{"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":"gpu","dataSources":[{"sourceId":59093,"databundleVersionId":7469972,"sourceType":"competition"},{"sourceId":7392775,"sourceType":"datasetVersion","datasetId":4297782},{"sourceId":7447509,"sourceType":"datasetVersion","datasetId":4334995}],"dockerImageVersionId":30648,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":true}},"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\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","execution":{"iopub.status.busy":"2024-03-12T05:55:33.049123Z","iopub.execute_input":"2024-03-12T05:55:33.049799Z","iopub.status.idle":"2024-03-12T05:55:33.443573Z","shell.execute_reply.started":"2024-03-12T05:55:33.049765Z","shell.execute_reply":"2024-03-12T05:55:33.442180Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import albumentations as A\nimport gc\nimport matplotlib.pyplot as plt\nimport math\nimport multiprocessing\nimport numpy as np\nimport os\nimport pandas as pd\nimport random\nimport time\nimport timm\nimport torch\nimport torch.nn as nn\nimport torch.nn.functional as F\n\n\nfrom albumentations.pytorch import ToTensorV2\nfrom glob import glob\nfrom torch.utils.data import DataLoader, Dataset\nfrom tqdm import tqdm\nfrom typing import Dict, List\n\nos.environ[\"CUDA_VISIBLE_DEVICES\"] = \"0,1\"\ndevice = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\nprint('Using', torch.cuda.device_count(), 'GPU(s)')\n","metadata":{"execution":{"iopub.status.busy":"2024-03-12T05:55:33.445552Z","iopub.execute_input":"2024-03-12T05:55:33.446495Z","iopub.status.idle":"2024-03-12T05:55:43.326688Z","shell.execute_reply.started":"2024-03-12T05:55:33.446457Z","shell.execute_reply":"2024-03-12T05:55:43.325598Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class config:\n model = \"resnet50\"\n epoch = 10\n lr = 1\n batchsize = 32\n splits = 5\n momentum = 0.9\n MAX_GRAD_NORM = 1e7\n WEIGHT_DECAY = 0.5\n device = \"cpu\"\n FOLDS=5\n AMP= True\nclass paths:\n preloadedeeg = \"/kaggle/input/brain-eeg-spectrograms/eeg_specs.npy\"\n train_eeg_dir = \"/kaggle/input/hms-harmful-brain-activity-classification/train_eegs\"\n train_spec_dir = \"/kaggle/input/hms-harmful-brain-activity-classification/train_spectrograms\"\n train_csv = \"/kaggle/input/hms-harmful-brain-activity-classification/train.csv\"\n test_csv = \"/kaggle/input/hms-harmful-brain-activity-classification/test.csv\"\n test_eeg = \"/kaggle/input/hms-harmful-brain-activity-classification/test_eegs\"\n test_spec = \"/kaggle/input/hms-harmful-brain-activity-classification/test_spectrograms\"\n out = \"/kaggle/working/\"","metadata":{"execution":{"iopub.status.busy":"2024-03-12T07:39:49.849151Z","iopub.execute_input":"2024-03-12T07:39:49.849573Z","iopub.status.idle":"2024-03-12T07:39:49.857149Z","shell.execute_reply.started":"2024-03-12T07:39:49.849524Z","shell.execute_reply":"2024-03-12T07:39:49.856107Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df = pd.read_csv(paths.train_csv)\ntargets = df.columns[-6:]\n","metadata":{"execution":{"iopub.status.busy":"2024-03-12T05:55:43.336702Z","iopub.execute_input":"2024-03-12T05:55:43.337033Z","iopub.status.idle":"2024-03-12T05:55:43.605446Z","shell.execute_reply.started":"2024-03-12T05:55:43.337004Z","shell.execute_reply":"2024-03-12T05:55:43.604237Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"targets","metadata":{"execution":{"iopub.status.busy":"2024-03-12T05:55:43.608519Z","iopub.execute_input":"2024-03-12T05:55:43.608968Z","iopub.status.idle":"2024-03-12T05:55:43.616843Z","shell.execute_reply.started":"2024-03-12T05:55:43.608930Z","shell.execute_reply":"2024-03-12T05:55:43.615709Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_df = df.groupby('eeg_id')[['spectrogram_id','spectrogram_label_offset_seconds']].agg({\n 'spectrogram_id':'first',\n 'spectrogram_label_offset_seconds':'min'\n})\ntrain_df.columns = ['spectogram_id','min']\n\naux = df.groupby('eeg_id')[['spectrogram_id','spectrogram_label_offset_seconds']].agg({\n 'spectrogram_label_offset_seconds':'max'\n})\ntrain_df['max'] = aux\n\naux = df.groupby('eeg_id')[['patient_id']].agg('first')\ntrain_df['patient_id'] = aux\n\naux = df.groupby('eeg_id')[targets].agg('sum')\nfor label in targets:\n train_df[label] = aux[label].values\n \ny_data = train_df[targets].values\ny_data = y_data / y_data.sum(axis=1,keepdims=True)\ntrain_df[targets] = y_data\n\naux = df.groupby('eeg_id')[['expert_consensus']].agg('first')\ntrain_df['target'] = aux\n\ntrain_df = train_df.reset_index()\nprint('Train non-overlapp eeg_id shape:', train_df.shape )\ntrain_df.head()","metadata":{"execution":{"iopub.status.busy":"2024-03-12T05:55:43.618337Z","iopub.execute_input":"2024-03-12T05:55:43.618700Z","iopub.status.idle":"2024-03-12T05:55:43.726788Z","shell.execute_reply.started":"2024-03-12T05:55:43.618666Z","shell.execute_reply":"2024-03-12T05:55:43.725673Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"spectrograms = np.load('/kaggle/input/brain-spectrograms/specs.npy',allow_pickle=True).item()\neegs = np.load(paths.preloadedeeg,allow_pickle = True).item()\n","metadata":{"execution":{"iopub.status.busy":"2024-03-12T05:55:43.727932Z","iopub.execute_input":"2024-03-12T05:55:43.728259Z","iopub.status.idle":"2024-03-12T05:57:52.483743Z","shell.execute_reply.started":"2024-03-12T05:55:43.728213Z","shell.execute_reply":"2024-03-12T05:57:52.482828Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from sklearn.model_selection import KFold, GroupKFold\ngfold = GroupKFold(n_splits = config.splits)\nfor fold,(trainindex,validationindex) in enumerate(gfold.split(train_df,train_df[\"target\"],train_df[\"patient_id\"])):\n train_df.loc[validationindex,\"fold\"] = int(fold)\n","metadata":{"execution":{"iopub.status.busy":"2024-03-12T05:57:52.485590Z","iopub.execute_input":"2024-03-12T05:57:52.486019Z","iopub.status.idle":"2024-03-12T05:57:52.517139Z","shell.execute_reply.started":"2024-03-12T05:57:52.485982Z","shell.execute_reply":"2024-03-12T05:57:52.516317Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"\ntrain_df","metadata":{"execution":{"iopub.status.busy":"2024-03-12T05:57:52.518525Z","iopub.execute_input":"2024-03-12T05:57:52.519332Z","iopub.status.idle":"2024-03-12T05:57:52.551448Z","shell.execute_reply.started":"2024-03-12T05:57:52.519291Z","shell.execute_reply":"2024-03-12T05:57:52.550092Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class CustomDataset():\n def __init__(self,traindf,config,mode:str=\"train\",specs:dict[int,np.ndarray]=spectrograms,eegs:dict[int,np.ndarray]=eegs):\n self.traindf = traindf\n self.specs = specs\n self.eeg = eegs\n self.mode = mode\n def __len__(self):\n return len(self.traindf)\n def __getitem__(self,idx):\n X = np.zeros((128, 256, 8), dtype='float32')\n y = np.zeros(6, dtype='float32')\n img = np.ones((128,256), dtype='float32')\n row = self.traindf.iloc[idx]\n if self.mode=='test': \n r = 0\n else: \n r = int((row['min'] + row['max']) // 4)\n \n for region in range(4):\n img = self.specs[row.spectogram_id][r:r+300, region*100:(region+1)*100].T\n \n # Log transform spectogram\n img = np.clip(img, np.exp(-4), np.exp(8))\n img = np.log(img)\n\n # Standarize per image\n ep = 1e-6\n mu = np.nanmean(img.flatten())\n std = np.nanstd(img.flatten())\n img = (img-mu)/(std+ep)\n img = np.nan_to_num(img, nan=0.0)\n X[14:-14, :, region] = img[:, 22:-22] / 2.0\n img = self.eeg[row.eeg_id]\n X[:, :, 4:] = img\n \n X = torch.tensor(X)\n spectograms = [X[:, :, i:i+1] for i in range(4)]\n spectograms = torch.cat(spectograms, dim=0)\n \n # === Get EEG spectograms ===\n eegs = [X[:, :, i:i+1] for i in range(4,8)]\n eegs = torch.cat(eegs, dim=0)\n \n # === Reshape (512,512,3) ===\n x = torch.cat([spectograms, eegs], dim=1)\n x = torch.cat([x,x,x], dim=2)\n x = x.permute(2, 0, 1)\n if self.mode != 'test':\n y = row[targets].values.astype(np.float32) \n \n return {\"data\":x,\n \"target\":y}\n ","metadata":{"execution":{"iopub.status.busy":"2024-03-12T06:16:43.328025Z","iopub.execute_input":"2024-03-12T06:16:43.328900Z","iopub.status.idle":"2024-03-12T06:16:43.347061Z","shell.execute_reply.started":"2024-03-12T06:16:43.328862Z","shell.execute_reply":"2024-03-12T06:16:43.346094Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"customdataset = CustomDataset(train_df,config)","metadata":{"execution":{"iopub.status.busy":"2024-03-12T06:16:43.517451Z","iopub.execute_input":"2024-03-12T06:16:43.517882Z","iopub.status.idle":"2024-03-12T06:16:43.522712Z","shell.execute_reply.started":"2024-03-12T06:16:43.517848Z","shell.execute_reply":"2024-03-12T06:16:43.521652Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"customdataset[100][\"data\"].shape","metadata":{"execution":{"iopub.status.busy":"2024-03-12T06:16:43.712481Z","iopub.execute_input":"2024-03-12T06:16:43.712926Z","iopub.status.idle":"2024-03-12T06:16:43.730942Z","shell.execute_reply.started":"2024-03-12T06:16:43.712892Z","shell.execute_reply":"2024-03-12T06:16:43.729828Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from torch.utils.data import DataLoader\ntrainloader = DataLoader(customdataset,batch_size = config.batchsize)","metadata":{"execution":{"iopub.status.busy":"2024-03-12T06:16:43.889339Z","iopub.execute_input":"2024-03-12T06:16:43.889753Z","iopub.status.idle":"2024-03-12T06:16:43.894898Z","shell.execute_reply.started":"2024-03-12T06:16:43.889723Z","shell.execute_reply":"2024-03-12T06:16:43.893946Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import timm","metadata":{"execution":{"iopub.status.busy":"2024-03-12T06:16:44.091374Z","iopub.execute_input":"2024-03-12T06:16:44.091729Z","iopub.status.idle":"2024-03-12T06:16:44.096052Z","shell.execute_reply.started":"2024-03-12T06:16:44.091703Z","shell.execute_reply":"2024-03-12T06:16:44.095153Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from torchvision.transforms import transforms\ntras = transforms.Compose([\n transforms.Resize((224,224))\n])","metadata":{"execution":{"iopub.status.busy":"2024-03-12T06:16:44.444318Z","iopub.execute_input":"2024-03-12T06:16:44.444795Z","iopub.status.idle":"2024-03-12T06:16:44.450851Z","shell.execute_reply.started":"2024-03-12T06:16:44.444758Z","shell.execute_reply":"2024-03-12T06:16:44.449686Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import torchvision.models as models\n","metadata":{"execution":{"iopub.status.busy":"2024-03-12T06:16:44.680650Z","iopub.execute_input":"2024-03-12T06:16:44.681088Z","iopub.status.idle":"2024-03-12T06:16:44.685724Z","shell.execute_reply.started":"2024-03-12T06:16:44.681038Z","shell.execute_reply":"2024-03-12T06:16:44.684595Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class Custommodel(nn.Module):\n def __init__(self,config,numclass:int=6):\n super(Custommodel,self).__init__()\n self.model = timm.create_model(\n config.model,\n pretrained = True,\n drop_rate = 0.5\n \n )\n in_features = self.model.fc.out_features\n self.freeze_layers()\n self.fc = nn.Sequential(\n nn.Dropout(p=0.5),\n nn.ReLU(),\n nn.Linear(in_features = in_features,out_features = numclass)\n )\n def freeze_layers(self):\n num_layers_to_freeze = 10\n #the reset has 61 layer out of which I am freezing the top 6 ones\n for i, param in enumerate(self.model.parameters()):\n if i < num_layers_to_freeze:\n param.requires_grad = False\n\n def forward(self,x):\n x = self.model(x)\n x = self.fc(x)\n #x = self.customlayer(x)\n return x\n ","metadata":{"execution":{"iopub.status.busy":"2024-03-12T07:39:58.708181Z","iopub.execute_input":"2024-03-12T07:39:58.708657Z","iopub.status.idle":"2024-03-12T07:39:58.717774Z","shell.execute_reply.started":"2024-03-12T07:39:58.708602Z","shell.execute_reply":"2024-03-12T07:39:58.716677Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model = Custommodel(config)","metadata":{"execution":{"iopub.status.busy":"2024-03-12T07:39:59.073488Z","iopub.execute_input":"2024-03-12T07:39:59.074450Z","iopub.status.idle":"2024-03-12T07:40:00.744022Z","shell.execute_reply.started":"2024-03-12T07:39:59.074399Z","shell.execute_reply":"2024-03-12T07:40:00.742948Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"**The resne model has 61 layers and in order to prevent the overfitting i am freezing the top 6 layer by making them require grad = False so now their parameters wont be upadated because of gradient descent**","metadata":{}},{"cell_type":"code","source":"for name, param in model.named_parameters():\n print(f'{name}: requires_grad={param.requires_grad}')\n","metadata":{"execution":{"iopub.status.busy":"2024-03-12T07:40:03.359610Z","iopub.execute_input":"2024-03-12T07:40:03.360010Z","iopub.status.idle":"2024-03-12T07:40:03.368526Z","shell.execute_reply.started":"2024-03-12T07:40:03.359977Z","shell.execute_reply":"2024-03-12T07:40:03.367292Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"custom = Custommodel(config)\ncustom(torch.randn(size = (1,3,224,224)))","metadata":{"execution":{"iopub.status.busy":"2024-03-12T07:40:03.580424Z","iopub.execute_input":"2024-03-12T07:40:03.581245Z","iopub.status.idle":"2024-03-12T07:40:04.325874Z","shell.execute_reply.started":"2024-03-12T07:40:03.581211Z","shell.execute_reply":"2024-03-12T07:40:04.324736Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"criterion = torch.nn.KLDivLoss(reduction=\"mean\")\nfrom torch.optim.lr_scheduler import OneCycleLR\n","metadata":{"execution":{"iopub.status.busy":"2024-03-12T07:40:04.327862Z","iopub.execute_input":"2024-03-12T07:40:04.328603Z","iopub.status.idle":"2024-03-12T07:40:04.333647Z","shell.execute_reply.started":"2024-03-12T07:40:04.328565Z","shell.execute_reply":"2024-03-12T07:40:04.332586Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def trainer(model,trainloader,optimizer,criterion,device):\n model.train()\n iterationloss = 0\n counter = 0\n for data in tqdm(trainloader):\n message = data['data'].to(device)\n target = data['target'].to(device).squeeze()\n optimizer.zero_grad()\n out = model(message)\n loss = criterion(F.log_softmax(out, dim=1), target)\n loss.backward()\n optimizer.step()\n iterationloss+=loss.item()*message.shape[0]\n counter+=message.shape[0]\n return iterationloss/counter","metadata":{"execution":{"iopub.status.busy":"2024-03-12T07:40:04.334905Z","iopub.execute_input":"2024-03-12T07:40:04.335368Z","iopub.status.idle":"2024-03-12T07:40:04.349308Z","shell.execute_reply.started":"2024-03-12T07:40:04.335336Z","shell.execute_reply":"2024-03-12T07:40:04.348345Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def tester(model,testloader,criterion,device):\n model.eval()\n iterationloss = 0\n counter = 0\n for data in testloader:\n message = data['data'].to(device)\n target = data['target'].to(device).squeeze()\n with torch.no_grad():\n out = model(message)\n loss = criterion(F.log_softmax(out, dim=1), target)\n iterationloss+=loss*message.shape[0]\n #print(loss) \n counter+=message.shape[0]\n return float(iterationloss)/float(counter)\n","metadata":{"execution":{"iopub.status.busy":"2024-03-12T07:40:04.351336Z","iopub.execute_input":"2024-03-12T07:40:04.351955Z","iopub.status.idle":"2024-03-12T07:40:04.359477Z","shell.execute_reply.started":"2024-03-12T07:40:04.351913Z","shell.execute_reply":"2024-03-12T07:40:04.358587Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def trainepoch(fold,train_loss,val_loss):\n traindf = train_df[train_df[\"fold\"]!=fold].reset_index(drop = True)\n validdf = train_df[train_df[\"fold\"]==fold].reset_index(drop = True)\n \n traindataset = CustomDataset(traindf,config,mode = \"train\")\n validdataset = CustomDataset(validdf,config,mode = \"train\")\n \n trainloader = DataLoader(traindataset,batch_size = config.batchsize)\n validloader = DataLoader(validdataset,batch_size = config.batchsize)\n \n model = Custommodel(config)\n model.to(device)\n \n optimizer = torch.optim.AdamW(model.parameters(), lr=1e-3, weight_decay=config.WEIGHT_DECAY)\n \n scheduler = OneCycleLR(\n optimizer,\n max_lr=1e-3,\n epochs=config.epoch,\n steps_per_epoch=len(trainloader),\n pct_start=0.1,\n anneal_strategy=\"cos\",\n final_div_factor=100,\n )\n \n criterion = torch.nn.KLDivLoss(reduction=\"batchmean\")\n bestloss = np.inf\n \n bestloss = np.inf\n for i in (range(10)):\n print(\"-\"*120)\n print(\"Began iteration no.\",i+1)\n print(\":\"*50,\"=\"*20,\":\"*50)\n trainloss = trainer(model,trainloader,optimizer,criterion,device)\n testloss = tester(model,validloader,criterion,device)\n train_loss.append(trainloss)\n val_loss.append(testloss)\n print(\"train loss = \",trainloss)\n print(\"val loss=\",testloss)\n print(\"=\"*80,\"\\n\")\n if testloss