{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.10.12","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"none","dataSources":[{"sourceId":59093,"databundleVersionId":7469972,"isSourceIdPinned":false,"sourceType":"competition"},{"sourceId":11484844,"sourceType":"datasetVersion","datasetId":7198309},{"sourceId":349989,"sourceType":"modelInstanceVersion","isSourceIdPinned":false,"modelInstanceId":292122,"modelId":312776},{"sourceId":351618,"sourceType":"modelInstanceVersion","isSourceIdPinned":true,"modelInstanceId":286627,"modelId":307445}],"dockerImageVersionId":30918,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"# This Python 3 environment comes with many helpful analytics libraries installed\n# It is defined by the kaggle/python Docker image: https://github.com/kaggle/docker-python\n# For example, here's several helpful packages to load\n\nimport numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\n\n# Input data files are available in the read-only \"../input/\" directory\n# For example, running this (by clicking run or pressing Shift+Enter) will list all files under the input directory\n\nimport os\n#for dirname, _, filenames in os.walk('/kaggle/input'):\n #for filename in filenames:\n #print(os.path.join(dirname, filename))\n\n# You can write up to 20GB to the current directory (/kaggle/working/) that gets preserved as output when you create a version using \"Save & Run All\" \n# You can also write temporary files to /kaggle/temp/, but they won't be saved outside of the current session","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:09.842419Z","iopub.execute_input":"2025-04-22T10:23:09.842852Z","iopub.status.idle":"2025-04-22T10:23:10.840004Z","shell.execute_reply.started":"2025-04-22T10:23:09.842817Z","shell.execute_reply":"2025-04-22T10:23:10.839069Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Importing libraries\n\nfrom pathlib import Path\nimport re\n\nimport matplotlib.pyplot as plt\n\nimport json","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:10.841202Z","iopub.execute_input":"2025-04-22T10:23:10.841643Z","iopub.status.idle":"2025-04-22T10:23:10.845832Z","shell.execute_reply.started":"2025-04-22T10:23:10.841617Z","shell.execute_reply":"2025-04-22T10:23:10.844886Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Setting the seed for np\n\nnp.random.seed(42)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:10.847232Z","iopub.execute_input":"2025-04-22T10:23:10.847538Z","iopub.status.idle":"2025-04-22T10:23:10.863340Z","shell.execute_reply.started":"2025-04-22T10:23:10.847511Z","shell.execute_reply":"2025-04-22T10:23:10.862613Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# Overview","metadata":{}},{"cell_type":"markdown","source":"This project is based on the Kaggle competetion to detect and classify seizures and other types of harmful brain activity.\nHere the main goal is to develop a model trained on electroencephalography (EEG) signals recorded from critically ill hospital patients.","metadata":{}},{"cell_type":"markdown","source":"**What is EEG?**\n\n**Electroencephalography**, commonly referred to as **EEG**, is a non-invasive method used to record electrical activity in the brain. This technique involves placing electrodes on the scalp, which detect tiny electrical charges that result from the activity of brain cells. The signals captured by these electrodes are amplified and recorded, typically resulting in a series of wavy lines that are analyzed by specialists","metadata":{}},{"cell_type":"markdown","source":"**Patterns for classification**\n\nThere are six patterns of interest for this competition: seizure (SZ), generalized periodic discharges (GPD), lateralized periodic discharges (LPD), lateralized rhythmic delta activity (LRDA), generalized rhythmic delta activity (GRDA), or “other","metadata":{}},{"cell_type":"markdown","source":"The EEG segments used in this competition have been annotated, or classified, by a group of experts. \nIn some cases experts completely agree about the correct label.On other cases the experts disagree. \n\n* We call segments where there are high levels of agreement **“idealized” patterns**.\n* Cases where ~1/2 of experts give a label as “other” and ~1/2 give one of the remaining five labels, we call **“proto patterns”**.\n* Cases where experts are approximately split between 2 of the 5 named patterns, we call **“edge cases”**.","metadata":{}},{"cell_type":"markdown","source":"# Exploring the Data","metadata":{}},{"cell_type":"code","source":"#exploring the files\ncurrent_path=Path(os.getcwd())\ninput_path=Path('/kaggle/input/hms-harmful-brain-activity-classification')\n\nos.listdir(input_path)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:15.575778Z","iopub.execute_input":"2025-04-22T10:23:15.576111Z","iopub.status.idle":"2025-04-22T10:23:15.583852Z","shell.execute_reply.started":"2025-04-22T10:23:15.576084Z","shell.execute_reply":"2025-04-22T10:23:15.582993Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Defing the spectrograms and eegs