harmful-brain-activity.ipynb

{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.10.13","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,"sourceType":"competition"}],"dockerImageVersionId":30646,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"source":"<a href=\"https://www.kaggle.com/code/yujansaya/harmful-brain-activity?scriptVersionId=161975137\" target=\"_blank\"><img align=\"left\" alt=\"Kaggle\" title=\"Open in Kaggle\" src=\"https://kaggle.com/static/images/open-in-kaggle.svg\"></a>","metadata":{},"cell_type":"markdown"},{"cell_type":"code","source":"import numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\nimport random\nimport matplotlib.pyplot as plt, gc\nimport joblib  # to pipeline files reading and transforming\nfrom tqdm.notebook import tqdm  # progress bar\nimport keras_cv\nimport os\n# for dirname, _, filenames in os.walk('/kaggle/input'):\n#     for filename in filenames:\n#         #print()\n#         os.path.join(dirname, filename)","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2024-02-06T21:33:23.018556Z","iopub.execute_input":"2024-02-06T21:33:23.018931Z","iopub.status.idle":"2024-02-06T21:33:43.388478Z","shell.execute_reply.started":"2024-02-06T21:33:23.018899Z","shell.execute_reply":"2024-02-06T21:33:43.387558Z"},"trusted":true},"execution_count":1,"outputs":[{"name":"stderr","text":"2024-02-06 21:33:29.253603: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:9261] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered\n2024-02-06 21:33:29.253737: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:607] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered\n2024-02-06 21:33:29.404813: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1515] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered\n","output_type":"stream"},{"name":"stdout","text":"Using TensorFlow backend\n","output_type":"stream"}]},{"cell_type":"code","source":"pip install tensorflow[and-cuda]","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import tensorflow as tf","metadata":{"execution":{"iopub.status.busy":"2024-02-06T21:33:43.390458Z","iopub.execute_input":"2024-02-06T21:33:43.391791Z","iopub.status.idle":"2024-02-06T21:33:43.395935Z","shell.execute_reply.started":"2024-02-06T21:33:43.391761Z","shell.execute_reply":"2024-02-06T21:33:43.394862Z"},"trusted":true},"execution_count":2,"outputs":[]},{"cell_type":"code","source":"BASE_PATH = \"/kaggle/input/hms-harmful-brain-activity-classification\"\n\nSPEC_DIR = \"/kaggle/working\"\nos.makedirs(f'{SPEC_DIR}/train', exist_ok=True)\nos.makedirs(f'{SPEC_DIR}/test', exist_ok=True)","metadata":{"execution":{"iopub.status.busy":"2024-02-06T21:33:45.932427Z","iopub.execute_input":"2024-02-06T21:33:45.932796Z","iopub.status.idle":"2024-02-06T21:33:45.940523Z","shell.execute_reply.started":"2024-02-06T21:33:45.932766Z","shell.execute_reply":"2024-02-06T21:33:45.939174Z"},"trusted":true},"execution_count":3,"outputs":[]},{"cell_type":"code","source":"num_classes = 6\nimage_size = [200, 600]\nclass_names = ['Seizure', 'LPD', 'GPD', 'LRDA','GRDA', 'Other']\nlabel2name = dict(enumerate(class_names))\nname2label = {v:k for k, v in label2name.items()}\nbatch_size = 64\nLOSS = tf.keras.losses.KLDivergence()","metadata":{"execution":{"iopub.status.busy":"2024-02-06T21:33:47.529103Z","iopub.execute_input":"2024-02-06T21:33:47.530086Z","iopub.status.idle":"2024-02-06T21:33:47.536004Z","shell.execute_reply.started":"2024-02-06T21:33:47.530048Z","shell.execute_reply":"2024-02-06T21:33:47.534911Z"},"trusted":true},"execution_count":4,"outputs":[]},{"cell_type":"code","source":"# Train + Valid\ndf = pd.read_csv(f'{BASE_PATH}/train.csv')\ndf['eeg_path'] = f'{BASE_PATH}/train_eegs/'+df['eeg_id'].astype(str)+'.parquet'\ndf['spec_path'] = f'{BASE_PATH}/train_spectrograms/'+df['spectrogram_id'].astype(str)+'.parquet'\ndf['spec2_path'] = f'{SPEC_DIR}/train/'+df['spectrogram_id'].astype(str)+'.npy'\ndf['eeg_to_spec'] = f'{SPEC_DIR}/train/'+df['eeg_id'].astype(str)+'.npy'\ndf['class_name'] = df.expert_consensus.copy()\ndf['class_label'] = df.expert_consensus.map(name2label)\ndisplay(df.head(2))\n\n# Test\ntest_df = pd.read_csv(f'{BASE_PATH}/test.csv')\ntest_df['eeg_path'] = f'{BASE_PATH}/test_eegs/'+test_df['eeg_id'].astype(str)+'.parquet'\ntest_df['spec_path'] = f'{BASE_PATH}/test_spectrograms/'+test_df['spectrogram_id'].astype(str)+'.parquet'\ntest_df['spec2_path'] = f'{SPEC_DIR}/test/'+test_df['spectrogram_id'].astype(str)+'.npy'\ntest_df['eeg_to_spec'] = f'{SPEC_DIR}/test/'+test_df['eeg_id'].astype(str)+'.npy'\ndisplay(test_df.head(2))","metadata":{"execution":{"iopub.status.busy":"2024-02-06T21:33:49.063365Z","iopub.execute_input":"2024-02-06T21:33:49.063881Z","iopub.status.idle":"2024-02-06T21:33:49.778884Z","shell.execute_reply.started":"2024-02-06T21:33:49.063849Z","shell.execute_reply":"2024-02-06T21:33:49.77766Z"},"trusted":true},"execution_count":5,"outputs":[{"output_type":"display_data","data":{"text/plain":"       eeg_id  eeg_sub_id  eeg_label_offset_seconds  spectrogram_id  \\\n0  1628180742           0                       0.0          353733   \n1  1628180742           1                       6.0          353733   \n\n   spectrogram_sub_id  spectrogram_label_offset_seconds    label_id  \\\n0                   0                               0.0   127492639   \n1                   1                               6.0  3887563113   \n\n   patient_id expert_consensus  seizure_vote  ...  gpd_vote  lrda_vote  \\\n0       42516          Seizure             3  ...         0          0   \n1       42516          Seizure             3  ...         0          0   \n\n   grda_vote  other_vote                                           eeg_path  \\\n0          0           0  /kaggle/input/hms-harmful-brain-activity-class...   \n1          0           0  /kaggle/input/hms-harmful-brain-activity-class...   \n\n                                           spec_path  \\\n0  /kaggle/input/hms-harmful-brain-activity-class...   \n1  /kaggle/input/hms-harmful-brain-activity-class...   \n\n                         spec2_path                           eeg_to_spec  \\\n0  /kaggle/working/train/353733.npy  /kaggle/working/train/1628180742.npy   \n1  /kaggle/working/train/353733.npy  /kaggle/working/train/1628180742.npy   \n\n  class_name class_label  \n0    Seizure           0  \n1    Seizure           0  \n\n[2 rows x 21 columns]","text/html":"<div>\n<style scoped>\n    .dataframe tbody tr th:only-of-type {\n        vertical-align: middle;\n    }\n\n    .dataframe tbody tr th {\n        vertical-align: top;\n    }\n\n    .dataframe thead th {\n        text-align: right;\n    }\n</style>\n<table border=\"1\" class=\"dataframe\">\n  <thead>\n    <tr style=\"text-align: right;\">\n      <th></th>\n      <th>eeg_id</th>\n      <th>eeg_sub_id</th>\n      <th>eeg_label_offset_seconds</th>\n      <th>spectrogram_id</th>\n      <th>spectrogram_sub_id</th>\n      <th>spectrogram_label_offset_seconds</th>\n      <th>label_id</th>\n      <th>patient_id</th>\n      <th>expert_consensus</th>\n      <th>seizure_vote</th>\n      <th>...</th>\n      <th>gpd_vote</th>\n      <th>lrda_vote</th>\n      <th>grda_vote</th>\n      <th>other_vote</th>\n      <th>eeg_path</th>\n      <th>spec_path</th>\n      <th>spec2_path</th>\n      <th>eeg_to_spec</th>\n      <th>class_name</th>\n      <th>class_label</th>\n    </tr>\n  </thead>\n  <tbody>\n    <tr>\n      <th>0</th>\n      <td>1628180742</td>\n      <td>0</td>\n      <td>0.0</td>\n      <td>353733</td>\n      <td>0</td>\n      <td>0.0</td>\n      <td>127492639</td>\n      <td>42516</td>\n      <td>Seizure</td>\n      <td>3</td>\n      <td>...</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>/kaggle/input/hms-harmful-brain-activity-class...</td>\n      <td>/kaggle/input/hms-harmful-brain-activity-class...</td>\n      <td>/kaggle/working/train/353733.npy</td>\n      <td>/kaggle/working/train/1628180742.npy</td>\n      <td>Seizure</td>\n      <td>0</td>\n    </tr>\n    <tr>\n      <th>1</th>\n      <td>1628180742</td>\n      <td>1</td>\n      <td>6.0</td>\n      <td>353733</td>\n      <td>1</td>\n      <td>6.0</td>\n      <td>3887563113</td>\n      <td>42516</td>\n      <td>Seizure</td>\n      <td>3</td>\n      <td>...</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>0</td>\n      <td>/kaggle/input/hms-harmful-brain-activity-class...</td>\n      <td>/kaggle/input/hms-harmful-brain-activity-class...