{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.11.13","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"none","dataSources":[{"sourceId":41880,"databundleVersionId":5677426,"sourceType":"competition"},{"sourceId":122144951,"sourceType":"kernelVersion"}],"dockerImageVersionId":31089,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"import numpy as np\nimport pandas as pd\nfrom sklearn.preprocessing import StandardScaler\nimport glob\nfrom torch import nn\nfrom sklearn import *\nimport gc\nimport pickle as p\nimport torch.functional as F\nfrom torch.utils.data import Dataset, DataLoader, TensorDataset\nimport torch\nimport torch.optim as optim\nfrom sklearn.metrics import classification_report","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"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\nfor dirname, _, filenames in os.walk('/kaggle/input'):\n for filename in filenames:\n print(os.path.join(dirname, filename))\n break\nprint(\"Required Files obtained\")\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},"outputs":[],"execution_count":null},{"cell_type":"code","source":"def get_data(f):\n df = pd.read_csv(f)\n df['Id'] = f.split('/')[-1].split('.')[0]\n df['data_type'] = f.split('/')[-2]\n return df\ndef prepare_data():\n #-----------------------------------\n path=\"/kaggle/input/tlvmc-parkinsons-freezing-gait-prediction/\"\n train_defog = glob.glob(path+'train/defog/**')\n train_tdcsfog = glob.glob(path+'train/tdcsfog/**')\n #-----------------------------------\n df_train_defog = pd.concat([get_data(f) for f in train_defog])\n df_train_tdcsfog = pd.concat([get_data(f) for f in train_tdcsfog])\n #-----------------------------------\n df_train=pd.concat([df_train_defog,df_train_tdcsfog])\n #-----------------------------------\n print(df_train_defog.shape)\n print(df_train_tdcsfog.shape)\n \n print(df_train.shape)\n #-----------------------------------\n df_train.fillna(0,inplace=True)\n print(df_train.isnull().sum().sum())\n return df_train, df_train_defog, df_train_tdcsfog\ndef present_basic_info(*args):\n print(df_train[df_train['data_type'] == 'defog'].head())\n print(df_train[df_train['data_type'] == 'tdcsfog'].head())\n print(df_train_defog.head())\n print(df_train_tdcsfog.head())\n df_train.head()\n\ndef create_datasets(df_train):\n X_train, X_valid, y_train, y_valid = model_selection.train_test_split(df_train[features], df_train[Targets], test_size=.30, random_state=42)\n print(f\"X_train : {X_train.shape}; X_valid : {X_valid.shape}\")\n print(f\"y_train : {y_train.shape}; y_valid : {y_valid.shape}\")\n return X_train, X_valid, y_train, y_valid\ndef give_Valid_Values(df_train_defog):\n return df_train_defog[df_train_defog['Valid'] != False]","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"df_train, df_train_defog, df_train_tdcsfog = prepare_data()\npresent_basic_info(df_train, df_train_defog, df_train_tdcsfog)\nfeatures=['Time', 'AccV', 'AccML', 'AccAP']\nTargets=['StartHesitation', 'Turn' , 'Walking']","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"\nX_train, X_vald, y_train, y_valid = create_datasets(df_train)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-06T09:13:53.928181Z","iopub.execute_input":"2025-11-06T09:13:53.928912Z","iopub.status.idle":"2025-11-06T09:13:59.150653Z","shell.execute_reply.started":"2025-11-06T09:13:53.928881Z","shell.execute_reply":"2025-11-06T09:13:59.149138Z"}},"outputs":[{"name":"stdout","text":"X_train : (14411861, 4); X_valid : (6176513, 4)\ny_train : (14411861, 3); y_valid : (6176513, 3)\n","output_type":"stream"}],"execution_count":9},{"cell_type":"markdown","source":"

Random Forest Model(as used in the reference notebook)

