{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"## Thanks to great works from \nJEROENVDD: https://www.kaggle.com/code/jeroenvdd/time-series-tsflex\n\n&\n\n此般浅薄: https://www.kaggle.com/code/xzj19013742/groupkfold-cross-validation-tsflex\n\n## Improvements\n- Use LightGBM instead of ExtraTrees regression\n- Simple hyperparameter search (RandomizedSearch with only 3 rounds for each target and a random dataset sample of 10%)\n- Define average_precision_score as eval-metric within the lgb-regressor\n- Adjust Multioutput-regressor class from sk-learn to add additional parameters for lgb\n\n\nHope this notebook supports further improvements!","metadata":{}},{"cell_type":"code","source":"# Install tsflex and seglearn\n!pip install tsflex --no-index --find-links=file:///kaggle/input/time-series-tools\n!pip install seglearn --no-index --find-links=file:///kaggle/input/time-series-tools","metadata":{"execution":{"iopub.status.busy":"2023-03-29T12:57:06.1353Z","iopub.execute_input":"2023-03-29T12:57:06.13604Z","iopub.status.idle":"2023-03-29T12:57:31.569856Z","shell.execute_reply.started":"2023-03-29T12:57:06.136001Z","shell.execute_reply":"2023-03-29T12:57:31.568339Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import numpy as np\nimport pandas as pd\nfrom tqdm.auto import tqdm\nfrom sklearn import *\nimport glob\n\np = '/kaggle/input/tlvmc-parkinsons-freezing-gait-prediction/'\n\ntrain = glob.glob(p+'train/**/**')\ntest = glob.glob(p+'test/**/**')\nsubjects = pd.read_csv(p+'subjects.csv')\ntasks = pd.read_csv(p+'tasks.csv')\nsub = pd.read_csv(p+'sample_submission.csv')\n\ntdcsfog_metadata=pd.read_csv('/kaggle/input/tlvmc-parkinsons-freezing-gait-prediction/tdcsfog_metadata.csv')\ndefog_metadata=pd.read_csv('/kaggle/input/tlvmc-parkinsons-freezing-gait-prediction/defog_metadata.csv')\n# daily_metadata=pd.read_csv('/kaggle/input/tlvmc-parkinsons-freezing-gait-prediction/daily_metadata.csv')\ntdcsfog_metadata['Module']='tdcsfog'\ndefog_metadata['Module']='defog'\n# daily_metadata['Module']='daily'\nmetadata=pd.concat([tdcsfog_metadata,defog_metadata])\nmetadata","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2023-03-29T12:57:31.573871Z","iopub.execute_input":"2023-03-29T12:57:31.574263Z","iopub.status.idle":"2023-03-29T12:57:33.363907Z","shell.execute_reply.started":"2023-03-29T12:57:31.574228Z","shell.execute_reply":"2023-03-29T12:57:33.362751Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# https://www.kaggle.com/code/jazivxt/familiar-solvs\ntasks['Duration'] = tasks['End'] - tasks['Begin']\ntasks = pd.pivot_table(tasks, values=['Duration'], index=['Id'], columns=['Task'], aggfunc='sum', fill_value=0)\ntasks.columns = [c[-1] for c in tasks.columns]\ntasks = tasks.reset_index()\ntasks['t_kmeans'] = cluster.KMeans(n_clusters=10, random_state=3).fit_predict(tasks[tasks.columns[1:]])\n\nsubjects = subjects.fillna(0).groupby('Subject').median()\nsubjects = subjects.reset_index()\n# subjects.rename(columns={'Subject':'Id'}, inplace=True)\nsubjects['s_kmeans'] = cluster.KMeans(n_clusters=10, random_state=3).fit_predict(subjects[subjects.columns[1:]])\nsubjects=subjects.rename(columns={'Visit':'s_Visit','Age':'s_Age','YearsSinceDx':'s_YearsSinceDx','UPDRSIII_On':'s_UPDRSIII_On','UPDRSIII_Off':'s_UPDRSIII_Off','NFOGQ':'s_NFOGQ'})\n\ndisplay(tasks)\ndisplay(subjects)","metadata":{"execution":{"iopub.status.busy":"2023-03-29T12:57:33.373239Z","iopub.execute_input":"2023-03-29T12:57:33.373559Z","iopub.status.idle":"2023-03-29T12:57:33.565049Z","shell.execute_reply.started":"2023-03-29T12:57:33.373528Z","shell.execute_reply":"2023-03-29T12:57:33.563668Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"complex_featlist=['Visit','Test','Medication','s_Visit','s_Age','s_YearsSinceDx','s_UPDRSIII_On','s_UPDRSIII_Off','s_NFOGQ','s_kmeans']\nmetadata_complex=metadata.merge(subjects,how='left',on='Subject').copy()\nmetadata_complex['Medication']=metadata_complex['Medication'].factorize()[0]\n\ndisplay(metadata_complex)","metadata":{"execution":{"iopub.status.busy":"2023-03-29T12:57:33.567046Z","iopub.execute_input":"2023-03-29T12:57:33.567825Z","iopub.status.idle":"2023-03-29T12:57:33.611078Z","shell.execute_reply.started":"2023-03-29T12:57:33.567772Z","shell.execute_reply":"2023-03-29T12:57:33.609945Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Create