We can see some graphs shows correlation, so let's check correlation next
","metadata":{}},{"cell_type":"markdown","source":"### Feature Correlation HeatMap","metadata":{}},{"cell_type":"code","source":"# correlation matrix for single eeg offset file\nplt.figure(figsize=(10,6))\ndf_corr = eeg_offset_file.corr(method='pearson')\nsns.heatmap(df_corr[:10],annot=True , cmap=\"BuPu\")\nplt.tight_layout()\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2024-01-28T17:50:44.073769Z","iopub.execute_input":"2024-01-28T17:50:44.074126Z","iopub.status.idle":"2024-01-28T17:50:45.156130Z","shell.execute_reply.started":"2024-01-28T17:50:44.074096Z","shell.execute_reply":"2024-01-28T17:50:45.155264Z"},"_kg_hide-input":true,"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Highly correlated feature pairs :\n","metadata":{}},{"cell_type":"code","source":"# strong correlated features pairs\nthershold = 0.6\n\nhighly_correlated_pairs = (df_corr.abs() > thershold) & (df_corr.abs() < 1)\ncorrelated_feature_pairs = set() \nfor i in range(len(eeg_offset_file.columns)):\n for j in range(i + 1, len(eeg_offset_file.columns)):\n if highly_correlated_pairs.iloc[i, j]:\n correlated_feature_pairs.add((eeg_offset_file.columns[i], eeg_offset_file.columns[j]))\n\n\nx = PrettyTable()\nfor feature_pair in correlated_feature_pairs:\n x.add_row(feature_pair)\nx.field_names = ['Feature 1', 'Feature 2']\nprint(x)","metadata":{"execution":{"iopub.status.busy":"2024-01-28T17:50:45.157261Z","iopub.execute_input":"2024-01-28T17:50:45.157548Z","iopub.status.idle":"2024-01-28T17:50:45.174293Z","shell.execute_reply.started":"2024-01-28T17:50:45.157523Z","shell.execute_reply":"2024-01-28T17:50:45.173189Z"},"_kg_hide-input":true,"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"del eeg_file_single, eeg_offset_file","metadata":{"execution":{"iopub.status.busy":"2024-01-28T17:50:45.175413Z","iopub.execute_input":"2024-01-28T17:50:45.175694Z","iopub.status.idle":"2024-01-28T17:50:45.188236Z","shell.execute_reply.started":"2024-01-28T17:50:45.175670Z","shell.execute_reply":"2024-01-28T17:50:45.187355Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# | Load All EEG Files","metadata":{}},{"cell_type":"code","source":"# reading all parquet file takes around 15 minutes. to seva time we \n# loaded all files and calculated features mean, max & min, saved these\n# features in a csv file. we can directly load this file.\n# change READ_EEG_PARQUET = True to read all data\n\neeg_columns = ['Fp1', 'F3', 'C3', 'P3', 'F7', 'T3', 'T5', 'O1', 'Fz',\n 'Cz', 'Pz','Fp2', 'F4', 'C4', 'P4', 'F8', 'T4', 'T6', 'O2', 'EKG']\n \nif READ_EEG_PARQUET:\n path = \"/kaggle/input/hms-harmful-brain-activity-classification/train_eegs/\"\n files = os.listdir(path)\n eeg = {}\n for i, f in enumerate(files):\n eeg_single_f = pd.read_parquet(f\"{path}{f}\") \n file_name = int(f.split(\".