","metadata":{}},{"cell_type":"code","source":"model_Reg = ensemble.RandomForestRegressor(n_estimators=100, max_depth=7, n_jobs=-1, random_state=42)\nmodel_Reg.fit(X_train, y_train)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-06T09:14:53.212781Z","iopub.execute_input":"2025-11-06T09:14:53.214925Z","iopub.status.idle":"2025-11-06T09:52:47.155227Z","shell.execute_reply.started":"2025-11-06T09:14:53.214871Z","shell.execute_reply":"2025-11-06T09:52:47.15277Z"}},"outputs":[{"execution_count":12,"output_type":"execute_result","data":{"text/plain":"RandomForestRegressor(max_depth=7, n_jobs=-1, random_state=42)","text/html":"
RandomForestRegressor(max_depth=7, n_jobs=-1, random_state=42)
In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
"},"metadata":{}}],"execution_count":12},{"cell_type":"code","source":"print(metrics.average_precision_score(y_valid, model_Reg.predict(X_valid).clip(0.0,1.0)))","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-06T09:52:47.159964Z","iopub.execute_input":"2025-11-06T09:52:47.160457Z","iopub.status.idle":"2025-11-06T09:52:47.251118Z","shell.execute_reply.started":"2025-11-06T09:52:47.160423Z","shell.execute_reply":"2025-11-06T09:52:47.249093Z"}},"outputs":[{"traceback":["\u001b[0;31m---------------------------------------------------------------------------\u001b[0m","\u001b[0;31mNameError\u001b[0m Traceback (most recent call last)","\u001b[0;32m/tmp/ipykernel_36/746985419.py\u001b[0m in \u001b[0;36m\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mprint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmetrics\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0maverage_precision_score\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0my_valid\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmodel_Reg\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpredict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mX_valid\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mclip\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m0.0\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m1.0\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m","\u001b[0;31mNameError\u001b[0m: name 'X_valid' is not defined"],"ename":"NameError","evalue":"name 'X_valid' is not defined","output_type":"error"}],"execution_count":13},{"cell_type":"code","source":"model_Reg.score(X_valid, y_valid)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-06T09:52:47.251755Z","iopub.status.idle":"2025-11-06T09:52:47.252179Z","shell.execute_reply.started":"2025-11-06T09:52:47.251926Z","shell.execute_reply":"2025-11-06T09:52:47.251939Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model_Reg.score(X_train, y_train)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-06T09:52:47.253289Z","iopub.status.idle":"2025-11-06T09:52:47.253693Z","shell.execute_reply.started":"2025-11-06T09:52:47.253554Z","shell.execute_reply":"2025-11-06T09:52:47.253568Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"

Dropping all false values

","metadata":{}},{"cell_type":"code","source":"df_train_AllValid = give_Valid_Values(df_train_defog)\nfeatures=['Time', 'AccV', 'AccML', 'AccAP']\nTargets=['StartHesitation', 'Turn' , 'Walking']","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-06T09:52:47.254385Z","iopub.status.idle":"2025-11-06T09:52:47.254769Z","shell.execute_reply.started":"2025-11-06T09:52:47.254586Z","shell.execute_reply":"2025-11-06T09:52:47.254605Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"x_train, x_test, y_train, y_test = model_selection.train_test_split(df_train_AllValid[features], df_train_AllValid[Targets], test_size=.30, random_state=42)\nprint(f\"X_train : {x_train.shape}; X_valid : {x_test.shape}\")\nprint(f\"y_train : {y_train.shape}; y_valid : {y_test.shape}\")","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-06T09:52:47.255904Z","iopub.status.idle":"2025-11-06T09:52:47.256291Z","shell.execute_reply.started":"2025-11-06T09:52:47.256116Z","shell.execute_reply":"2025-11-06T09:52:47.256133Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"x_train, x_test, y_train, y_test = model_selection.train_test_split(df_train_AllValid[features], df_train_AllValid[Targets], test_size=.30, random_state=42)\nprint(f\"X_train : {x_train.shape}; X_valid : {x_test.shape}\")\nprint(f\"y_train : {y_train.shape}; y_valid : {y_test.shape}\")\nmodel_Reg = ensemble.RandomForestRegressor(n_estimators=100, max_depth=7, random_state=42)\nmodel_Reg.fit(x_train, y_train)\nprint(f\"Precision : {metrics.average_precision_score(y_test, model_Reg.predict(x_test).clip(0.0,1.0))}\")\nprint(f\"Training Accuracy : {model_Reg.score(x_train, y_train)}\")\nprint(f\"Testing Accuracy : {model_Reg.score(x_test, y_test)}\")","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-06T09:52:47.258078Z","iopub.status.idle":"2025-11-06T09:52:47.258546Z","shell.execute_reply.started":"2025-11-06T09:52:47.258298Z","shell.execute_reply":"2025-11-06T09:52:47.258318Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"