dir\ntrain_spectrogram_dir=input_path/'train_spectrograms'\ntrain_eegs_dir=input_path/'train_eegs'\n\ntest_spectrogram_dir=input_path/'test_spectrograms'\ntest_eegs_dir=input_path/'test_eegs'","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:15.721163Z","iopub.execute_input":"2025-04-22T10:23:15.721482Z","iopub.status.idle":"2025-04-22T10:23:15.725355Z","shell.execute_reply.started":"2025-04-22T10:23:15.721456Z","shell.execute_reply":"2025-04-22T10:23:15.724670Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Checking the train data\ntrain_data=pd.read_csv('/kaggle/input/hms-harmful-brain-activity-classification/train.csv')\ntrain_data","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:15.813059Z","iopub.execute_input":"2025-04-22T10:23:15.813354Z","iopub.status.idle":"2025-04-22T10:23:16.051573Z","shell.execute_reply.started":"2025-04-22T10:23:15.813332Z","shell.execute_reply":"2025-04-22T10:23:16.050537Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_data.columns","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:16.052610Z","iopub.execute_input":"2025-04-22T10:23:16.052847Z","iopub.status.idle":"2025-04-22T10:23:16.058095Z","shell.execute_reply.started":"2025-04-22T10:23:16.052826Z","shell.execute_reply":"2025-04-22T10:23:16.057396Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"**Columns of dataset:**\n\n**eeg_id** - A unique identifier for the entire EEG recording.\n\n**eeg_sub_id** - An ID for the specific 50 second long subsample this row's labels apply to.\n\n**eeg_label_offset_seconds** - The time between the beginning of the consolidated EEG and this subsample.\n\n**spectrogram_id** - A unique identifier for the entire EEG recording.\n\n**spectrogram_sub_id** - An ID for the specific 10 minute subsample this row's labels apply to.\n\n**spectogram_label_offset_seconds** - The time between the beginning of the consolidated spectrogram and this subsample.\n\n**label_id** - An ID for this set of labels.\n\n**patient_id** - An ID for the patient who donated the data.\n\n**expert_consensus** - The consensus annotator label. Provided for convenience only.\n\n**seizure/lpd/gpd/lrda/grda/other_vote** - The count of annotator votes for a given brain activity class. The full names of the activity classes are as follows: lpd: lateralized periodic discharges, gpd: generalized periodic discharges, lrd: lateralized rhythmic delta activity, and grda: generalized rhythmic delta activity \n","metadata":{}},{"cell_type":"code","source":"train_data.info()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:19.751366Z","iopub.execute_input":"2025-04-22T10:23:19.751730Z","iopub.status.idle":"2025-04-22T10:23:19.780805Z","shell.execute_reply.started":"2025-04-22T10:23:19.751701Z","shell.execute_reply":"2025-04-22T10:23:19.779956Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"print(f\"No of patients:{train_data['patient_id'].nunique()}\")","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:19.879799Z","iopub.execute_input":"2025-04-22T10:23:19.880093Z","iopub.status.idle":"2025-04-22T10:23:19.886848Z","shell.execute_reply.started":"2025-04-22T10:23:19.880071Z","shell.execute_reply":"2025-04-22T10:23:19.885916Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_data['expert_consensus'].value_counts().plot(kind='bar')","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:23.292196Z","iopub.execute_input":"2025-04-22T10:23:23.292507Z","iopub.status.idle":"2025-04-22T10:23:23.509336Z","shell.execute_reply.started":"2025-04-22T10:23:23.292482Z","shell.execute_reply":"2025-04-22T10:23:23.508612Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"vote_columns=['seizure_vote', 'lpd_vote','gpd_vote', 'lrda_vote', 'grda_vote', 'other_vote']\ntrain_data[vote_columns].value_counts()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:23.510287Z","iopub.execute_input":"2025-04-22T10:23:23.510624Z","iopub.status.idle":"2025-04-22T10:23:23.530903Z","shell.execute_reply.started":"2025-04-22T10:23:23.510589Z","shell.execute_reply":"2025-04-22T10:23:23.530046Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"We can see that there are various combinations of types of votes, and we will classify and add a column based on whether they are :\n* idealized- high levels of agreement\n* proto- Cases where ~1/2 of experts give a label as “other” and ~1/2 give one of the remaining five labels\n* edge-Cases where experts are approximately split between 2 of the 5 named patterns,\n","metadata":{}},{"cell_type":"code","source":"\n#Calculating total votes\ntrain_data['total_vote']=train_data[vote_columns].sum(axis=1)\n\n#Calculating the normalized vote for each\nfor col in vote_columns:\n train_data[col+'_n']=train_data[col]/train_data['total_vote']\n\ntrain_data\n","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:23.893849Z","iopub.execute_input":"2025-04-22T10:23:23.894180Z","iopub.status.idle":"2025-04-22T10:23:24.000150Z","shell.execute_reply.started":"2025-04-22T10:23:23.894151Z","shell.execute_reply":"2025-04-22T10:23:23.998967Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"vote_columns_n=[x+'_n' for x in vote_columns]\ntrain_data[vote_columns_n].value_counts()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:24.001671Z","iopub.execute_input":"2025-04-22T10:23:24.002018Z","iopub.status.idle":"2025-04-22T10:23:24.026145Z","shell.execute_reply.started":"2025-04-22T10:23:24.001987Z","shell.execute_reply":"2025-04-22T10:23:24.025457Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Defining