a tsflex feature collection","metadata":{}},{"cell_type":"code","source":"from seglearn.feature_functions import base_features, emg_features\n\nfrom tsflex.features import FeatureCollection, MultipleFeatureDescriptors\nfrom tsflex.features.integrations import seglearn_feature_dict_wrapper\n\n\nbasic_feats = MultipleFeatureDescriptors(\n    functions=seglearn_feature_dict_wrapper(base_features()),\n    series_names=['AccV', 'AccML', 'AccAP'],\n    windows=[5_000],\n    strides=[5_000],\n)\n\nemg_feats = emg_features()\ndel emg_feats['simple square integral'] # is same as abs_energy (which is in base_features)\n\nemg_feats = MultipleFeatureDescriptors(\n    functions=seglearn_feature_dict_wrapper(emg_feats),\n    series_names=['AccV', 'AccML', 'AccAP'],\n    windows=[5_000],\n    strides=[5_000],\n)\n\nfc = FeatureCollection([basic_feats, emg_feats])","metadata":{"execution":{"iopub.status.busy":"2023-03-29T12:57:33.61258Z","iopub.execute_input":"2023-03-29T12:57:33.612938Z","iopub.status.idle":"2023-03-29T12:57:33.677206Z","shell.execute_reply.started":"2023-03-29T12:57:33.612874Z","shell.execute_reply":"2023-03-29T12:57:33.675656Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Extract the features","metadata":{}},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import pathlib\ndef reader(f):\n    try:\n        df = pd.read_csv(f, index_col=\"Time\", usecols=['Time', 'AccV', 'AccML', 'AccAP', 'StartHesitation', 'Turn' , 'Walking'])\n        \n        #df[\"mean_acc_v\"] = df[\"AccV\"].mean()\n        #df[\"mean_acc_ml\"] = df[\"AccML\"].mean()\n        #df[\"mean_acc_ap\"] = df[\"AccAP\"].mean()\n        #df[\"min_acc_v\"] = df[\"AccV\"].mean()\n        #df[\"min_acc_ml\"] = df[\"AccML\"].mean()\n        #df[\"min_acc_ap\"] = df[\"AccAP\"].mean()\n        #df[\"max_acc_v\"] = df[\"AccV\"].mean()\n        #df[\"max_acc_ml\"] = df[\"AccML\"].mean()\n        #df[\"max_acc_ap\"] = df[\"AccAP\"].mean()\n        #df[\"median_acc_v\"] = df[\"AccV\"].mean()\n        #df[\"median_acc_ml\"] = df[\"AccML\"].mean()\n        #df[\"median_acc_ap\"] = df[\"AccAP\"].mean()\n        #df[\"std_acc_v\"] = df[\"AccV\"].mean()\n        #df[\"std_acc_ml\"] = df[\"AccML\"].mean()\n        #df[\"std_acc_ap\"] = df[\"AccAP\"].mean()\n        #df[\"rms_acc_v\"] = np.sqrt(np.mean(np.square(df[\"AccV\"])))\n        #df[\"rms_acc_ml\"] = np.sqrt(np.mean(np.square(df[\"AccML\"])))\n        #df[\"rms_acc_ap\"] = np.sqrt(np.mean(np.square(df[\"AccAP\"])))\n        #df[\"zcr_acc_v\"] = np.mean(np.diff(np.sign(df[\"AccV\"])) != 0)\n        #df[\"zcr_acc_ml\"] = np.mean(np.diff(np.sign(df[\"AccML\"])) != 0)\n        #df[\"zcr_acc_ap\"] = np.mean(np.diff(np.sign(df[\"AccAP\"])) != 0)\n        \n        #lags = [1,2]\n        #for lag in lags:\n            # Moving average\n        #    df[f'ma_{lag}_acc_v'] = df[\"AccV\"].rolling(lag).mean()\n        #    df[f'ma_{lag}_acc_ml'] = df[\"AccML\"].rolling(lag).mean()\n        #    df[f'ma_{lag}_acc_ap'] = df[\"AccAP\"].rolling(lag).mean()\n            # Moving std\n        #    df[f'std_{lag}_acc_v'] = df[\"AccV\"].rolling(lag).std()\n        #    df[f'std_{lag}_acc_ml'] = df[\"AccML\"].rolling(lag).std()\n        #    df[f'std_{lag}_acc_ap'] = df[\"AccAP\"].rolling(lag).std()\n            \n        \n        df['Id'] = f.split('/')[-1].split('.')