{"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":"nvidiaTeslaT4","dataSources":[{"sourceId":59093,"databundleVersionId":7469972,"sourceType":"competition"},{"sourceId":1493925,"sourceType":"datasetVersion","datasetId":877193},{"sourceId":7392775,"sourceType":"datasetVersion","datasetId":4297782},{"sourceId":7447509,"sourceType":"datasetVersion","datasetId":4334995},{"sourceId":9841,"sourceType":"modelInstanceVersion","modelInstanceId":8013}],"dockerImageVersionId":30646,"isInternetEnabled":false,"language":"python","sourceType":"notebook","isGpuEnabled":true}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"import pandas as pd\nimport numpy as np\nimport matplotlib.pyplot as plt\nimport tensorflow as tf\nimport sys\nimport pickle\nimport re\nimport os, gc\nos.environ[\"CUDA_VISIBLE_DEVICES\"]=\"0,1\"\nprint('TensorFlow version =',tf.__version__)","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2024-02-10T01:12:12.349996Z","iopub.execute_input":"2024-02-10T01:12:12.350443Z","iopub.status.idle":"2024-02-10T01:12:27.614726Z","shell.execute_reply.started":"2024-02-10T01:12:12.350416Z","shell.execute_reply":"2024-02-10T01:12:27.613806Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Data Preprocessing\nThis notebook is based off of Chris Deotte:\n- https://www.kaggle.com/code/cdeotte/efficientnetb0-starter-lb-0-43","metadata":{}},{"cell_type":"code","source":"df = pd.read_csv(\"/kaggle/input/hms-harmful-brain-activity-classification/train.csv\")\n\n# absDimensions of the Dataset\nprint(f\"DF Shape: {df.shape}\")\nprint(f'Unique EEG: {df[\"eeg_id\"].nunique()}')\nprint(f'Unique Spectrogram: {df[\"spectrogram_id\"].nunique()}')\ndf.head()","metadata":{"execution":{"iopub.status.busy":"2024-02-10T01:12:27.616364Z","iopub.execute_input":"2024-02-10T01:12:27.616905Z","iopub.status.idle":"2024-02-10T01:12:27.876623Z","shell.execute_reply.started":"2024-02-10T01:12:27.616877Z","shell.execute_reply":"2024-02-10T01:12:27.875629Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# --------- Transform subsampels into sampels ---------\n# subsamples ===> samples\n\nVOTES = ['seizure_vote', 'lpd_vote', 'gpd_vote', 'lrda_vote', 'grda_vote', 'other_vote']\n\n\n\ntrain = df.groupby('eeg_id').agg({\n 'spectrogram_id': 'first',\n 'patient_id': 'first',\n}).reset_index()\n\n# Adds together all the votes based on the EEG_ID\ntmp = df.groupby('eeg_id')[VOTES].agg('sum')\nfor v in VOTES:\n train[v] = tmp[v].values\n\ntrain['target'] = train[VOTES].sum(axis=1)\ntrain[VOTES] = train[VOTES].div(train['target'], axis=0)\ntrain['target'] = train[VOTES].idxmax(axis=1)\n\n# Shape and View\nprint(f\"Train Shape: {train.shape}\")\ntrain.head()","metadata":{"execution":{"iopub.status.busy":"2024-02-10T01:12:27.881475Z","iopub.execute_input":"2024-02-10T01:12:27.881789Z","iopub.status.idle":"2024-02-10T01:12:27.946854Z","shell.execute_reply.started":"2024-02-10T01:12:27.881764Z","shell.execute_reply":"2024-02-10T01:12:27.945943Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"\n#### DATASET","metadata":{}},{"cell_type":"code","source":"%%time\n\nREAD = False\n\n\n# NPY Data taken from Chris Deotte. Appreciate the smaller mem usage of .npy format \nif READ:\n spectrograms = np.load('/kaggle/input/brain-spectrograms/specs.npy',allow_pickle=True).item()\n eegs = np.load('/kaggle/input/brain-eeg-spectrograms/eeg_specs.npy',allow_pickle=True).item()\n \n example = train[\"spectrogram_id\"].iloc[0]\n example_eeg = train[\"eeg_id\"].iloc[0]\n\n print(f\"Size of EEG Dictionary: {len(all_eegs)}\")\n print(f\"Shape of EEG Item: {all_eegs[example_eeg].shape}\")\n\n print(f\"Size of Spectogram Dictionary: {len(spectrograms)}\")\n print(f\"Shape of Spectogram Item: {spectrograms[example].shape}\")\nelse:\n spectrograms = {}\n eegs = {}\n ","metadata":{"execution":{"iopub.status.busy":"2024-02-10T01:12:27.947912Z","iopub.execute_input":"2024-02-10T01:12:27.948171Z","iopub.status.idle":"2024-02-10T01:12:27.954744Z","shell.execute_reply.started":"2024-02-10T01:12:27.948148Z","shell.execute_reply":"2024-02-10T01:12:27.953903Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#### Building