\")[0])\n eeg[file_name] = eeg_single_f.values\n \n for c in eeg_columns:\n train[c+\"_mean\"] = None\n train[c+\"_max\"] = None\n train[c+\"_min\"] = None\n for i, eeg_id in enumerate(train['eeg_id']):\n eeg_offset_second = train.loc[i,\"eeg_label_offset_seconds\"]\n for j,col in enumerate(eeg_columns):\n temp = eeg[eeg_id][int(eeg_offset_second*200):int(eeg_offset_second+50)*200,j]\n train.at[i,col+\"_mean\"] = np.nanmean(temp)\n train.at[i,col+\"_max\"] = np.nanmax(temp)\n train.at[i,col+\"_min\"] = np.nanmin(temp)\n\nelse:\n train_eeg = pd.read_csv(\"/kaggle/input/preprocessed-data/train_eegs.csv\")","metadata":{"execution":{"iopub.status.busy":"2024-01-28T17:50:45.189362Z","iopub.execute_input":"2024-01-28T17:50:45.189680Z","iopub.status.idle":"2024-01-28T17:50:46.915946Z","shell.execute_reply.started":"2024-01-28T17:50:45.189648Z","shell.execute_reply":"2024-01-28T17:50:46.915133Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_eeg.head()","metadata":{"execution":{"iopub.status.busy":"2024-01-28T17:50:46.916997Z","iopub.execute_input":"2024-01-28T17:50:46.917284Z","iopub.status.idle":"2024-01-28T17:50:46.941003Z","shell.execute_reply.started":"2024-01-28T17:50:46.917259Z","shell.execute_reply":"2024-01-28T17:50:46.940022Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Load Test EEG Files\n\npath = \"/kaggle/input/hms-harmful-brain-activity-classification/test_eegs/\"\nfiles = os.listdir(path)\neeg = {}\nfor i, f in enumerate(files):\n eeg_single_f = pd.read_parquet(f\"{path}{f}\") \n file_name = int(f.split(\".\")[0])\n eeg[file_name] = eeg_single_f.values\n\nfor c in eeg_columns:\n test[c+\"_mean\"] = None\n test[c+\"_max\"] = None\n test[c+\"_min\"] = None\nfor i, eeg_id in enumerate(test['eeg_id']):\n for j,col in enumerate(eeg_columns):\n temp = eeg[eeg_id]\n test.at[i,col+\"_mean\"] = np.nanmean(temp)\n test.at[i,col+\"_max\"] = np.nanmax(temp)\n test.at[i,col+\"_min\"] = np.nanmin(temp)\n","metadata":{"execution":{"iopub.status.busy":"2024-01-28T17:56:06.296600Z","iopub.execute_input":"2024-01-28T17:56:06.296966Z","iopub.status.idle":"2024-01-28T17:56:06.329326Z","shell.execute_reply.started":"2024-01-28T17:56:06.296936Z","shell.execute_reply":"2024-01-28T17:56:06.328534Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test","metadata":{"execution":{"iopub.status.busy":"2024-01-28T17:56:22.755611Z","iopub.execute_input":"2024-01-28T17:56:22.756011Z","iopub.status.idle":"2024-01-28T17:56:22.777120Z","shell.execute_reply.started":"2024-01-28T17:56:22.755981Z","shell.execute_reply":"2024-01-28T17:56:22.776142Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# plot target variable\n\ntarget_counts = train[TARGET].value_counts()\nlabels= ['seizure','lpd','gpd','lrda','grda','other']\n\nfig, axs = plt.subplots(1,2,figsize=(10,4))\n\naxs[0].pie(\n train[TARGET].value_counts(),\n labels=labels,\n shadow=True, \n explode=[.1 for i in range(train[TARGET].nunique())], # Adjust the explode based on the number of unique values\n autopct='%1.f%%',\n)\nplt.suptitle('Target label in Train Dataset', fontsize=13, fontweight='bold')\n\nsns.countplot(train,x = TARGET,ax=axs[1])\nplt.tight_layout()\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2024-01-28T17:50:46.942121Z","iopub.execute_input":"2024-01-28T17:50:46.942466Z","iopub.status.idle":"2024-01-28T17:50:47.430290Z","shell.execute_reply.started":"2024-01-28T17:50:46.942434Z","shell.execute_reply":"2024-01-28T17:50:47.429361Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# | Model Training","metadata":{}},{"cell_type":"code","source":"FEATURES = [ 'Fp1_mean','Fp1_max', 'Fp1_min', 'F3_mean', 'F3_max', 'F3_min', 'C3_mean',\n 'C3_max', 'C3_min', 'P3_mean', 'P3_max', 'P3_min', 'F7_mean', 'F7_max',\n 'F7_min', 'T3_mean', 'T3_max', 'T3_min', 