Using LSTM

","metadata":{}},{"cell_type":"code","source":"","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"class Model_LSTM(nn.Module):\n def __init__(self, input_dim = 4, hidden_dim = 64, layer_dim = 2):\n super(Model_LSTM, self).__init__()\n self.input_dim = input_dim\n self.hidden_dim = hidden_dim\n self.layer_dim = layer_dim\n self.lstm = nn.LSTM(\n input_dim,\n hidden_dim,\n layer_dim,\n batch_first = True\n )\n self.fc = nn.Linear(hidden_dim, 4)\n def forward(self, x):\n x = x.unsqueeze(1)\n out, _ = self.lstm(x)\n out = out[:,-1,:]\n \n out = self.fc(out)\n return out\n#FocalLoss for imbalanced Dataset\nclass FocalLoss(nn.Module):\n def __init__(self, alpha = None, gamma = 2, reduction = 'mean'):\n super(FocalLoss, self).__init__()\n self.alpha = alpha\n self.gamma = gamma\n self.reduction = reduction\n\n def forward(self, inputs, targets):\n ce_loss = nn.functional.cross_entropy(inputs, targets, weight = self.alpha, reduction='none') \n # Compute pt (model confidence on true class)\n pt = torch.exp(-ce_loss)\n \n # Apply the focal adjustment\n focal_loss = (1 - pt) ** self.gamma * ce_loss\n\n # Apply reduction (mean, sum, or no reduction)\n if self.reduction == 'mean':\n return focal_loss.mean()\n elif self.reduction == 'sum':\n return focal_loss.sum()\n else:\n return focal_loss\n \n \ndef transform_target_labels(y_train, y_test):\n vector = {0:0, 1:1, 2:2, 4:3}\n y_train = y_train[\"StartHesitation\"] * 4 + y_train[\"Turn\"] * 2 + y_train[\"Walking\"]\n y_test = y_test[\"StartHesitation\"] * 4 + y_test[\"Turn\"] * 2 + y_test[\"Walking\"]\n y_train = np.vectorize(vector.get)(y_train)\n y_test = np.vectorize(vector.get)(y_test)\n y_train = torch.tensor(y_train, dtype=torch.long) # must be long for CrossEntropyLoss\n \n y_test = torch.tensor(y_test, dtype=torch.long)\n return y_train, y_test\n\ndef createDataLoaders(x_train, x_test, y_train, y_test):\n x_train = torch.tensor(x_train, dtype=torch.float32)\n x_test = torch.tensor(x_test, dtype=torch.float32)\n y_train, y_test = transform_target_labels(y_train, y_test)\n \n trainData = TensorDataset(x_train, y_train)\n testData = TensorDataset(x_test, y_test)\n\n trainLoader = DataLoader(trainData, batch_size=64, shuffle=True)\n testLoader = DataLoader(testData, batch_size=64, shuffle=True)\n return trainLoader, testLoader\n\n\nfrom sklearn.metrics import classification_report\nimport torch.nn as nn\nimport torch.optim as optim\n\ndef train_model(model, TrainingSet, TestingSet, epoch=10, lr=0.01, device=\"cuda\"):\n \n class_counts = torch.tensor([2383254, 68987, 410776, 354], dtype=torch.float32)\n class_weights = 1.0 / torch.sqrt(class_counts)\n class_weights = class_weights / class_weights.sum() * len(class_counts)\n loss_fn = FocalLoss(alpha=class_weights, gamma=1.5)\n optimizer = optim.Adam(model.parameters(), lr)\n\n for i in range(epoch):\n model.train()\n total_loss = 0\n all_preds, all_labels = [], []\n\n # Training\n for batch, label in TrainingSet:\n logits = model(batch) # [batch, 4]\n loss = loss_fn(logits, label)\n\n optimizer.zero_grad()\n loss.backward()\n optimizer.step()\n\n total_loss += loss.item()\n \n preds = torch.argmax(logits, dim=1)\n all_preds.extend(preds.cpu().numpy())\n all_labels.extend(label.cpu().numpy())\n\n print(\"-----------------------------\")\n print(f\"Epoch : {i+1}/{epoch}\")\n print(f\"Train Loss : {total_loss/len(TrainingSet):.4f}\")\n print(\"Train Classification Report:\")\n print(classification_report(all_labels, all_preds, target_names=[\"0\",\"1\",\"2\",\"4\"], zero_division=0))\n\n # Evaluation\n model.eval()\n test_preds, test_labels = [], []\n with torch.no_grad():\n for batch, label in TestingSet:\n logits = model(batch)\n\n preds = torch.argmax(logits, dim=1)\n test_preds.extend(preds.cpu().numpy())\n test_labels.extend(label.cpu().numpy())\n\n print(\"Test Classification Report:\")\n