a Data Generator\nMain Reference:\nhttps://stanford.edu/~shervine/blog/keras-how-to-generate-data-on-the-fly\n\nThings to do:\n - Load from spectorgram & eeg dictionary\n ","metadata":{}},{"cell_type":"code","source":"import numpy as np\nimport keras\n\nclass DataGenerator(keras.utils.Sequence):\n 'Generates data for Keras'\n def __init__(self,\n data,\n state=\"train\", \n batch_size=32,\n n_classes=6,\n shuffle=True, \n specs=spectrograms,\n eegs=eegs):\n 'Initialization'\n self.data = data\n self.state = state\n self.batch_size = batch_size\n self.n_classes = n_classes\n self.specs = specs\n self.eegs = eegs\n self.shuffle = shuffle\n self.on_epoch_end()\n\n def __len__(self):\n 'Denotes the number of batches per epoch'\n return int(np.floor(len(self.data) / self.batch_size))\n \n def __getitem__(self, index):\n 'Generate one batch of data'\n # Generate indexes of the batch\n indexes = self.indexes[index*self.batch_size:(index+1)*self.batch_size]\n\n # Generate data\n X, y = self.__data_generation(indexes)\n\n return X, y\n\n def on_epoch_end(self):\n 'Updates indexes after each epoch'\n self.indexes = np.arange(len(self.data))\n if self.shuffle == True:\n np.random.shuffle(self.indexes)\n\n def __data_generation(self, indexes):\n 'Generates data containing batch_size samples' # X : (n_samples, *dim, n_channels)\n # Initialization\n X = np.zeros((len(indexes),128,256,4),dtype='float32')\n y = np.zeros((len(indexes),self.n_classes),dtype='float32')\n img = np.ones((128,256),dtype='float32')\n \n for n, i in enumerate(indexes):\n \n # date row\n cur = self.data.iloc[i]\n \n # images of spec and eeg\n spec = self.specs[cur.spectrogram_id] # Not in use as of now\n eeg = self.eegs[cur.eeg_id]\n\n # add the data on\n X[n] = eeg\n \n if self.state == \"train\":\n y[n,] = cur[VOTES]\n \n return X, y\n ","metadata":{"execution":{"iopub.status.busy":"2024-02-10T01:12:27.955916Z","iopub.execute_input":"2024-02-10T01:12:27.956222Z","iopub.status.idle":"2024-02-10T01:12:27.968925Z","shell.execute_reply.started":"2024-02-10T01:12:27.956200Z","shell.execute_reply":"2024-02-10T01:12:27.968224Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Checking the Data Shape and for Testing\nif READ:\n gen = DataGenerator(train,\n batch_size=6)\n\n# gen.__getitem__(0)","metadata":{"execution":{"iopub.status.busy":"2024-02-10T01:12:27.970301Z","iopub.execute_input":"2024-02-10T01:12:27.970595Z","iopub.status.idle":"2024-02-10T01:12:27.983307Z","shell.execute_reply.started":"2024-02-10T01:12:27.970572Z","shell.execute_reply":"2024-02-10T01:12:27.982457Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#### Model Definitions\n\nBuilding the Model","metadata":{}},{"cell_type":"code","source":"import tensorflow as tf\nfrom tensorflow.keras.layers import Dense\n\n# building model\ndef build_cnn():\n inp = tf.keras.Input(shape=(128, 256, 4))\n \n # EfficientNetB0 Base Model\n base_model = tf.keras.applications.efficientnet.EfficientNetB0(include_top=False, weights=\"imagenet\", input_shape=None)\n\n base_model.load_weights('/kaggle/input/keras-pretrained-models/EfficientNetB0_NoTop_ImageNet.h5')\n \n x = [inp[:,:,:,i:i+1] for i in range(4)]\n x = tf.keras.layers.Concatenate(axis=1)(x)\n \n \n # Model and Additional layera\n x = base_model(x)\n x = tf.keras.layers.GlobalAveragePooling2D()(x)\n x = tf.keras.layers.Dense(6,activation='softmax', dtype='float32')(x)\n\n model = tf.keras.Model(inputs=inp, outputs=x)\n opt = tf.keras.optimizers.Adam(learning_rate=1e-3)\n loss = tf.keras.losses.KLDivergence()\n \n model.compile(loss=loss, optimizer=opt, metrics=['accuracy', 'kullback_leibler_divergence'])\n \n return model\n","metadata":{"execution":{"iopub.status.busy":"2024-02-10T01:12:27.984342Z","iopub.execute_input":"2024-02-10T01:12:27.984621Z","iopub.status.idle":"2024-02-10T01:12:28.109817Z","shell.execute_reply.started":"2024-02-10T01:12:27.984599Z","shell.execute_reply":"2024-02-10T01:12:28.108971Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Build \n\n* Using Multiple GPU's in order to Train and have enough RAM. This is defined in the strategy variable.