</td>\n      <td>/kaggle/working/train/353733.npy</td>\n      <td>/kaggle/working/train/1628180742.npy</td>\n      <td>Seizure</td>\n      <td>0</td>\n    </tr>\n  </tbody>\n</table>\n<p>2 rows × 21 columns</p>\n</div>"},"metadata":{}},{"output_type":"display_data","data":{"text/plain":"   spectrogram_id      eeg_id  patient_id  \\\n0          853520  3911565283        6885   \n\n                                            eeg_path  \\\n0  /kaggle/input/hms-harmful-brain-activity-class...   \n\n                                           spec_path  \\\n0  /kaggle/input/hms-harmful-brain-activity-class...   \n\n                        spec2_path                          eeg_to_spec  \n0  /kaggle/working/test/853520.npy  /kaggle/working/test/3911565283.npy  ","text/html":"<div>\n<style scoped>\n    .dataframe tbody tr th:only-of-type {\n        vertical-align: middle;\n    }\n\n    .dataframe tbody tr th {\n        vertical-align: top;\n    }\n\n    .dataframe thead th {\n        text-align: right;\n    }\n</style>\n<table border=\"1\" class=\"dataframe\">\n  <thead>\n    <tr style=\"text-align: right;\">\n      <th></th>\n      <th>spectrogram_id</th>\n      <th>eeg_id</th>\n      <th>patient_id</th>\n      <th>eeg_path</th>\n      <th>spec_path</th>\n      <th>spec2_path</th>\n      <th>eeg_to_spec</th>\n    </tr>\n  </thead>\n  <tbody>\n    <tr>\n      <th>0</th>\n      <td>853520</td>\n      <td>3911565283</td>\n      <td>6885</td>\n      <td>/kaggle/input/hms-harmful-brain-activity-class...</td>\n      <td>/kaggle/input/hms-harmful-brain-activity-class...</td>\n      <td>/kaggle/working/test/853520.npy</td>\n      <td>/kaggle/working/test/3911565283.npy</td>\n    </tr>\n  </tbody>\n</table>\n</div>"},"metadata":{}}]},{"cell_type":"code","source":"import librosa\n\nNAMES = ['LL','LP','RP','RR']\n\nFEATS = [['Fp1','F7','T3','T5','O1'],\n         ['Fp1','F3','C3','P3','O1'],\n         ['Fp2','F8','T4','T6','O2'],\n         ['Fp2','F4','C4','P4','O2']]\n\n\ndef spectrogram_from_eeg(eeg_id, split, display=False,):\n    \n    # LOAD MIDDLE 50 SECONDS OF EEG SERIES\n    #eeg_path = f\"{BASE_PATH}/{split}_eegs/{eeg_id}.parquet\"\n    PATH = f'/kaggle/input/hms-harmful-brain-activity-classification/{split}_eegs/'\n    eeg = pd.read_parquet(f'{PATH}{eeg_id}.parquet')\n    \n    middle = (len(eeg)-10_000)//2\n    eeg = eeg.iloc[middle:middle+10_000]\n    \n    # VARIABLE TO HOLD SPECTROGRAM\n    img = np.zeros((150, 200, 4),dtype='float32')\n    \n    #if display: plt.figure(figsize=(10,7))\n    signals = []\n    for k in range(4):\n        COLS = FEATS[k]\n        \n        for kk in range(4):\n        \n            # COMPUTE PAIR DIFFERENCES\n            x = eeg[COLS[kk]].values - eeg[COLS[kk+1]].values\n\n            # FILL NANS\n            m = np.nanmean(x)\n            if np.isnan(x).mean()<1: x = np.nan_to_num(x,nan=m)\n            else: x[:] = 0\n\n            # DENOISE\n#             if USE_WAVELET:\n#                 x = denoise(x, wavelet=USE_WAVELET)\n            signals.append(x)\n\n            # RAW SPECTROGRAM\n            mel_spec = librosa.feature.melspectrogram(y=x, sr=200, hop_length=len(x)//200, \n                  n_fft=1024, n_mels=150, fmin=0, fmax=20, win_length=150)\n\n            # LOG TRANSFORM\n            width = (mel_spec.shape[1]//10)*10\n            mel_spec_db = librosa.power_to_db(mel_spec, ref=np.max).astype(np.float32)[:,:width]\n\n            # STANDARDIZE TO -1 TO 1\n            mel_spec_db = (mel_spec_db+40)/40 \n            img[:,:,k] += mel_spec_db\n                \n        # AVERAGE THE 4 MONTAGE DIFFERENCES\n        img[:,:,k] /= 4.0\n        \n#         if display:\n#             plt.subplot(2,2,k+1)\n#             plt.imshow(img[:,:,k],aspect='auto',origin='lower')\n#             plt.title(f'EEG {eeg_id} - Spectrogram {NAMES[k]}')\n            \n#     if display: \n#         plt.show()\n#         plt.figure(figsize=(10,5))\n#         offset = 0\n#         for k in range(4):\n#             if k>0: offset -= signals[3-k].min()\n#             plt.plot(range(10_000),signals[k]+offset,label=NAMES[3-k])\n#             offset += signals[3-k].max()\n#         plt.legend()\n#         plt.title(f'EEG {eeg_id} Signals')\n#         plt.show()\n#         print(); print('#'*25); print()\n#         