a function to classify eeg data according to vote distribution\n\ndef classify_pattern(x):\n named_columns=['seizure_vote_n', 'lpd_vote_n', 'gpd_vote_n', 'lrda_vote_n', 'grda_vote_n']\n #Agreeing on one type\n if np.any(x[vote_columns_n]>0.95):\n pattern='idealized'\n #Agreeing on other and named\n elif np.any(x[named_columns]>0.45) and x['other_vote_n']>0.45:\n pattern='proto'\n else:\n pattern= 'edge'\n return pattern","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:25.630171Z","iopub.execute_input":"2025-04-22T10:23:25.630481Z","iopub.status.idle":"2025-04-22T10:23:25.635150Z","shell.execute_reply.started":"2025-04-22T10:23:25.630458Z","shell.execute_reply":"2025-04-22T10:23:25.634327Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_data['pattern']=train_data.apply(lambda x:classify_pattern(x),axis=1)\ntrain_data\n","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T12:15:06.272995Z","iopub.execute_input":"2025-04-22T12:15:06.273344Z","iopub.status.idle":"2025-04-22T12:16:13.928485Z","shell.execute_reply.started":"2025-04-22T12:15:06.273322Z","shell.execute_reply":"2025-04-22T12:16:13.927650Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Visualizing distribution of pattern\ntrain_data['pattern'].value_counts().plot(kind='bar')","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T12:16:37.022873Z","iopub.execute_input":"2025-04-22T12:16:37.023205Z","iopub.status.idle":"2025-04-22T12:16:37.233976Z","shell.execute_reply.started":"2025-04-22T12:16:37.023181Z","shell.execute_reply":"2025-04-22T12:16:37.232932Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Checking the test data\ntest_data=pd.read_csv('/kaggle/input/hms-harmful-brain-activity-classification/test.csv')\ntest_data","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:26.220964Z","iopub.execute_input":"2025-04-22T10:23:26.221234Z","iopub.status.idle":"2025-04-22T10:23:26.234361Z","shell.execute_reply.started":"2025-04-22T10:23:26.221213Z","shell.execute_reply":"2025-04-22T10:23:26.233392Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"## EEG Data","metadata":{}},{"cell_type":"code","source":"print(f'Nos. of train eeg files:{len(os.listdir(train_eegs_dir))}')\neegs_list=os.listdir(train_eegs_dir)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:26.787395Z","iopub.execute_input":"2025-04-22T10:23:26.787764Z","iopub.status.idle":"2025-04-22T10:23:26.956108Z","shell.execute_reply.started":"2025-04-22T10:23:26.787735Z","shell.execute_reply":"2025-04-22T10:23:26.955304Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"sample_eeg=train_eegs_dir/eegs_list[2]\nsample_eeg","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:26.957184Z","iopub.execute_input":"2025-04-22T10:23:26.957476Z","iopub.status.idle":"2025-04-22T10:23:26.962627Z","shell.execute_reply.started":"2025-04-22T10:23:26.957453Z","shell.execute_reply":"2025-04-22T10:23:26.961877Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"eeg_id=re.search(r'[0-9]+',str(sample_eeg)).group()\ntrain_data[train_data['eeg_id']==int(eeg_id)]","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:26.977409Z","iopub.execute_input":"2025-04-22T10:23:26.977662Z","iopub.status.idle":"2025-04-22T10:23:26.997129Z","shell.execute_reply.started":"2025-04-22T10:23:26.977640Z","shell.execute_reply":"2025-04-22T10:23:26.996433Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Reading a sample eeg file\n\npd.read_parquet(sample_eeg)\n","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:27.239189Z","iopub.execute_input":"2025-04-22T10:23:27.239535Z","iopub.status.idle":"2025-04-22T10:23:27.380079Z","shell.execute_reply.started":"2025-04-22T10:23:27.239507Z","shell.execute_reply":"2025-04-22T10:23:27.379187Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"**EEG data from one or more overlapping samples.** \nThe column names are the names of the individual electrode locations for EEG leads, with one exception. The EKG column is for an electrocardiogram lead that records data from the heart. All of the EEG data (for both train and test) was collected at a frequency of 200 samples per second.","metadata":{}},{"cell_type":"markdown","source":"There are 20 columns of 10000 rows.\n\nNo of seconds for which data was collected=10000/200 sec=50 sec","metadata":{}},{"cell_type":"code","source":"#Defining a function to plot EEG\n\ndef plot_eeg(eeg_path):\n sample_eeg=pd.read_parquet(eeg_path)\n\n for i,col in enumerate(sample_eeg.columns):\n fig=plt.figure(figsize=(5,30))\n ax=plt.subplot(20,1,i+1)\n ax.plot(sample_eeg[col])\n ax.set_xlabel(col)\n plt.grid()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:30.129817Z","iopub.execute_input":"2025-04-22T10:23:30.130119Z","iopub.status.idle":"2025-04-22T10:23:30.135317Z","shell.execute_reply.started":"2025-04-22T10:23:30.130096Z","shell.execute_reply":"2025-04-22T10:23:30.134227Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"plot_eeg(eeg_path=sample_eeg)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:30.262895Z","iopub.execute_input":"2025-04-22T10:23:30.263213Z","iopub.status.idle":"2025-04-22T10:23:33.255531Z","shell.execute_reply.started":"2025-04-22T10:23:30.263184Z","shell.execute_reply":"2025-04-22T10:23:33.254758Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"**Insights