LibTest.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import math, time\n",
    "from datetime import datetime\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "from torch.autograd import Variable\n",
    "import torchtext, spacy\n",
    "from torchtext.data import Field, BucketIterator\n",
    "from torchtext.data.metrics import bleu_score\n",
    "from torch import optim\n",
    "\n",
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from lib.transformers import *\n",
    "from lib.training import *\n",
    "from lib.evaluation import *\n",
    "from lib.datasets import *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "bs = 64\n",
    "field_src, field_trg, src_vocab, trg_vocab, src_pad_idx, trg_pad_idx, sp_src, sp_trg, device, data, iterators = get_multi30k(bs)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Trainer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "model, optimizer, criterion = build_transformer(src_vocab, trg_vocab, src_pad_idx, trg_pad_idx, device)\n",
    "trainer = Trainer(model, iterators, criterion, optimizer, device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch.utils.tensorboard import SummaryWriter\n",
    "\n",
    "writer = SummaryWriter(f'.tf_runs/LibTest')\n",
    "writer.close()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "eps = 10"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "trainer.train_log(eps, writer, f'{datetime.now().strftime(\"%Y-%m-%d %H-%M\")}', \"T-\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "VerboseTrainer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "model, optimizer, criterion = build_transformer(src_vocab, trg_vocab, src_pad_idx, trg_pad_idx, device)\n",
    "trainer = VerboseTrainer(model, iterators, criterion, optimizer, device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "eps = 10"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Epoch 0: | Time: 1m 23s\n",
      "Train loss: 6.0353017630556085\n",
      "Valid loss: 4.218024522066116\n",
      "\n",
      "Epoch 1: | Time: 1m 22s\n",
      "Train loss: 3.8261806203405238\n",
      "Valid loss: 3.2277660071849823\n",
      "\n",
      "Epoch 2: | Time: 1m 22s\n",
      "Train loss: 3.093068240497606\n",
      "Valid loss: 2.6603732109069824\n",
      "\n",
      "Epoch 3: | Time: 1m 23s\n",
      "Train loss: 2.565169703592813\n",
      "Valid loss: 2.282218113541603\n",
      "\n",
      "Epoch 4: | Time: 1m 23s\n",
      "Train loss: 2.169322951249614\n",
      "Valid loss: 1.97156273573637\n",
      "\n",
      "Epoch 5: | Time: 1m 26s\n",
      "Train loss: 1.9188138173015108\n",
      "Valid loss: 1.7927093803882599\n",
      "\n",
      "Epoch 6: | Time: 1m 22s\n",
      "Train loss: 1.7767694797284803\n",
      "Valid loss: 1.7358791306614876\n",
      "\n",
      "Epoch 7: | Time: 1m 23s\n",
      "Train loss: 1.6956382081372097\n",
      "Valid loss: 1.7434918284416199\n",
      "\n",
      "Epoch 8: | Time: 1m 23s\n",
      "Train loss: 1.671780839102909\n",
      "Valid loss: 1.6617712453007698\n",
      "\n",
      "Epoch 9: | Time: 1m 22s\n",
      "Train loss: 1.6172071298838713\n",
      "Valid loss: 1.6636941581964493\n",
      "\n",
      "Total Time: 13m 54s\n",
      "Final test loss: 1.6861108019948006\n"
     ]
    }
   ],
   "source": [
    "trainer.train(eps, 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Evaluate Trainer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['drei', 'köche', 'spielen', 'fußball', 'fußball', 'fußball', 'fußball', 'fußball', 'fußball', 'fußball', 'fußball', 'fußball', 'fußball', 'fußball', 'fußball', 'fußball', 'fußball', 'fußball', 'fußball', 'fußball']\n"
     ]
    }
   ],
   "source": [
    "translator = Translator(model, sp_src, field_src, field_trg, device)\n",
    "translation = translator.translate(\"Three brothers are playing football\")\n",
    "print(translation)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "22.996988892555237\n"
     ]
    }
   ],
   "source": [
    "bleu = translator.calculate_bleu(data[2])\n",
    "print(bleu)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "TranslatingTrainer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "model, optimizer, criterion = build_transformer(src_vocab, trg_vocab, src_pad_idx, trg_pad_idx, device)\n",
    "translator = Translator(model, sp_src, field_src, field_trg, device)\n",
    "trainer = TranslatingTrainer(model, iterators, criterion, optimizer, translator, device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Epoch 0: | Time: 1m 33s\n",
      "Train loss: 6.037329666414975\n",
      "Valid loss: 4.17379991710186\n",
      "['zwei', 'männer', 'in', 'einem', '<unk>', '.', '<eos>']\n",
      "\n",
      "Epoch 1: | Time: 1m 29s\n",
      "Train loss: 3.81827574206869\n",