plt.close() \n    #np.save(f'{SPEC_DIR}/{split}/{eeg_id}',img)\n    return {eeg_id: img}","metadata":{"execution":{"iopub.status.busy":"2024-02-06T21:37:01.444002Z","iopub.execute_input":"2024-02-06T21:37:01.444425Z","iopub.status.idle":"2024-02-06T21:37:01.463631Z","shell.execute_reply.started":"2024-02-06T21:37:01.44439Z","shell.execute_reply":"2024-02-06T21:37:01.462594Z"},"trusted":true},"execution_count":6,"outputs":[]},{"cell_type":"code","source":"%%time\nDISPLAY = 4\nEEG_IDS = df.eeg_id.unique()\nEEG_IDS_TEST = test_df.eeg_id.unique()\n\n#train eegs\neegs = joblib.Parallel(n_jobs=-1, backend=\"loky\")(\n    joblib.delayed(spectrogram_from_eeg)(eeg_id, \"train\")\n    for eeg_id in tqdm(EEG_IDS, total=len(EEG_IDS))\n)\nall_eegs = {k: v for d in eegs for k, v in d.items()}\n\n#test eegs\n# _ = joblib.Parallel(n_jobs=-1, backend=\"loky\")(\n#     joblib.delayed(spectrogram_from_eeg)(eeg_id, \"test\")\n#     for eeg_id in tqdm(EEG_IDS_TEST, total=len(EEG_IDS_TEST))\n# )","metadata":{"_kg_hide-input":false,"_kg_hide-output":true,"execution":{"iopub.status.busy":"2024-02-06T21:37:19.577701Z","iopub.execute_input":"2024-02-06T21:37:19.578094Z","iopub.status.idle":"2024-02-06T21:54:49.559023Z","shell.execute_reply.started":"2024-02-06T21:37:19.578062Z","shell.execute_reply":"2024-02-06T21:54:49.557853Z"},"trusted":true},"execution_count":7,"outputs":[{"output_type":"display_data","data":{"text/plain":"  0%|          | 0/17089 [00:00<?, ?it/s]","application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"bc69e7d541fb4b5ea5ea9f85b329c2a0"}},"metadata":{}},{"name":"stdout","text":"CPU times: user 4min 14s, sys: 18.3 s, total: 4min 32s\nWall time: 17min 29s\n","output_type":"stream"}]},{"cell_type":"code","source":"np.save('eeg_specs',all_eegs)","metadata":{"execution":{"iopub.status.busy":"2024-02-06T21:55:47.55282Z","iopub.execute_input":"2024-02-06T21:55:47.55388Z","iopub.status.idle":"2024-02-06T21:56:18.694799Z","shell.execute_reply.started":"2024-02-06T21:55:47.553843Z","shell.execute_reply":"2024-02-06T21:56:18.693355Z"},"trusted":true},"execution_count":10,"outputs":[]},{"cell_type":"code","source":"# Define a function to process a single eeg_id\ndef process_spec(spec_id, split=\"train\"):\n    spec_path = f\"{BASE_PATH}/{split}_spectrograms/{spec_id}.parquet\"\n    spec = pd.read_parquet(spec_path)\n    spec = spec.fillna(0).values[:, 1:].T # fill NaN values with 0, transpose for (Time, Freq) -> (Freq, Time)\n    spec = spec.astype(\"float32\")\n    #np.save(f\"{SPEC_DIR}/{split}/{spec_id}.npy\", spec)\n    return {spec_id: spec}\n\n# Get unique spec_ids of train and valid data\nspec_ids = df[\"spectrogram_id\"].unique()\n\n# Parallelize the processing using joblib for training data\nspecs = joblib.Parallel(n_jobs=-1, backend=\"loky\")(\n    joblib.delayed(process_spec)(spec_id, \"train\")\n    for spec_id in tqdm(spec_ids, total=len(spec_ids))\n)\nall_specs = {k: v for d in specs for k, v in d.items()}\n\n# Get unique spec_ids of test data\n# test_spec_ids = test_df[\"spectrogram_id\"].unique()\n\n# # Parallelize the processing using joblib for test data\n# _ = joblib.Parallel(n_jobs=-1, backend=\"loky\")(\n#     joblib.delayed(process_spec)(spec_id, \"test\")\n#     for spec_id in tqdm(test_spec_ids, total=len(test_spec_ids))\n# )","metadata":{"execution":{"iopub.status.busy":"2024-02-06T21:57:57.688433Z","iopub.execute_input":"2024-02-06T21:57:57.689371Z","iopub.status.idle":"2024-02-06T22:01:32.783547Z","shell.execute_reply.started":"2024-02-06T21:57:57.689327Z","shell.execute_reply":"2024-02-06T22:01:32.782456Z"},"trusted":true},"execution_count":11,"outputs":[{"output_type":"display_data","data":{"text/plain":"  0%|          | 0/11138 [00:00<?, ?it/s]","application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"3ef49b6aa20445b5ae048fbdb3eadf06"}},"metadata":{}}]},{"cell_type":"code","source":"np.save('kaggle_specs',all_specs)","metadata":{"execution":{"iopub.status.busy":"2024-02-06T22:02:32.509321Z","iopub.execute_input":"2024-02-06T22:02:32.510319Z","iopub.status.idle":"2024-02-06T22:02:59.260121Z","shell.execute_reply.started":"2024-02-06T22:02:32.510279Z","shell.execute_reply":"2024-02-06T22:02:59.257955Z"},"trusted":true},"execution_count":13,"outputs":[]},{"cell_type":"code","source":"def