print(classification_report(test_labels, test_preds, target_names=[\"0\",\"1\",\"2\",\"4\"]))\n","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-06T09:52:47.259804Z","iopub.status.idle":"2025-11-06T09:52:47.260135Z","shell.execute_reply.started":"2025-11-06T09:52:47.259963Z","shell.execute_reply":"2025-11-06T09:52:47.259975Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"y_train, y_test = transform_target_labels(y_train, y_test)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-06T09:52:47.262092Z","iopub.status.idle":"2025-11-06T09:52:47.262501Z","shell.execute_reply.started":"2025-11-06T09:52:47.262324Z","shell.execute_reply":"2025-11-06T09:52:47.262344Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import matplotlib.pyplot as plt\nimport numpy as np\n\n# Count frequencies for train and test\nlabels, counts_train = np.unique(y_train, return_counts=True)\n_, counts_test = np.unique(y_test, return_counts=True)\n\nx = np.arange(len(labels)) # label positions\nwidth = 0.35 # bar width\n\nplt.bar(x - width/2, counts_train, width, label='Train')\nplt.bar(x + width/2, counts_test, width, label='Test')\n\nplt.xticks(x, labels)\nplt.xlabel(\"Class Labels\")\nplt.ylabel(\"Count\")\nplt.title(\"Class Distribution (Train vs Test)\")\nplt.legend()\nplt.show()\n","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-06T09:13:59.213874Z","iopub.status.idle":"2025-11-06T09:13:59.214236Z","shell.execute_reply.started":"2025-11-06T09:13:59.214101Z","shell.execute_reply":"2025-11-06T09:13:59.214115Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model = Model_LSTM()\ntrainData, testData = createDataLoaders(\n x_train.to_numpy(),\n x_test.to_numpy(),\n y_train,\n y_test\n)\ntrain_accuracy, test_accuracy, train_loss = train_model(\n model,\n trainData,\n testData,\n epoch = 1\n)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-06T09:13:59.215233Z","iopub.status.idle":"2025-11-06T09:13:59.215526Z","shell.execute_reply.started":"2025-11-06T09:13:59.215397Z","shell.execute_reply":"2025-11-06T09:13:59.21541Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"total_true = np.zeros((0))\ntotal_pred = np.zeros((0))\nidx = 0\nfor batch, label in trainData:\n y_pred = model(batch)\n mapped = torch.tensor([0, 1, 2, 4])\n t_pred = torch.argmax(y_pred, dim=1)\n print(t_pred)\n #print(f\"Accuracy : {corrc/count}\")\n break","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-06T09:13:59.217487Z","iopub.status.idle":"2025-11-06T09:13:59.218087Z","shell.execute_reply.started":"2025-11-06T09:13:59.217696Z","shell.execute_reply":"2025-11-06T09:13:59.217713Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"y_train[\"FinalLabel\"] = y_train[\"StartHesitation\"] * 4 + y_train[\"Turn\"] * 2 + y_train[\"Walking\"]\n","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-06T09:13:59.218844Z","iopub.status.idle":"2025-11-06T09:13:59.219175Z","shell.execute_reply.started":"2025-11-06T09:13:59.218993Z","shell.execute_reply":"2025-11-06T09:13:59.219005Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"y_train.groupby(\"Label Count\").agg(\n count = pd.NamedAgg(column = \"Turn\", aggfunc = \"count\")\n)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-06T09:13:59.220168Z","iopub.status.idle":"2025-11-06T09:13:59.220482Z","shell.execute_reply.started":"2025-11-06T09:13:59.220301Z","shell.execute_reply":"2025-11-06T09:13:59.220311Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"29531 + 176053 + 146","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-11-06T09:13:59.222174Z","iopub.status.idle":"2025-11-06T09:13:59.223525Z","shell.execute_reply.started":"2025-11-06T09:13:59.22334Z","shell.execute_reply":"2025-11-06T09:13:59.223365Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"

Further Simplification of the Dataset

\n0 in this class label represents cases where no FOG event takes place\nThe rest are events where some FOG event do in fact take place\nBy combining the data into a simple 0 and 1 binary, the class imbalance can be potentiall dealt with. It must still be kept in mind that the 0s would still far outweight the 1s","metadata":{}},{"cell_type":"code","source":"","metadata":{"trusted":true},"outputs":[],"execution_count":null}]}