inception.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "74a8df04",
   "metadata": {},
   "outputs": [],
   "source": [
    "from helper_code import *\n",
    "import numpy as np, os, sys, joblib\n",
    "import ecg_plot\n",
    "import pandas as pd\n",
    "from glob import glob\n",
    "import os\n",
    "from torch.utils.data import Dataset, DataLoader\n",
    "from sklearn.model_selection import train_test_split\n",
    "from tsai.all import *\n",
    "import torch\n",
    "import optuna\n",
    "from optuna.integration import FastAIPruningCallback\n",
    "from sklearn.metrics import classification_report\n",
    "import transformation_funcs as tfs\n",
    "import seaborn as sns\n",
    "\n",
    "torch.cuda.set_device(0) \n",
    "def snomedConvert(label_df,snomed=True):\n",
    "    codes =  pd.read_csv(\"data/snomed_codes.csv\",sep=\",\")[[\"Dx\",\"SNOMEDCTCode\"]]\n",
    "    if snomed:\n",
    "        label_df.columns = [codes[codes[\"SNOMEDCTCode\"] == int(x)][\"Dx\"].item() for x in label_df.columns]\n",
    "        return label_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "b323e637",
   "metadata": {},
   "outputs": [],
   "source": [
    "X = np.load('./data/big_numpy_datasets/WFDB_CPSC2018.npy', mmap_mode='c')\n",
    "label_df = pd.read_csv(\"data/WFDB_CPSC2018.csv\").drop(columns=[\"headers\",\"leads\"])\n",
    "\n",
    "y = snomedConvert(label_df)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "323302fb",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1152x36 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "((#5503) [3730,369,1166,1104,1256,1925,1397,6851,5431,5658...],\n",
       " (#687) [3536,5567,3053,1591,268,2755,6308,2063,3139,3885...],\n",
       " (#687) [4921,3129,4389,2920,319,6726,55,1263,3608,2788...])"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "splits = get_splits(y.to_numpy(), valid_size=.1,test_size=0.1, stratify=True, random_state=23, shuffle=True)\n",
    "splits"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "1d707492",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Counts by label: {'right bundle branch block': 1857, 'sinus rhythm': 918, 'atrial fibrillation': 1221, 'ventricular ectopics': 700, 'st depression': 869, 'left bundle branch block': 236, 'st elevation': 220, '1st degree av block': 722, 'premature atrial contraction': 616}\n"
     ]
    }
   ],
   "source": [
    "df = pd.read_csv(\"data/WFDB_CPSC2018.csv\").drop(columns=[\"headers\",\"leads\"])\n",
    "y = snomedConvert(df)\n",
    "y_multi = []\n",
    "for i,row in y.iterrows():\n",
    "    sample_labels = []\n",
    "    for i,r in enumerate(row):\n",
    "        if r == True:\n",
    "            sample_labels.append(y.columns[i])\n",
    "        \n",
    "    y_multi.append(list(tuple(sample_labels)))\n",
    "label_counts = collections.Counter([a for r in y_multi for a in r])\n",
    "print('Counts by label:', dict(label_counts))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "601cb62a",
   "metadata": {},
   "outputs": [],
   "source": [
    "sf = 0.5\n",
    "tfms = [None,TSMultiLabelClassification()]\n",
    "batch_tfms = [\n",
    "#              TSStandardize(by_sample=True),\n",
    "              tfs.Normalize(),\n",
    "              tfs.Scale(scale_factor=0.5),\n",
    "              tfs.BandPass(int(sf*500),3,45),\n",
    "#               tfs.RandomShift(0.1),\n",
    "#               tfs.MulNoise(6),\n",
    "#               tfs.CutOutWhenTraining(),\n",
    "             ]\n",
    "dsets = TSDatasets(X.astype(float), y_multi, tfms=tfms, splits=splits)\n",
    "dls   = TSDataLoaders.from_dsets(dsets.train,dsets.valid, bs=[64, 128], batch_tfms=batch_tfms, num_workers=0) \n",
    "# reorganize y dataframe using dl's label/vocab mapping and count pos_weights\n",
    "y = y[dls.vocab.o2i.keys()]\n",
    "pos_weight = torch.tensor((len(y)-y.sum())/(y.sum()+1e-5))\n",
    "metrics =[ precision_multi, recall_multi, specificity_multi, F1_multi] \n",
    "model = InceptionTimePlus(dls.vars, dls.c, dls.len, depth=10,)\n",
    "learn = Learner(dls, model, metrics=metrics, loss_func=nn.BCEWithLogitsLoss(),cbs=ShowGraph())\n",
    "learn.fit_one_cycle(200, lr_max=0.004178080960048705)\n",
    "learn.recorder.plot_metrics()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.8.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}