build_augmenter(dim=image_size):\n    augmenters = [\n        keras_cv.layers.MixUp(alpha=2.0),\n        keras_cv.layers.RandomCutout(height_factor=(1.0, 1.0),\n                                     width_factor=(0.06, 0.1)), # freq-masking\n        keras_cv.layers.RandomCutout(height_factor=(0.06, 0.1),\n                                     width_factor=(1.0, 1.0)), # time-masking\n    ]\n    \n    def augment(img, label):\n        data = {\"images\":img, \"labels\":label}\n        for augmenter in augmenters:\n            if tf.random.uniform([]) < 0.5:\n                data = augmenter(data, training=True)\n        return data[\"images\"], data[\"labels\"]\n    \n    return augment\n\n\ndef build_decoder(with_labels=True, target_size=image_size, dtype=32):\n    def decode_signal(path, path_eeg, offset=None):\n        # Read .npy files and process the signal\n        file_bytes = tf.io.read_file(path)\n        sig = tf.io.decode_raw(file_bytes, tf.float32)\n        sig = sig[1024//dtype:]  # Remove header tag\n        sig = tf.reshape(sig, [200, -1])\n        \n        #Extract labeled subsample from full spectrogram using \"offset\"\n        if offset is not None: \n            #offset = offset // 2  # Only odd values are given\n            sig = sig[:, offset:offset+600]\n            \n            # Pad spectrogram to ensure the same input shape of [400, 300]\n            pad_size = tf.math.maximum(0, 600 - tf.shape(sig)[1])\n            sig = tf.pad(sig, [[0, 0], [0, pad_size]])\n            sig = tf.reshape(sig, [200, 600])\n        \n        # Log spectrogram \n        sig = tf.clip_by_value(sig, tf.math.exp(-4.0), tf.math.exp(8.0)) # avoid 0 in log\n        sig = tf.math.log(sig)\n        \n        # Normalize spectrogram\n        sig -= tf.math.reduce_mean(sig)\n        sig /= tf.math.reduce_std(sig) + 1e-6\n        \n        # Adding our eeg_spectogram\n        file_eeg = tf.io.read_file(path_eeg)\n        eeg = tf.io.decode_raw(file_eeg, tf.float32)\n        eeg = eeg[1024//dtype:]\n        eeg = tf.transpose(eeg)\n        eeg = tf.reshape(eeg, [200, -1])\n        \n        sig = tf.concat([sig, eeg], axis = 0)\n        \n        # Mono channel to 3 channels to use \"ImageNet\" weights\n        sig = tf.tile(sig[..., None], [1, 1, 3])\n        return sig\n    \n    def decode_label(label):\n        label = tf.one_hot(label, num_classes)\n        label = tf.cast(label, tf.float32)\n        label = tf.reshape(label, [num_classes])\n        return label\n    \n    def decode_with_labels(path, path_eeg, offset=None, label=None):\n        sig = decode_signal(path, path_eeg, offset)\n        label = decode_label(label)\n        return (sig, label)\n    \n    return decode_with_labels if with_labels else decode_signal\n\n\ndef build_dataset(paths, paths_eeg, offsets=None, labels=None, batch_size=32, cache=True,\n                  decode_fn=None, augment_fn=None,\n                  augment=False, repeat=True, shuffle=1024, \n                  cache_dir=\"\", drop_remainder=False):\n    if cache_dir != \"\" and cache is True:\n        os.makedirs(cache_dir, exist_ok=True)\n    \n    if decode_fn is None:\n        decode_fn = build_decoder(labels is not None)\n    \n    if augment_fn is None:\n        augment_fn = build_augmenter()\n    \n    AUTO = tf.data.experimental.AUTOTUNE\n    slices = (paths, paths_eeg, offsets) if labels is None else (paths, paths_eeg, offsets, labels)\n    \n    ds = tf.data.Dataset.from_tensor_slices(slices)\n    ds = ds.map(decode_fn, num_parallel_calls=AUTO)\n    ds = ds.cache(cache_dir) if cache else ds\n    ds = ds.repeat() if repeat else ds\n    if shuffle: \n        ds = ds.shuffle(shuffle, seed=42)\n        opt = tf.data.Options()\n        opt.experimental_deterministic = False\n        ds = ds.with_options(opt)\n    ds = ds.batch(batch_size, drop_remainder=drop_remainder)\n    ds = ds.map(augment_fn, num_parallel_calls=AUTO) if augment else ds\n    ds = ds.prefetch(AUTO)\n    return ds","metadata":{"execution":{"iopub.status.busy":"2024-02-06T17:27:31.540407Z","iopub.execute_input":"2024-02-06T17:27:31.540998Z","iopub.status.idle":"2024-02-06T17:27:31.562197Z","shell.execute_reply.started":"2024-02-06T17:27:31.540965Z","shell.execute_reply":"2024-02-06T17:27:31.561131Z"},"trusted":true},"execution_count":14,"outputs":[]},{"cell_type":"code","source":"from