\n* I have choosen to not use a validation data set though plan to in the future.\n* I also do not use the Spectrogram and only use EEG's for this model\n\n\n#### v0_HMS_ConvNet:\n\n534/534 [==============================] - 263s 371ms/step - loss: 0.7295 - accuracy: 0.6474 - kullback_leibler_divergence: 0.7295\nEpoch 2/4\n\n534/534 [==============================] - 198s 370ms/step - loss: 0.5857 - accuracy: 0.7108 - kullback_leibler_divergence: 0.5857\nEpoch 3/4\n\n534/534 [==============================] - 197s 369ms/step - loss: 0.5344 - accuracy: 0.7309 - kullback_leibler_divergence: 0.5344\nEpoch 4/4\n\n534/534 [==============================] - 197s 370ms/step - loss: 0.4868 - accuracy: 0.7488 - kullback_leibler_divergence: 0.4868","metadata":{}},{"cell_type":"code","source":"strategy = tf.distribute.MirroredStrategy()\n\nTRAIN = False\nSAVE = False\n\nif TRAIN:\n\n gen = DataGenerator(data=train, \n batch_size=32, \n n_classes=6,\n shuffle=True)\n\n with strategy.scope():\n v0_HMS_ConvNet = build_cnn() \n\n v0_HMS_ConvNet._name = \"v0_HMS_ConvNet\"\n\n v0_HMS_ConvNet.fit(gen,\n verbose=1, \n epochs=4)\n if SAVE:\n\n filename = \"v0_HMS_ConvNet.pkl\"\n\n with open(filename, 'wb') as file: \n pickle.dump(v0_HMS_ConvNet, file)","metadata":{"execution":{"iopub.status.busy":"2024-02-10T01:12:28.110878Z","iopub.execute_input":"2024-02-10T01:12:28.111183Z","iopub.status.idle":"2024-02-10T01:12:29.204626Z","shell.execute_reply.started":"2024-02-10T01:12:28.111152Z","shell.execute_reply":"2024-02-10T01:12:29.203818Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# PREDICTIONS\n\nBegin with the correct formating of the data","metadata":{}},{"cell_type":"code","source":"test_df = pd.read_csv(\"/kaggle/input/hms-harmful-brain-activity-classification/test.csv\")\n\ntest_df = test_df.groupby('eeg_id').agg({\n 'spectrogram_id': 'first',\n 'patient_id': 'first',\n}).reset_index()\n\n\n# Shape and View\nprint(f\"Train Shape: {test_df.shape}\")\ntest_df.head()","metadata":{"execution":{"iopub.status.busy":"2024-02-10T01:12:29.207319Z","iopub.execute_input":"2024-02-10T01:12:29.207639Z","iopub.status.idle":"2024-02-10T01:12:29.224601Z","shell.execute_reply.started":"2024-02-10T01:12:29.207612Z","shell.execute_reply":"2024-02-10T01:12:29.223709Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#### Processing the Parquets\n\nCleaning the data is also taken from Chris Deotte\n- This includes his code to read the parquets\n","metadata":{}},{"cell_type":"code","source":"import pywt, librosa\n\nUSE_WAVELET = None \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\ndef spectrogram_from_eeg(parquet_path, display=False):\n \n # LOAD MIDDLE 50 SECONDS OF EEG SERIES\n eeg = pd.read_parquet(parquet_path)\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((128,256,4),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)//256, \n n_fft=1024, n_mels=128, fmin=0, fmax=20, win_length=128)\n\n # LOG TRANSFORM\n width = (mel_spec.shape[1]//32)*32\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 \n return img","metadata":{"execution":{"iopub.status.busy":"2024-02-10T01:12:29.226248Z","iopub.execute_input":"2024-02-10T01:12:29.226635Z","iopub.status.idle":"2024-02-10T01:12:29.325282Z","shell.execute_reply.started":"2024-02-10T01:12:29.226599Z","shell.execute_reply":"2024-02-10T01:12:29.324556Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# READ ALL EEG SPECTROGRAMS\nPATH2 = '/kaggle/input/hms-harmful-brain-activity-classification/test_eegs/'\nDISPLAY = 1\nEEG_IDS2 = test_df.eeg_id.unique()\nall_eegs2 = {}\n\nprint('Converting Test EEG to Spectrograms...'); print()\nfor i,eeg_id in enumerate(EEG_IDS2):\n \n # CREATE SPECTROGRAM FROM EEG PARQUET\n img = spectrogram_from_eeg(f'{PATH2}{eeg_id}.parquet', i