from the plot:**\n* Data has a lot of peaks and troughs\n* Most of variation is within -100 to +100 except EKG","metadata":{}},{"cell_type":"code","source":"#Checking the test eeg\nprint(f'Nos. of test eeg files:{len(os.listdir(test_eegs_dir))}')\ntest_eegs_list=os.listdir(test_eegs_dir)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:38.246250Z","iopub.execute_input":"2025-04-22T10:23:38.246606Z","iopub.status.idle":"2025-04-22T10:23:38.256886Z","shell.execute_reply.started":"2025-04-22T10:23:38.246546Z","shell.execute_reply":"2025-04-22T10:23:38.255969Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"## Spectrogram data\n\n","metadata":{}},{"cell_type":"code","source":"print(f'Nos. of spectrogram files:{len(os.listdir(train_spectrogram_dir))}')\nspg_list=os.listdir(train_spectrogram_dir)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:38.615152Z","iopub.execute_input":"2025-04-22T10:23:38.615496Z","iopub.status.idle":"2025-04-22T10:23:38.715911Z","shell.execute_reply.started":"2025-04-22T10:23:38.615467Z","shell.execute_reply":"2025-04-22T10:23:38.715125Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Reading a sample spectrogram file\nsample_spectrogram=train_spectrogram_dir/spg_list[2]\npd.read_parquet(sample_spectrogram)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:38.795977Z","iopub.execute_input":"2025-04-22T10:23:38.796270Z","iopub.status.idle":"2025-04-22T10:23:38.854515Z","shell.execute_reply.started":"2025-04-22T10:23:38.796247Z","shell.execute_reply":"2025-04-22T10:23:38.853758Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"Spectrograms assembled EEG data. The **column names indicate the frequency in hertz** and the recording regions of the EEG electrodes. The latter are abbreviated as LL = left lateral; RL = right lateral; LP = left parasagittal; RP = right parasagittal.","metadata":{}},{"cell_type":"code","source":"def plot_spectrogram(spectrogram_path):\n sample_spect = pd.read_parquet(spectrogram_path)\n \n split_spect = {\n \"LL\": sample_spect.filter(regex='^LL', axis=1),\n \"RL\": sample_spect.filter(regex='^RL', axis=1),\n \"RP\": sample_spect.filter(regex='^RP', axis=1),\n \"LP\": sample_spect.filter(regex='^LP', axis=1),\n }\n \n fig, axes = plt.subplots(nrows=2, ncols=2, figsize=(15, 12))\n axes = axes.flatten()\n label_interval = 5\n for i, split_name in enumerate(split_spect.keys()):\n ax = axes[i]\n img = ax.imshow(np.log(split_spect[split_name]).T, cmap='viridis', aspect='auto', origin='lower')\n cbar = fig.colorbar(img, ax=ax)\n cbar.set_label('Log(Value)')\n ax.set_title(split_name)\n ax.set_ylabel(\"Frequency (Hz)\")\n ax.set_xlabel(\"Time\")\n\n ax.set_yticks(np.arange(len(split_spect[split_name].columns)))\n ax.set_yticklabels([column_name[3:] for column_name in split_spect[split_name].columns])\n frequencies = [column_name[3:] for column_name in split_spect[split_name].columns]\n ax.set_yticks(np.arange(0, len(split_spect[split_name].columns), label_interval))\n ax.set_yticklabels(frequencies[::label_interval])\n plt.tight_layout()\n plt.show()\n","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:39.165255Z","iopub.execute_input":"2025-04-22T10:23:39.165622Z","iopub.status.idle":"2025-04-22T10:23:39.173389Z","shell.execute_reply.started":"2025-04-22T10:23:39.165554Z","shell.execute_reply":"2025-04-22T10:23:39.172448Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"plot_spectrogram(sample_spectrogram)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:23:39.263760Z","iopub.execute_input":"2025-04-22T10:23:39.264017Z","iopub.status.idle":"2025-04-22T10:23:41.441605Z","shell.execute_reply.started":"2025-04-22T10:23:39.263995Z","shell.execute_reply":"2025-04-22T10:23:41.440672Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# Understanding the data ","metadata":{}},{"cell_type":"code","source":"#Getting the first eeg_id and spectogram_id\nsample_eeg_id=train_data['eeg_id'][0]\nsample_spectogram_id=train_data['spectrogram_id'][0]\n\nsample_data=train_data[train_data['eeg_id']==sample_eeg_id]\nsample_data","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:20.941960Z","iopub.execute_input":"2025-04-22T10:24:20.942299Z","iopub.status.idle":"2025-04-22T10:24:20.967478Z","shell.execute_reply.started":"2025-04-22T10:24:20.942272Z","shell.execute_reply":"2025-04-22T10:24:20.966510Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Reading eeg data\nsample_eeg=train_eegs_dir/(str(sample_eeg_id)+'.parquet')\npd.read_parquet(sample_eeg)\n","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:20.970692Z","iopub.execute_input":"2025-04-22T10:24:20.970965Z","iopub.status.idle":"2025-04-22T10:24:21.021805Z","shell.execute_reply.started":"2025-04-22T10:24:20.970943Z","shell.execute_reply":"2025-04-22T10:24:21.020883Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Visualizing eeg data\nplot_eeg(sample_eeg)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:21.026320Z","iopub.execute_input":"2025-04-22T10:24:21.026581Z","iopub.status.idle":"2025-04-22T10:24:24.159330Z","shell.execute_reply.started":"2025-04-22T10:24:21.026544Z","shell.execute_reply":"2025-04-22T10:24:24.158335Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"* From the above EEG data, we can observe that the training data consists of several sub ids which are each 50 sec long.