sklearn.model_selection import StratifiedGroupKFold\n\nsgkf = StratifiedGroupKFold(n_splits=5, shuffle=True, random_state=42)\n\ndf[\"fold\"] = -1\ndf.reset_index(drop=True, inplace=True)\nfor fold, (train_idx, valid_idx) in enumerate(\n    sgkf.split(df, y=df[\"class_label\"], groups=df[\"patient_id\"])\n):\n    df.loc[valid_idx, \"fold\"] = fold\ndf.groupby([\"fold\", \"class_name\"])[[\"eeg_id\"]].count().T","metadata":{"execution":{"iopub.status.busy":"2024-02-06T17:27:38.965976Z","iopub.execute_input":"2024-02-06T17:27:38.966458Z","iopub.status.idle":"2024-02-06T17:27:40.65246Z","shell.execute_reply.started":"2024-02-06T17:27:38.966426Z","shell.execute_reply":"2024-02-06T17:27:40.651566Z"},"trusted":true},"execution_count":15,"outputs":[{"execution_count":15,"output_type":"execute_result","data":{"text/plain":"fold           0                                     1                    ...  \\\nclass_name   GPD  GRDA   LPD  LRDA Other Seizure   GPD  GRDA   LPD  LRDA  ...   \neeg_id      2050  2605  1168  3793  3462    4007  4832  4856  4463  3117  ...   \n\nfold           3                         4                                  \nclass_name   LPD  LRDA Other Seizure   GPD  GRDA   LPD  LRDA Other Seizure  \neeg_id      2318  4296  3884    4787  1889  3700  4249  3002  4013    4045  \n\n[1 rows x 30 columns]","text/html":"<div>\n<style scoped>\n    .dataframe tbody tr th:only-of-type {\n        vertical-align: middle;\n    }\n\n    .dataframe tbody tr th {\n        vertical-align: top;\n    }\n\n    .dataframe thead tr th {\n        text-align: left;\n    }\n</style>\n<table border=\"1\" class=\"dataframe\">\n  <thead>\n    <tr>\n      <th>fold</th>\n      <th colspan=\"6\" halign=\"left\">0</th>\n      <th colspan=\"4\" halign=\"left\">1</th>\n      <th>...</th>\n      <th colspan=\"4\" halign=\"left\">3</th>\n      <th colspan=\"6\" halign=\"left\">4</th>\n    </tr>\n    <tr>\n      <th>class_name</th>\n      <th>GPD</th>\n      <th>GRDA</th>\n      <th>LPD</th>\n      <th>LRDA</th>\n      <th>Other</th>\n      <th>Seizure</th>\n      <th>GPD</th>\n      <th>GRDA</th>\n      <th>LPD</th>\n      <th>LRDA</th>\n      <th>...</th>\n      <th>LPD</th>\n      <th>LRDA</th>\n      <th>Other</th>\n      <th>Seizure</th>\n      <th>GPD</th>\n      <th>GRDA</th>\n      <th>LPD</th>\n      <th>LRDA</th>\n      <th>Other</th>\n      <th>Seizure</th>\n    </tr>\n  </thead>\n  <tbody>\n    <tr>\n      <th>eeg_id</th>\n      <td>2050</td>\n      <td>2605</td>\n      <td>1168</td>\n      <td>3793</td>\n      <td>3462</td>\n      <td>4007</td>\n      <td>4832</td>\n      <td>4856</td>\n      <td>4463</td>\n      <td>3117</td>\n      <td>...</td>\n      <td>2318</td>\n      <td>4296</td>\n      <td>3884</td>\n      <td>4787</td>\n      <td>1889</td>\n      <td>3700</td>\n      <td>4249</td>\n      <td>3002</td>\n      <td>4013</td>\n      <td>4045</td>\n    </tr>\n  </tbody>\n</table>\n<p>1 rows × 30 columns</p>\n</div>"},"metadata":{}}]},{"cell_type":"code","source":"# Sample from full data\nsample_df = df.groupby(\"spectrogram_id\").head(1).reset_index(drop=True)\ntrain_df = sample_df[sample_df.fold != 0]\nvalid_df = sample_df[sample_df.fold == 0]\nprint(f\"# Num Train: {len(train_df)} | Num Valid: {len(valid_df)}\")\n\n# Train\ntrain_paths = train_df.spec2_path.values\ntrain_eeg_paths = train_df.eeg_to_spec\ntrain_offsets = train_df.spectrogram_label_offset_seconds.values.astype(int)\ntrain_labels = train_df.class_label.values\ntrain_ds = build_dataset(train_paths,train_eeg_paths, train_offsets, train_labels, batch_size=batch_size,\n                         repeat=True, shuffle=True, augment=True, cache=True)\n\n# Valid\nvalid_paths = valid_df.spec2_path.values \nvalid_eeg_paths = valid_df.eeg_to_spec\nvalid_offsets = valid_df.spectrogram_label_offset_seconds.values.astype(int)\nvalid_labels = valid_df.class_label.values\nvalid_ds = build_dataset(valid_paths, valid_eeg_paths, valid_offsets, valid_labels, batch_size=batch_size,\n                         