example.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Classifying Sheet Music\n",
    "\n",
    "An example notebook walking through each step in starting with a PDF page of sheet music and making a prediction on its composer"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Imports"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import pickle\n",
    "import subprocess\n",
    "import torch\n",
    "import numpy as np\n",
    "import utils.encoders as encoders\n",
    "\n",
    "from pathlib import Path\n",
    "from hmc_mir.bootleg_score import bootleg_score\n",
    "from transformers import AutoModelForSequenceClassification, AutoTokenizer"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Configurations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'Length of each fragment to inference on: int'"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "PDF = Path(\"bach.pdf\")\n",
    "'''PDF file path: PosixPath'''\n",
    "PAGE = 0\n",
    "'''Page index (0-indexed): int'''\n",
    "MODEL = Path(\"/home/ajain/ttmp/EWLLMs/experiments/100_way_linear_probes/dense_1_1/checkpoint-13128\")\n",
    "'''Model path: PosixPath'''\n",
    "FRAGMENT_LENGTH = 64\n",
    "'''Length of each fragment to inference on: int'''"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Extracting Bootleg Score From PDF"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Convert PDF page to PNG"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Converting bach.pdf\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "convert-im6.q16: profile 'icc': 'RGB ': RGB color space not permitted on grayscale PNG `bach.png' @ warning/png.c/MagickPNGWarningHandler/1667.\n"
     ]
    }
   ],
   "source": [
    "def convertPDF2PNG(pdffile, pngfile):\n",
    "    pngfile.parent.mkdir(parents=True, exist_ok=True)\n",
    "\n",
    "    print('Converting {}'.format(pdffile))\n",
    "    try:\n",
    "        subprocess.call(['convert', '-density', '300', '-alpha', 'remove', '-resize', '2550', pdffile, pngfile])\n",
    "    except FileNotFoundError:\n",
    "        pass\n",
    "png_path = PDF.parent / (PDF.stem + '.png')\n",
    "convertPDF2PNG(PDF, png_path)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Get bootleg score from PNG"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Processing bach-0.png\n"
     ]
    }
   ],
   "source": [
    "img_path = png_path.parent / (png_path.stem + '-' + str(PAGE) + png_path.suffix)\n",
    "bscore_path = img_path.parent / (img_path.stem + '.pkl')\n",
    "bootleg_score.processImageFile(img_path, bscore_path)\n",
    "\n",
    "with open(bscore_path, 'rb') as f:\n",
    "    bscore = pickle.load(f)['bscore']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Visualize bootleg score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1000x1000 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "bootleg_score.visualizeBootlegScore(bscore, lines = [13, 15, 17, 19, 21, 35, 37, 39, 41, 43])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Classify Page"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create 64 length fragments with 50% overlap for ensemble prediction in shape (64, # Possible Note Events)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "fragments = []\n",
    "for i in range(0, bscore.shape[1], int(FRAGMENT_LENGTH/2)):\n",
    "    fragments.append(bscore[:,i:i+FRAGMENT_LENGTH].T)\n",
    "    if i + FRAGMENT_LENGTH >= bscore.shape[1]:\n",
    "        break"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Encode fragments"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "fragments = [encoders.dense_encoder(fragment, block_size=[1,1]) for fragment in fragments]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Make fragment-level predictions with model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([-0.2461, -0.3558, -0.2287, -0.3302, -0.4072, -0.3212, -0.3867, -0.3994],\n",
       "       device='cuda:0')"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "classifier = AutoModelForSequenceClassification.from_pretrained(MODEL).to('cuda')\n",
    "tokenizer = AutoTokenizer.from_pretrained(MODEL)\n",
    "preds = []\n",
    "classifier.eval()\n",
    "with torch.no_grad():\n",
    "    for fragment in fragments:\n",
    "        input_ids = tokenizer.encode(fragment, return_tensors='pt').to('cuda')\n",
    "        logits = classifier(input_ids)[0]\n",
    "        preds.append(logits)\n",
    "torch.mean(torch.cat(preds), dim=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open('/mnt/data0/BSCRC/data/9_way_dataset.pkl', \"rb\") as f:\n",
    "    data = pickle.load(f)\n",
    "preds = np.load('/home/ajain/ttmp/EWLLMs/experiments/9_way_linear_probes/dense_1_1/test_preds.npy')\n",
    "y_test = data[5]\n",
    "\n",
    "evaluation.create_confusion_matrix(preds, y_test)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "EWLLMs",
   "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": "3.10.12"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}