\n* On the other hand, EEG is a consolidated data where each 2000 samples represent a second.\n* It is important to note that **test data only contains 50 secs of non-overlapping EEG data**\n\n","metadata":{}},{"cell_type":"code","source":"#Reading spectogram data\nsample_spg=train_spectrogram_dir/(str(sample_spectogram_id)+'.parquet')\npd.read_parquet(sample_spg)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:24.160436Z","iopub.execute_input":"2025-04-22T10:24:24.160707Z","iopub.status.idle":"2025-04-22T10:24:24.236343Z","shell.execute_reply.started":"2025-04-22T10:24:24.160685Z","shell.execute_reply":"2025-04-22T10:24:24.235660Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"plot_spectrogram(sample_spg)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:24.253724Z","iopub.execute_input":"2025-04-22T10:24:24.253964Z","iopub.status.idle":"2025-04-22T10:24:26.318183Z","shell.execute_reply.started":"2025-04-22T10:24:24.253944Z","shell.execute_reply":"2025-04-22T10:24:26.316946Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"* From the above Spectrogram data, we can observe that the training data consists of several sub ids which are each 10 mins long.\n* On the other hand, Spectrogram is a consolidated data where each row represents 2 secs \n* It is important to note that **test data only contains 10 mins of non-overlapping Spectrogram data**","metadata":{}},{"cell_type":"code","source":"#Checking the occurence of spectrograms against eeg\ntrain_data.groupby(['eeg_id'])['spectrogram_id'].nunique().unique()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:26.319870Z","iopub.execute_input":"2025-04-22T10:24:26.320274Z","iopub.status.idle":"2025-04-22T10:24:26.334449Z","shell.execute_reply.started":"2025-04-22T10:24:26.320242Z","shell.execute_reply":"2025-04-22T10:24:26.333619Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"Hence,each eeg is associated with a unique spectrogram data","metadata":{}},{"cell_type":"code","source":"train_data.groupby(['spectrogram_id'])['eeg_id'].nunique().unique()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:26.336319Z","iopub.execute_input":"2025-04-22T10:24:26.336549Z","iopub.status.idle":"2025-04-22T10:24:26.352958Z","shell.execute_reply.started":"2025-04-22T10:24:26.336530Z","shell.execute_reply":"2025-04-22T10:24:26.352025Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"There may be multiple EEGs for a single Spectrogram","metadata":{}},{"cell_type":"code","source":"#Each spectrogram ID may have multiple expert_consensus\ntrain_data.groupby(['spectrogram_id'])['expert_consensus'].nunique().value_counts()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:26.354556Z","iopub.execute_input":"2025-04-22T10:24:26.354896Z","iopub.status.idle":"2025-04-22T10:24:26.373791Z","shell.execute_reply.started":"2025-04-22T10:24:26.354874Z","shell.execute_reply":"2025-04-22T10:24:26.372986Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Each eeg ID may have multiple expert_consensus\ntrain_data.groupby(['eeg_id'])['expert_consensus'].nunique().value_counts()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:26.374752Z","iopub.execute_input":"2025-04-22T10:24:26.375005Z","iopub.status.idle":"2025-04-22T10:24:26.392855Z","shell.execute_reply.started":"2025-04-22T10:24:26.374984Z","shell.execute_reply":"2025-04-22T10:24:26.392120Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"From the above, we can see that both for spectrogram and EEG, there is usually a common expert consensus though a few may vary","metadata":{}},{"cell_type":"code","source":"train_data","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:26.393591Z","iopub.execute_input":"2025-04-22T10:24:26.393837Z","iopub.status.idle":"2025-04-22T10:24:26.442495Z","shell.execute_reply.started":"2025-04-22T10:24:26.393815Z","shell.execute_reply":"2025-04-22T10:24:26.441524Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"In the train data, each **eeg_sub_id** is associated with **50 sec of EEG data** and each **spectrogram_sub_id** is associated with **10 min of Spectrogram data**","metadata":{}},{"cell_type":"markdown","source":"# Extracting data corresponding to EEG Data","metadata":{}},{"cell_type":"code","source":"sample_eeg_id=train_data['eeg_id'][0]\nsample_eeg=train_eegs_dir/(str(sample_eeg_id)+'.parquet')\nsample_df=pd.read_parquet(sample_eeg)\nsample_df","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:26.443484Z","iopub.execute_input":"2025-04-22T10:24:26.443828Z","iopub.status.idle":"2025-04-22T10:24:26.471454Z","shell.execute_reply.started":"2025-04-22T10:24:26.443796Z","shell.execute_reply":"2025-04-22T10:24:26.470619Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"sample_train_data=train_data[train_data['eeg_id']==sample_eeg_id]\nsample_train_data","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:26.472408Z","iopub.execute_input":"2025-04-22T10:24:26.472778Z","iopub.status.idle":"2025-04-22T10:24:26.496175Z","shell.execute_reply.started":"2025-04-22T10:24:26.472744Z","shell.execute_reply":"2025-04-22T10:24:26.495286Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"All of the EEG data (for both train and test) was collected at a frequency of 200 samples per second.