'T5_mean', 'T5_max', 'T5_min',\n 'O1_mean', 'O1_max', 'O1_min', 'Fz_mean', 'Fz_max', 'Fz_min', 'Cz_mean',\n 'Cz_max', 'Cz_min', 'Pz_mean', 'Pz_max', 'Pz_min', 'Fp2_mean',\n 'Fp2_max', 'Fp2_min', 'F4_mean', 'F4_max', 'F4_min', 'C4_mean',\n 'C4_max', 'C4_min', 'P4_mean', 'P4_max', 'P4_min', 'F8_mean', 'F8_max',\n 'F8_min', 'T4_mean', 'T4_max', 'T4_min', 'T6_mean', 'T6_max', 'T6_min',\n 'O2_mean', 'O2_max', 'O2_min', 'EKG_mean', 'EKG_max', 'EKG_min']\n\nX = train_eeg[FEATURES]\ny = train_eeg[TARGET]\nX_test = test[FEATURES]","metadata":{"execution":{"iopub.status.busy":"2024-01-28T18:08:27.146400Z","iopub.execute_input":"2024-01-28T18:08:27.147079Z","iopub.status.idle":"2024-01-28T18:08:27.172873Z","shell.execute_reply.started":"2024-01-28T18:08:27.147048Z","shell.execute_reply":"2024-01-28T18:08:27.172037Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Normalize data\nscaler = StandardScaler()\nX[FEATURES] = scaler.fit_transform(X[FEATURES])\nX_test[FEATURES] = scaler.transform(X_test[FEATURES])","metadata":{"execution":{"iopub.status.busy":"2024-01-28T18:08:27.409424Z","iopub.execute_input":"2024-01-28T18:08:27.410189Z","iopub.status.idle":"2024-01-28T18:08:27.557169Z","shell.execute_reply.started":"2024-01-28T18:08:27.410155Z","shell.execute_reply":"2024-01-28T18:08:27.556339Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# train model\n\nskf = StratifiedKFold(n_splits = 5, shuffle=True, random_state=SEED)\nfor fold, (train_idx,val_idx) in enumerate(skf.split(X,y)):\n X_train, y_train = X.iloc[train_idx], y.iloc[train_idx]\n X_val, y_val = X.iloc[val_idx], y.iloc[val_idx]\n \n train_pool = Pool(X_train,y_train)\n val_pool = Pool(X_val,y_val)\n\n model = cat.CatBoostClassifier(\n task_type = \"GPU\",\n loss_function = \"MultiClass\"\n )\n \n model.fit(train_pool, verbose=300, eval_set=val_pool)\n#","metadata":{"execution":{"iopub.status.busy":"2024-01-28T18:05:45.748321Z","iopub.execute_input":"2024-01-28T18:05:45.748710Z","iopub.status.idle":"2024-01-28T18:06:25.818580Z","shell.execute_reply.started":"2024-01-28T18:05:45.748677Z","shell.execute_reply":"2024-01-28T18:06:25.817562Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# train on complete data\n\nmodel = cat.CatBoostClassifier(\n task_type = \"GPU\",\n loss_function = \"MultiClass\"\n )\n \nmodel.fit(X,y, verbose = 300)\npred = model.predict_proba(test[FEATURES])","metadata":{"execution":{"iopub.status.busy":"2024-01-28T17:56:33.651563Z","iopub.execute_input":"2024-01-28T17:56:33.652202Z","iopub.status.idle":"2024-01-28T17:56:41.291705Z","shell.execute_reply.started":"2024-01-28T17:56:33.652170Z","shell.execute_reply":"2024-01-28T17:56:41.290659Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# | Final Submission\n","metadata":{}},{"cell_type":"code","source":"sub = sample_sub.copy()\nsub[TARGET_COLUMNS] = pred\nsub.to_csv(\"submission.csv\",index=False)\nsub","metadata":{"execution":{"iopub.status.busy":"2024-01-28T17:57:38.175542Z","iopub.execute_input":"2024-01-28T17:57:38.176491Z","iopub.status.idle":"2024-01-28T17:57:38.192694Z","shell.execute_reply.started":"2024-01-28T17:57:38.176450Z","shell.execute_reply":"2024-01-28T17:57:38.191818Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"if you see any typos please let me know. Upvote if you like. Thanks, Happy Coding !!","metadata":{}}]}