'],loc='lower right',fontsize=12)\r\n",
        "ax=fig.add_subplot(111, label=\"3\",frame_on=False)\r\n",
        "ax.plot(lr, color='black')\r\n",
        "ax.set_title('Training curves')\r\n",
        "#ax.set_xlabel('epoch')\r\n",
        "ax.set_xticklabels([])\r\n",
        "ax.set_yticklabels([])\r\n",
        "ax.set_xticks([])\r\n",
        "ax.set_yticks([])\r\n",
        "ax.legend(['learning rate'],loc='upper left',fontsize=12)\r\n",
        "plt.xlabel('epoch',labelpad=20)\r\n",
        "print('')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    }
  ]
}