\nhence for 18000 rows, time provided for this **EEG data =18000/200 =90 sec**","metadata":{}},{"cell_type":"markdown","source":"## Creating and Loading Datset for EEG","metadata":{}},{"cell_type":"code","source":"import torch\nfrom torch.utils.data import Dataset,DataLoader,Subset","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:26.498167Z","iopub.execute_input":"2025-04-22T10:24:26.498425Z","iopub.status.idle":"2025-04-22T10:24:30.137758Z","shell.execute_reply.started":"2025-04-22T10:24:26.498402Z","shell.execute_reply":"2025-04-22T10:24:30.136991Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"device='cuda' if torch.cuda.is_available() else 'cpu'","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:30.139061Z","iopub.execute_input":"2025-04-22T10:24:30.139509Z","iopub.status.idle":"2025-04-22T10:24:30.197334Z","shell.execute_reply.started":"2025-04-22T10:24:30.139486Z","shell.execute_reply":"2025-04-22T10:24:30.196422Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"class EEGDataset(Dataset):\n \n def __init__(self,data,eeg_dir_path):\n self.data=data\n self.eeg_dir_path=eeg_dir_path\n self.targets=['seizure_vote',\t'lpd_vote',\t'gpd_vote',\t'lrda_vote',\t'grda_vote',\t'other_vote']\n \n def __len__(self):\n return len(self.data)\n\n def __getitem__(self,index):\n sample_eeg_id=self.data.iloc[index]['eeg_id'] #Getting the id\n sample_eeg_offset=self.data.iloc[index]['eeg_label_offset_seconds'] #Getting the offset\n\n #Getting the targets\n sample_eeg_target=self.data.iloc[index][self.targets]\n\n sample_eeg=self.eeg_dir_path/(str(sample_eeg_id)+'.parquet') #Getting the EEG file \n sample_eeg_df=pd.read_parquet(sample_eeg) #Reading the parquet\n\n #Imputing values\n \n #sample_eeg_df=sample_eeg_df.bfill(axis=0) #Filling null values across rows\n #sample_eeg_df=sample_eeg_df.ffill(axis=0)\n\n #Getting the EEG data\n X=self.get_features(eeg_df=sample_eeg_df,eeg_offset=sample_eeg_offset).T #Transposing to get in the shape of (Channel,Timesteps)\n \n y=np.array(sample_eeg_target.values.astype('float32')) \n y=torch.tensor(y)\n y=y.softmax(dim=-1) #Since we want probability distribution\n \n\n return torch.tensor(X),y\n\n\n def get_features(self,eeg_df,eeg_offset):\n\n #Each row in eeg_df contains 1 of 200 samples per second\n \n #Checking for existence of atleast 50 secs or 10000 rows of data\n #if int(eeg_offset)*200+50*200<=len(eeg_df):\n truncated_df=eeg_df.iloc[int(eeg_offset)*200:int(eeg_offset)*200+50*200] #Collecting 50 secs of data\n eeg_array=truncated_df.to_numpy() #Shape:(timesteps,features)\n \n \n #else:/\n #truncated_df=eeg_df.iloc[int(eeg_offset)*200:] #Collecting till last\n #eeg_array=truncated_df.to_numpy()\n\n #Padding to 10000 time steps\n #rows_to_add = 10000 - eeg_array.shape[0]\n #padding = ((0, rows_to_add), (0, 0)) # Pad rows at the end, no padding for columns\n #eeg_array = np.pad(eeg_array, padding, mode='mean')\n return eeg_array\n\n \n\n def target_dict(self):\n return {i:t for (i,t) in enumerate(self.targets)}\n \n \n \n \n \n\n \n ","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:30.198258Z","iopub.execute_input":"2025-04-22T10:24:30.198499Z","iopub.status.idle":"2025-04-22T10:24:30.209002Z","shell.execute_reply.started":"2025-04-22T10:24:30.198477Z","shell.execute_reply":"2025-04-22T10:24:30.208311Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"### Checking for null indexes","metadata":{}},{"cell_type":"code","source":"#Defining a function to get null indexes\ndef get_null_indexes(data,eeg_dir_path):\n null_index_list=[]\n for index in range(len(data)):\n eeg_id=data.iloc[index]['eeg_id'] #Getting the id\n eeg_offset=data.iloc[index]['eeg_label_offset_seconds'] #Getting the offset\n\n eeg_file=eeg_dir_path/(str(eeg_id)+'.parquet') #Getting the EEG file \n eeg_df=pd.read_parquet(eeg_file) #Reading the parquet\n\n\n truncated_df=eeg_df.iloc[int(eeg_offset)*200:int(eeg_offset)*200+50*200] #Collecting 50 secs of data\n eeg_array=truncated_df.to_numpy() #Shape:(timesteps,features).T #Transposing to get in the shape of (Channel,Timesteps)\n\n if np.isnan(eeg_array).any():\n null_index_list.append(index)\n\n return null_index_list\n ","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:30.209970Z","iopub.execute_input":"2025-04-22T10:24:30.210275Z","iopub.status.idle":"2025-04-22T10:24:30.227379Z","shell.execute_reply.started":"2025-04-22T10:24:30.210246Z","shell.execute_reply":"2025-04-22T10:24:30.226647Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Getting null indices\n\n#null_index_list=get_null_indexes(data=train_data,eeg_dir_path=train_eegs_dir)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:30.229226Z","iopub.execute_input":"2025-04-22T10:24:30.229428Z","iopub.status.idle":"2025-04-22T10:24:30.244241Z","shell.execute_reply.started":"2025-04-22T10:24:30.229410Z","shell.execute_reply":"2025-04-22T10:24:30.243399Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Null list