{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "from gensim.test.utils import common_texts\n",
    "from gensim.models import FastText\n",
    "import os.path\n",
    "import gzip\n",
    "import shutil\n",
    "import torch\n",
    "import torch.optim as optim"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "features = 100\n",
    "batch_size = 16\n",
    "criterion = torch.nn.BCELoss()\n",
    "\n",
    "with gzip.open('train/train.tsv.gz', 'rb') as f_in:\n",
    "    with open('train/train.tsv', 'wb') as f_out:\n",
    "        shutil.copyfileobj(f_in, f_out)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0        [mindaugas, budzinauskas, wierzy, w, odbudowę,...\n",
       "1        [przyjmujący, reprezentacji, polski, wrócił, d...\n",
       "2        [fen, 9:, zapowiedź, walki, róża, gumienna, vs...\n",
       "3        [aleksander, filipiak:, czuję, się, dobrze, w,...\n",
       "4        [victoria, carl, i, aleksiej, czerwotkin, mist...\n",
       "                               ...                        \n",
       "98127    [kamil, syprzak, zaczyna, kolekcjonować, trofe...\n",
       "98128    [holandia:, dwa, gole, piotra, parzyszka, piot...\n",
       "98129    [sparingowo:, korona, gorsza, od, stali., lett...\n",
       "98130    [vive, -, wisła., ośmiu, debiutantów, w, tegor...\n",
       "98131    [wta, miami:, timea, bacsinszky, pokonana,, sw...\n",
       "Name: Text, Length: 98132, dtype: object"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data = pd.read_csv('train/train.tsv', sep='\\t', names=[\"Ball\",\"Text\"])\n",
    "data[\"Text\"] = data[\"Text\"].str.lower().str.split()\n",
    "data[\"Text\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "ft_model = None\n",
    "if not os.path.isfile('fasttext.model'):\n",
    "    ft_model = FastText(size=features, window=3, min_count=1)\n",
    "    ft_model.build_vocab(sentences=data[\"Text\"])\n",
    "    ft_model.train(data[\"Text\"], total_examples=len(data[\"Text\"]), epochs=10)\n",
    "    ft_model.save(\"fasttext.model\")\n",
    "else:\n",
    "    ft_model = FastText.load(\"fasttext.model\")\n",
    "    \n",
    "def document_vector(doc):\n",
    "    result = ft_model.wv[doc]\n",
    "    return np.max(result, axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "X = [document_vector(x) for x in data[\"Text\"]]\n",
    "Y = data[\"Ball\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "class NeuralNetworkModel(torch.nn.Module):\n",
    "    def __init__(self):\n",
    "        super(NeuralNetworkModel, self).__init__()\n",
    "        self.fc1 = torch.nn.Linear(features,200)\n",
    "        self.fc2 = torch.nn.Linear(200,150)\n",
    "        self.fc3 = torch.nn.Linear(150,1)\n",
    "\n",
    "    def forward(self, x):\n",
    "        x = self.fc1(x)\n",
    "        x = torch.relu(x)\n",
    "        x = self.fc2(x)\n",
    "        x = torch.sigmoid(x)\n",
    "        x = self.fc3(x)\n",
    "        x = torch.sigmoid(x)\n",
    "        return x\n",
    "\n",
    "    \n",
    "def get_loss_acc(model, X_dataset, Y_dataset):\n",
    "    loss_score = 0\n",
    "    acc_score = 0\n",
    "    items_total = 0\n",
    "    model.eval()\n",
    "    for i in range(0, Y_dataset.shape[0], batch_size):\n",
    "        x = X_dataset[i:i+batch_size]\n",
    "        x = torch.tensor(x)\n",
    "        y = Y_dataset[i:i+batch_size]\n",
    "        y = torch.tensor(y.astype(np.float32).to_numpy()).reshape(-1,1)\n",
    "        y_predictions = model(x)\n",
    "        acc_score += torch.sum((y_predictions >= 0.5) == y).item()\n",
    "        items_total += y.shape[0] \n",
    "\n",
    "        loss = criterion(y_predictions, y)\n",
    "\n",
    "        loss_score += loss.item() * y.shape[0] \n",
    "    return (loss_score / items_total), (acc_score / items_total)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "model_path = 'nn.model'\n",
    "nn_model = NeuralNetworkModel()\n",
    "    \n",
    "if not os.path.isfile(model_path):\n",
    "    optimizer = optim.SGD(nn_model.parameters(), lr=0.1)\n",
    "\n",
    "    display(get_loss_acc(nn_model, X, Y))\n",
    "    for epoch in range(5):\n",
    "        nn_model.train()\n",
    "        for i in range(0, len(X), batch_size):\n",
    "            x = X[i:i+batch_size]\n",
    "            x = torch.tensor(x)\n",
    "\n",
    "            y = Y[i:i+batch_size]\n",
    "            y = torch.tensor(y.astype(np.float32).to_numpy()).reshape(-1,1)\n",
    "\n",
    "            y_predictions = nn_model(x)\n",
    "            loss = criterion(y_predictions, y)\n",
    "\n",
    "            optimizer.zero_grad()\n",
    "            loss.backward()\n",
    "            optimizer.step()\n",
    "        display(get_loss_acc(nn_model, X, Y))\n",
    "    torch.save(nn_model.state_dict(), model_path)\n",
    "else:\n",
    "    nn_model.load_state_dict(torch.load(model_path))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_dev = pd.read_csv('dev-0/in.tsv', sep='\\t', names=[\"Text\"])[\"Text\"]\n",
    "y_dev = pd.read_csv('dev-0/expected.tsv', sep='\\t', names=[\"Ball\"])[\"Ball\"]\n",
    "x_dev = [document_vector(x) for x in x_dev.str.lower().str.split()]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.45761072419184756, 0.7694424064563463)"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "get_loss_acc(nn_model, x_dev, y_dev)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_dev_prediction = nn_model(torch.tensor(x_dev))\n",
    "y_dev_prediction = np.array([round(y) for y in y_dev_prediction.flatten().tolist()])\n",
    "np.savetxt(\"dev-0/out.tsv\", y_dev_prediction,  fmt='%d')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_test = pd.read_csv('test-A/in.tsv', sep='\\t', names=[\"Text\"])[\"Text\"]\n",
    "x_test = [document_vector(x) for x in x_test.str.lower().str.split()]\n",
    "y_test_prediction = nn_model(torch.tensor(x_test))\n",
    "y_test_prediction = np.array([round(y) for y in y_test_prediction.flatten().tolist()])\n",
    "np.savetxt(\"test-A/out.tsv\", y_test_prediction,  fmt='%d')"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}