      "Valid loss: 3.254710853099823\n",
      "['drei', 'hunde', 'spielen', 'im', 'freien', '.', '<eos>']\n",
      "\n",
      "Epoch 2: | Time: 1m 31s\n",
      "Train loss: 3.139283850854714\n",
      "Valid loss: 2.694185122847557\n",
      "['drei', 'jungen', 'spielen', 'drei', 'jungen', '.', '<eos>']\n",
      "\n",
      "Epoch 3: | Time: 1m 32s\n",
      "Train loss: 2.6144469530046774\n",
      "Valid loss: 2.283024236559868\n",
      "['drei', '<unk>', 'spielen', 'fußball', 'fußball', '.', '<eos>']\n",
      "\n",
      "Epoch 4: | Time: 1m 36s\n",
      "Train loss: 2.1994088410280876\n",
      "Valid loss: 1.9997878074645996\n",
      "['drei', '<unk>', 'spielen', 'fußball', 'fußball', ',', 'die', 'fußball', 'spielen', '.', '<eos>']\n",
      "\n",
      "Epoch 5: | Time: 1m 36s\n",
      "Train loss: 1.9343254928546856\n",
      "Valid loss: 1.85004024207592\n",
      "['drei', 'ringer', 'spielen', 'fußball', ',', 'während', 'eines', 'fußballspieler', 'bei', 'dem', 'fußball', 'spielen', '.', '<eos>']\n",
      "\n",
      "Epoch 6: | Time: 1m 30s\n",
      "Train loss: 1.7799442799080836\n",
      "Valid loss: 1.768664687871933\n",
      "['drei', '<unk>', 'spielen', 'football', 'football', 'football', 'football', 'football', 'football', 'football', '.', '<eos>']\n",
      "\n",
      "Epoch 7: | Time: 1m 29s\n",
      "Train loss: 1.696174480578973\n",
      "Valid loss: 1.6992519870400429\n",
      "['drei', 'kajakfahrer', 'spielen', 'football', 'football', 'football', 'football', 'football', '.', '<eos>']\n",
      "\n",
      "Epoch 8: | Time: 1m 30s\n",
      "Train loss: 1.6664754283585737\n",
      "Valid loss: 1.684781439602375\n",
      "['drei', '<unk>', 'spielen', 'football', ',', 'drei', 'football', 'spielen', '.', '<eos>']\n",
      "\n",
      "Epoch 9: | Time: 1m 28s\n",
      "Train loss: 1.603775681115457\n",
      "Valid loss: 1.6144917011260986\n",
      "['drei', 'kajakfahrer', 'spielen', 'football', 'fußball', 'und', 'spielen', 'den', 'football', '.', '<eos>']\n",
      "\n",
      "Epoch 10: | Time: 1m 22s\n",
      "Train loss: 1.5207269186490433\n",
      "Valid loss: 1.6450646072626114\n",
      "['drei', 'chirurgen', 'spielen', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', '.', '<eos>']\n",
      "\n",
      "Epoch 11: | Time: 1m 20s\n",
      "Train loss: 1.4391635091819428\n",
      "Valid loss: 1.544962964951992\n",
      "['drei', 'segelboote', 'spielen', 'football', '.', '<eos>']\n",
      "\n",
      "Epoch 12: | Time: 1m 24s\n",
      "Train loss: 1.358447353220196\n",
      "Valid loss: 1.477102406322956\n",
      "['drei', '<unk>', 'spielen', 'football', 'football', 'football', '.', 'drei', 'schüler', 'spielen', 'football', 'football', 'football', 'football', '.', '<eos>']\n",
      "\n",
      "Epoch 13: | Time: 1m 21s\n",
      "Train loss: 1.2809442200324608\n",
      "Valid loss: 1.4925240650773048\n",
      "['drei', 'chirurgen', 'spielen', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football']\n",
      "\n",
      "Epoch 14: | Time: 1m 23s\n",
      "Train loss: 1.2220981555625732\n",
      "Valid loss: 1.4860398545861244\n",
      "['drei', 'dunkelhaarige', 'schüler', 'spielen', 'football', 'football', 'football', '.', '<eos>']\n",
      "\n",
      "Epoch 15: | Time: 1m 22s\n",
      "Train loss: 1.1609355645820434\n",
      "Valid loss: 1.4644689932465553\n",
      "['drei', 'sportler', 'spielen', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football', 'football']\n",
      "\n",
      "Epoch 16: | Time: 1m 24s\n",
      "Train loss: 1.0960785627102536\n",
      "Valid loss: 1.4674052000045776\n",
      "['drei', 'mönche', 'spielen', 'football', 'football', 'football', 'football', 'football', 'football', '.', '<eos>']\n",
      "\n",
      "Epoch 17: | Time: 1m 22s\n",
      "Train loss: 1.0481002594668434\n",
      "Valid loss: 1.485005035996437\n",
      "['drei', 'motocross-fahrer', 'spielen', 'football', ',', 'die', 'football', 'spielen', '.', '<eos>']\n",
      "\n",
      "Epoch 18: | Time: 1m 21s\n",
      "Train loss: 1.0057906402102652\n",
      "Valid loss: 1.465163715183735\n",
      "['drei', 'angestellte', 'spielen', 'football', ',', 'spielen', 'football', 'und', 'spielen', 'football', '.', '<eos>']\n",
      "\n",
      "Total Time: 27m 42s\n",
      "Final test loss: 1.5252329781651497\n",
      "['drei', 'angestellte', 'spielen', 'football', ',', 'spielen', 'football', 'und', 'spielen', 'football', '.', '<eos>']\n"
     ]
    }
   ],
   "source": [
    "trainer.train()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "trainer.plot_loss()"
   ]
  }
 ],
 "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": "3.9.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}