location\nnull_index_dir=Path('/kaggle/working')\nnull_index_path=null_index_dir/'null_index.json'","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:30.245101Z","iopub.execute_input":"2025-04-22T10:24:30.245296Z","iopub.status.idle":"2025-04-22T10:24:30.258466Z","shell.execute_reply.started":"2025-04-22T10:24:30.245278Z","shell.execute_reply":"2025-04-22T10:24:30.257649Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Null index download path\nnull_index_download_path=Path('/kaggle/input')/'null-index-json/null_index.json'","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:32.109341Z","iopub.execute_input":"2025-04-22T10:24:32.109816Z","iopub.status.idle":"2025-04-22T10:24:32.113686Z","shell.execute_reply.started":"2025-04-22T10:24:32.109776Z","shell.execute_reply":"2025-04-22T10:24:32.112787Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Storing files in an object\n'''\nwith open(null_index_path,'w') as f:\n json.dump(null_index_list,f)\n'''","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:32.295655Z","iopub.execute_input":"2025-04-22T10:24:32.295927Z","iopub.status.idle":"2025-04-22T10:24:32.301328Z","shell.execute_reply.started":"2025-04-22T10:24:32.295905Z","shell.execute_reply":"2025-04-22T10:24:32.300496Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Loading the list\nwith open(null_index_download_path,'r') as f:\n null_list=json.load(f)\n ","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:32.600234Z","iopub.execute_input":"2025-04-22T10:24:32.600554Z","iopub.status.idle":"2025-04-22T10:24:32.609969Z","shell.execute_reply.started":"2025-04-22T10:24:32.600529Z","shell.execute_reply":"2025-04-22T10:24:32.609289Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"print(f'No of null index:{len(null_list)}')","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:32.681765Z","iopub.execute_input":"2025-04-22T10:24:32.682044Z","iopub.status.idle":"2025-04-22T10:24:32.686921Z","shell.execute_reply.started":"2025-04-22T10:24:32.682022Z","shell.execute_reply":"2025-04-22T10:24:32.685934Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"#### Splitting data into train & val","metadata":{}},{"cell_type":"code","source":"#Getting the non-null indices\nnon_null_indices=list(set(range(len(train_data)))-set(null_list))\nprint(f'All indices:{len(train_data)}')\nprint(f'Non null indices:{len(non_null_indices)}')","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:34.601311Z","iopub.execute_input":"2025-04-22T10:24:34.601703Z","iopub.status.idle":"2025-04-22T10:24:34.615992Z","shell.execute_reply.started":"2025-04-22T10:24:34.601668Z","shell.execute_reply":"2025-04-22T10:24:34.615158Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Splitting the indices\ntrain_split=0.8\n\ntrain_indices=np.random.choice(non_null_indices,int((len(non_null_indices)*train_split)),replace=False).tolist()\nval_indices=list(set(non_null_indices)-set(train_indices))\n\n","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:34.749708Z","iopub.execute_input":"2025-04-22T10:24:34.750047Z","iopub.status.idle":"2025-04-22T10:24:34.778177Z","shell.execute_reply.started":"2025-04-22T10:24:34.750020Z","shell.execute_reply":"2025-04-22T10:24:34.777182Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Splitting the data based on indices\n\ntrain_set=train_data.iloc[train_indices]\nval_set=train_data.iloc[val_indices]","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:35.051222Z","iopub.execute_input":"2025-04-22T10:24:35.051550Z","iopub.status.idle":"2025-04-22T10:24:35.080401Z","shell.execute_reply.started":"2025-04-22T10:24:35.051526Z","shell.execute_reply":"2025-04-22T10:24:35.079660Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Initialializing the dataset\ntrain_eeg_ds=EEGDataset(data=train_set,eeg_dir_path=train_eegs_dir)\nval_eeg_ds=EEGDataset(data=val_set,eeg_dir_path=train_eegs_dir)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:35.102636Z","iopub.execute_input":"2025-04-22T10:24:35.102905Z","iopub.status.idle":"2025-04-22T10:24:35.106873Z","shell.execute_reply.started":"2025-04-22T10:24:35.102884Z","shell.execute_reply":"2025-04-22T10:24:35.106096Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Checking length of dataset\nprint(f'Length of train dataset:{len(train_eeg_ds)}')\nprint(f'Length of val dataset:{len(val_eeg_ds)}')","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:35.285974Z","iopub.execute_input":"2025-04-22T10:24:35.286270Z","iopub.status.idle":"2025-04-22T10:24:35.290902Z","shell.execute_reply.started":"2025-04-22T10:24:35.286248Z","shell.execute_reply":"2025-04-22T10:24:35.289919Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Checking shape of dataset\nprint(f'Shape of X in dataset:{train_eeg_ds[0][0].shape}')\nprint(f'Shape of y in dataset:{train_eeg_ds[0][1].shape}')","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:24:38.308707Z","iopub.execute_input":"2025-04-22T10:24:38.309059Z","iopub.status.idle":"2025-04-22T10:24:38.376502Z","shell.execute_reply.started":"2025-04-22T10:24:38.309032Z","shell.execute_reply":"2025-04-22T10:24:38.375555Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"Hence we get the dataset in which **feature** in the shape of **(channels,timesteps)** and **target** in the shape of (labels)","metadata":{}},{"cell_type":"code","source":"#Creating a subdataset for research