repeat=False, shuffle=False, augment=False, cache=True)","metadata":{"execution":{"iopub.status.busy":"2024-02-06T17:27:43.934275Z","iopub.execute_input":"2024-02-06T17:27:43.934906Z","iopub.status.idle":"2024-02-06T17:27:53.609673Z","shell.execute_reply.started":"2024-02-06T17:27:43.934875Z","shell.execute_reply":"2024-02-06T17:27:53.60871Z"},"trusted":true},"execution_count":16,"outputs":[{"name":"stdout","text":"# Num Train: 9166 | Num Valid: 1972\n","output_type":"stream"}]},{"cell_type":"code","source":"imgs, tars = next(iter(train_ds))\n\nnum_imgs = 1\nplt.figure(figsize=(16, 20))\nfor i in range(num_imgs):\n    #plt.subplot(1, 4, i + 1)\n    img = imgs[i].numpy()[...,0]  # Adjust as per your image data format\n    img -= img.min()\n    img /= img.max() + 1e-4\n    tar = label2name[np.argmax(tars[i].numpy())]\n    plt.imshow(img)\n    plt.title(f\"Target: {tar}\")\n    plt.axis('off')\n    \n#plt.tight_layout()\nplt.show()","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import math\n\ndef get_lr_callback(batch_size=8, mode='cos', epochs=5, plot=False):\n    lr_start, lr_max, lr_min = 5e-5, 6e-6 * batch_size, 1e-5\n    lr_ramp_ep, lr_sus_ep, lr_decay = 3, 0, 0.75\n\n    def lrfn(epoch):  # Learning rate update function\n        if epoch < lr_ramp_ep: lr = (lr_max - lr_start) / lr_ramp_ep * epoch + lr_start\n        elif epoch < lr_ramp_ep + lr_sus_ep: lr = lr_max\n        elif mode == 'exp': lr = (lr_max - lr_min) * lr_decay**(epoch - lr_ramp_ep - lr_sus_ep) + lr_min\n        elif mode == 'step': lr = lr_max * lr_decay**((epoch - lr_ramp_ep - lr_sus_ep) // 2)\n        elif mode == 'cos':\n            decay_total_epochs, decay_epoch_index = epochs - lr_ramp_ep - lr_sus_ep + 3, epoch - lr_ramp_ep - lr_sus_ep\n            phase = math.pi * decay_epoch_index / decay_total_epochs\n            lr = (lr_max - lr_min) * 0.5 * (1 + math.cos(phase)) + lr_min\n        return lr\n\n    if plot:  # Plot lr curve if plot is True\n        plt.figure(figsize=(10, 5))\n        plt.plot(np.arange(epochs), [lrfn(epoch) for epoch in np.arange(epochs)], marker='o')\n        plt.xlabel('epoch'); plt.ylabel('lr')\n        plt.title('LR Scheduler')\n        plt.show()\n\n    return tf.keras.callbacks.LearningRateScheduler(lrfn, verbose=False)  # Create lr callback\n\nlr_cb = get_lr_callback(64, mode='cos', plot=True)","metadata":{"execution":{"iopub.status.busy":"2024-02-06T17:27:58.683741Z","iopub.execute_input":"2024-02-06T17:27:58.684106Z","iopub.status.idle":"2024-02-06T17:27:59.232858Z","shell.execute_reply.started":"2024-02-06T17:27:58.684075Z","shell.execute_reply":"2024-02-06T17:27:59.2321Z"},"trusted":true},"execution_count":17,"outputs":[{"output_type":"display_data","data":{"text/plain":"<Figure size 1000x500 with 1 Axes>","image/png":"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"},"metadata":{}}]},{"cell_type":"code","source":"from tensorflow.keras.applications.efficientnet_v2 import EfficientNetV2L","metadata":{"execution":{"iopub.status.busy":"2024-02-06T17:28:01.519679Z","iopub.execute_input":"2024-02-06T17:28:01.520078Z","iopub.status.idle":"2024-02-06T17:28:01.525966Z","shell.execute_reply.started":"2024-02-06T17:28:01.520047Z","shell.execute_reply":"2024-02-06T17:28:01.52509Z"},"trusted":true},"execution_count":18,"outputs":[]},{"cell_type":"code","source":"#base_model = EfficientNetV2M(input_shape = (480, 480, 3), include_top = True, weights = 'imagenet')","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"base_model = EfficientNetV2L(input_shape = (400, 600, 3), include_top = False, weights = 'imagenet')","metadata":{"execution":{"iopub.status.busy":"2024-02-06T17:28:05.782397Z","iopub.execute_input":"2024-02-06T17:28:05.782767Z","iopub.status.idle":"2024-02-06T17:28:18.76687Z","shell.execute_reply.started":"2024-02-06T17:28:05.782736Z","shell.execute_reply":"2024-02-06T17:28:18.766014Z"},"trusted":true},"execution_count":19,"outputs":[{"name":"stdout","text":"Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/efficientnet_v2/efficientnetv2-l_notop.h5\n473176280/473176280 [==============================] - 2s 0us/step\n","output_type":"stream"}]},{"cell_type":"code","source":"for layer in base_model.layers:\n    layer.trainable = False","metadata":{"execution":{"iopub.status.busy":"2024-02-06T17:28:21.481551Z","iopub.execute_input":"2024-02-06T17:28:21.482331Z","iopub.status.idle":"2024-02-06T17:28:21.524618Z","shell.execute_reply.started":"2024-02-06T17:28:21.482296Z","shell.execute_reply":"2024-02-06T17:28:21.523781Z"},"trusted":true},"execution_count":20,"outputs":[]},{"cell_type":"code","source":"x = base_model.output\nx = tf.keras.layers.Flatten()(x)\nx = tf.keras.layers.Dense(32, activation=\"relu\")(x)\nx = tf.keras.layers.Dropout(0.1)(x)\n\n# Add a final sigmoid layer with 1 node for classification output\npredictions = tf.keras.layers.Dense(6, activation=\"softmax\")(x)\nmodel_final = tf.keras.models.Model(inputs = base_model.input, outputs = predictions)","metadata":{"execution":{"iopub.status.busy":"2024-02-06T17:28:27.288783Z","iopub.execute_input":"2024-02-06T17:28:27.289781Z","iopub.status.idle":"2024-02-06T17:28:27.416329Z","shell.execute_reply.started":"2024-02-06T17:28:27.289737Z","shell.execute_reply":"2024-02-06T17:28:27.415275Z"},"trusted":true},"execution_count":21,"outputs":[]},{"cell_type":"code","source":"model_final.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=1e-4),\n              loss=LOSS, metrics = [LOSS])","metadata":{"execution":{"iopub.status.busy":"2024-02-06T17:28:29.911863Z","iopub.execute_input":"2024-02-06T17:28:29.912242Z","iopub.status.idle":"2024-02-06T17:28:29.953436Z","shell.execute_reply.started":"2024-02-06T17:28:29.912197Z","shell.execute_reply":"2024-02-06T17:28:29.952673Z"},"trusted":true},"execution_count":22,"outputs":[]},{"cell_type":"code","source":"ckpt_cb = tf.keras.callbacks.ModelCheckpoint(\"best_model.keras\",\n                                         monitor='val_loss',\n                                         save_best_only=True,\n                                         save_weights_only=False,\n                                         mode='min')","metadata":{"execution":{"iopub.status.busy":"2024-02-06T17:28:31.939649Z","iopub.execute_input":"2024-02-06T17:28:31.940036Z","iopub.status.idle":"2024-02-06T17:28:31.94476Z","shell.execute_reply.started":"2024-02-06T17:28:31.940005Z","shell.execute_reply":"2024-02-06T17:28:31.943725Z"},"trusted":true},"execution_count":23,"outputs":[]},{"cell_type":"code","source":"history = model_final.fit(\n    train_ds, \n    epochs=5,\n    callbacks=[ckpt_cb, lr_cb], \n    steps_per_epoch=len(train_df)//64,\n    validation_data=valid_ds, \n    verbose=1\n)","metadata":{"execution":{"iopub.status.busy":"2024-02-06T17:28:35.031017Z","iopub.execute_input":"2024-02-06T17:28:35.031738Z"},"trusted":true},"execution_count":null,"outputs":[{"name":"stdout","text":"Epoch 1/5\n","output_type":"stream"},{"name":"stderr","text":"2024-02-06 17:29:02.405354: E tensorflow/core/grappler/optimizers/meta_optimizer.cc:961] layout failed: INVALID_ARGUMENT: Size of values 0 does not match size of permutation 4 @ fanin shape inmodel/block1b_drop/dropout/SelectV2-2-TransposeNHWCToNCHW-LayoutOptimizer\nWARNING: All log messages before absl::InitializeLog() is called are written to STDERR\nI0000 00:00:1707240555.396162     217 device_compiler.h:186] Compiled cluster using XLA!  This line is logged at most once for the lifetime of the process.\n","output_type":"stream"},{"name":"stdout","text":"143/143 [==============================] - ETA: 0s - loss: 1.7565 - kl_divergence: 1.7565","output_type":"stream"}]},{"cell_type":"markdown","source":"# Predictions","metadata":{}},{"cell_type":"code","source":"test_paths = test_df.spec2_path.values\ntest_eeg_paths = test_df.eeg_to_spec\ntest_ds = build_dataset(test_paths, test_eeg_paths, batch_size=min(CFG.batch_size, len(test_df)),\n                         repeat=False, shuffle=False, cache=False, augment=False)","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"preds = model_final.predict(test_ds)","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"pred_df = test_df[[\"eeg_id\"]].copy()\ntarget_cols = [x.lower()+'_vote' for x in class_names]\npred_df[target_cols] = preds.tolist()\n\nsub_df = pd.read_csv(f'{BASE_PATH}/sample_submission.csv')\nsub_df = sub_df[[\"eeg_id\"]].copy()\nsub_df = sub_df.merge(pred_df, on=\"eeg_id\", how=\"left\")\nsub_df.to_csv(\"submission.csv\", index=False)\nsub_df.head()","metadata":{},"execution_count":null,"outputs":[]}]}