purpose\n\ntrain_sub_size=30000\nval_sub_size=5000\n\ntrain_sub_indices=np.random.choice(range(len(train_eeg_ds)),train_sub_size,replace=False)\nval_sub_indices=np.random.choice(range(len(val_eeg_ds)),val_sub_size,replace=False)\n\ntrain_sub_ds=Subset(train_eeg_ds,indices=train_sub_indices)\nval_sub_ds=Subset(val_eeg_ds,indices=val_sub_indices)\nprint(f'Length of sub dataset:{len(train_sub_ds)}')","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:25:59.206470Z","iopub.execute_input":"2025-04-22T10:25:59.206872Z","iopub.status.idle":"2025-04-22T10:25:59.224093Z","shell.execute_reply.started":"2025-04-22T10:25:59.206839Z","shell.execute_reply":"2025-04-22T10:25:59.223287Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"batch_size=32\n#Creating a sub dataloader\ntrain_sub_dl=DataLoader(train_sub_ds,batch_size=batch_size,shuffle=True)\nval_sub_dl=DataLoader(val_sub_ds,batch_size=batch_size,shuffle=False)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:26:01.788396Z","iopub.execute_input":"2025-04-22T10:26:01.788764Z","iopub.status.idle":"2025-04-22T10:26:01.792940Z","shell.execute_reply.started":"2025-04-22T10:26:01.788735Z","shell.execute_reply":"2025-04-22T10:26:01.792059Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"print(f'Shape of X in dataloader:{next(iter(train_sub_dl))[0].shape}')\nprint(f'Shape of y in dataloader:{next(iter(train_sub_dl))[1].shape}')\n","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:26:05.537417Z","iopub.execute_input":"2025-04-22T10:26:05.537817Z","iopub.status.idle":"2025-04-22T10:26:08.219076Z","shell.execute_reply.started":"2025-04-22T10:26:05.537783Z","shell.execute_reply":"2025-04-22T10:26:08.217893Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"sample_dl_X=next(iter(train_sub_dl))[0].to(device)\nprint(f'Shape of sample_dl={sample_dl_X.shape}')\n","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:26:08.220267Z","iopub.execute_input":"2025-04-22T10:26:08.220596Z","iopub.status.idle":"2025-04-22T10:26:09.372977Z","shell.execute_reply.started":"2025-04-22T10:26:08.220547Z","shell.execute_reply":"2025-04-22T10:26:09.372114Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Getting full validation dataloader\nval_eeg_dl=DataLoader(val_eeg_ds,batch_size=64)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:26:09.374468Z","iopub.execute_input":"2025-04-22T10:26:09.375267Z","iopub.status.idle":"2025-04-22T10:26:09.379871Z","shell.execute_reply.started":"2025-04-22T10:26:09.375220Z","shell.execute_reply":"2025-04-22T10:26:09.378782Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#Getting the dict\n\ntarget_dict=val_eeg_ds.target_dict()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T12:13:19.525907Z","iopub.execute_input":"2025-04-22T12:13:19.526269Z","iopub.status.idle":"2025-04-22T12:13:19.530159Z","shell.execute_reply.started":"2025-04-22T12:13:19.526239Z","shell.execute_reply":"2025-04-22T12:13:19.529345Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# Defining the model","metadata":{}},{"cell_type":"markdown","source":"## Defining Utility functions","metadata":{}},{"cell_type":"code","source":"import torch\nfrom torch import nn\nfrom torch.nn import functional as F\n\nimport torchinfo","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:25:15.095053Z","iopub.execute_input":"2025-04-22T10:25:15.095430Z","iopub.status.idle":"2025-04-22T10:25:15.109410Z","shell.execute_reply.started":"2025-04-22T10:25:15.095397Z","shell.execute_reply":"2025-04-22T10:25:15.108426Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import gc\n\ngc.collect()\n\ntorch.cuda.empty_cache()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-04-22T10:25:15.333109Z","iopub.execute_input":"2025-04-22T10:25:15.333445Z","iopub.status.idle":"2025-04-22T10:25:15.461236Z","shell.execute_reply.started":"2025-04-22T10:25:15.333421Z","shell.execute_reply":"2025-04-22T10:25:15.460215Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"def train_model(train_dl,val_dl,model,loss_fn,optimizer,epochs,model_path=None,scheduler=None,best_loss=None):\n epoch_train_loss=[]\n epoch_val_loss=[]\n if best_loss:\n best_loss=best_loss\n else:\n best_loss=np.inf\n \n for i in range(epochs):\n #Training the model\n model.train()\n train_loss=0\n for n,(X,y) in enumerate(train_dl):\n X,y=X.to(device),y.to(device)\n y_pred=model(X) #y_pred is softmax\n \n loss=loss_fn(torch.log(y_pred),y) \n \n train_loss+=loss*y.shape[0]\n #Zero grad the optimizer\n optimizer.zero_grad()\n #Backpropagate\n loss.backward()\n #Updating parameters\n optimizer.step()\n #Tuning the scheduler\n if scheduler:\n scheduler.step()\n if n%10==0:\n print(f'Batch:{i}_{n} | Batch loss:{loss.item():.2f}')\n\n \n\n train_loss=(train_loss/len(train_dl.dataset)).item()\n print(f'Epoch:{i} | Train loss:{train_loss:.2f}')\n epoch_train_loss.append(train_loss)\n\n #Evaluating model\n model.eval()\n val_loss=0\n with torch.no_grad():\n for X,y in val_dl:\n X,y=X.to(device),y.to(device)\n y_pred=model(X) #y_pred is softmax\n loss=loss_fn(torch.log(y_pred),y)\n val_loss+=loss*y.shape[0]\n val_loss=(val_loss/len(val_dl.dataset)).item()\n \n print(f'Epoch:{i} | Val loss:{val_loss:.2f}')\n epoch_val_loss.append(val_loss)\n #Saving model\n if model_path:\n model_path.parent.mkdir(parents=True,exist_ok=True)\n if val_loss