[0]\n        df['Module'] = pathlib.Path(f).parts[-2]\n        df = pd.merge(df, tasks[['Id','t_kmeans']], how='left', on='Id').fillna(-1)\n#         df = pd.merge(df, subjects[['Id','s_kmeans']], how='left', on='Id').fillna(-1)\n        df = pd.merge(df, metadata_complex[['Id','Subject']+['Visit','Test','Medication','s_kmeans']], how='left', on='Id').fillna(-1)\n        df_feats = fc.calculate(df, return_df=True, include_final_window=True, approve_sparsity=True, window_idx=\"begin\").astype(np.float32)\n        df = df.merge(df_feats, how=\"left\", left_index=True, right_index=True)\n        df.fillna(method=\"ffill\", inplace=True)\n        return df\n    except: pass\ntrain = pd.concat([reader(f) for f in tqdm(train)]).fillna(0); print(train.shape)\ncols = [c for c in train.columns if c not in ['Id','Subject','Module', 'Time', 'StartHesitation', 'Turn' , 'Walking', 'Valid', 'Task','Event']]\npcols = ['StartHesitation', 'Turn' , 'Walking']\nscols = ['Id', 'StartHesitation', 'Turn' , 'Walking']","metadata":{"execution":{"iopub.status.busy":"2023-03-29T12:57:33.682044Z","iopub.execute_input":"2023-03-29T12:57:33.682824Z","iopub.status.idle":"2023-03-29T12:57:42.618495Z","shell.execute_reply.started":"2023-03-29T12:57:33.682758Z","shell.execute_reply":"2023-03-29T12:57:42.616737Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train=train.reset_index(drop=True)","metadata":{"execution":{"iopub.status.busy":"2023-03-29T12:57:42.620599Z","iopub.execute_input":"2023-03-29T12:57:42.620953Z","iopub.status.idle":"2023-03-29T12:57:42.900005Z","shell.execute_reply.started":"2023-03-29T12:57:42.620919Z","shell.execute_reply":"2023-03-29T12:57:42.898592Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Train the model","metadata":{}},{"cell_type":"code","source":"import xgboost as xgb\nimport lightgbm as lgb\nfrom sklearn.multioutput import MultiOutputRegressor\nfrom sklearn.metrics import mean_squared_error, r2_score\n","metadata":{"execution":{"iopub.status.busy":"2023-03-29T12:57:42.901999Z","iopub.execute_input":"2023-03-29T12:57:42.902424Z","iopub.status.idle":"2023-03-29T12:57:44.366256Z","shell.execute_reply.started":"2023-03-29T12:57:42.902386Z","shell.execute_reply":"2023-03-29T12:57:44.365076Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import numpy as np\nimport xgboost as xgb\nfrom sklearn.model_selection import RandomizedSearchCV, train_test_split\nfrom scipy.stats import uniform, randint\nfrom sklearn.metrics import average_precision_score, make_scorer\n\nbest_params_ = {'estimator__colsample_bytree': 0.5282057895135501, \n 'estimator__learning_rate': 0.22659963168004743, \n 'estimator__max_depth': 8, \n 'estimator__min_child_weight': 3.1233911067827616, \n 'estimator__n_estimators': 291, \n 'estimator__subsample': 0.9961057796456088}","metadata":{"execution":{"iopub.status.busy":"2023-03-29T12:57:44.369822Z","iopub.execute_input":"2023-03-29T12:57:44.370277Z","iopub.status.idle":"2023-03-29T13:00:31.358171Z","shell.execute_reply.started":"2023-03-29T12:57:44.37023Z","shell.execute_reply":"2023-03-29T13:00:31.356675Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"best_params_ = {kk: v for k, v in best_params_.items() for kk in k.split('__')}; del best_params_['estimator']","metadata":{"execution":{"iopub.status.busy":"2023-03-29T13:00:31.359825Z","iopub.execute_input":"2023-03-29T13:00:31.360682Z","iopub.status.idle":"2023-03-29T13:00:31.368241Z","shell.execute_reply.started":"2023-03-29T13:00:31.360637Z","shell.execute_reply":"2023-03-29T13:00:31.366953Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from sklearn.base import clone\n\ndef custom_average_precision(y_true, y_pred):\n    score = average_precision_score(y_true, y_pred)\n    return 'average_precision', score, True\n\nclass LGBMMultiOutputRegressor(MultiOutputRegressor):\n    def fit(self, X, y, eval_set=None, **fit_params):\n        self.estimators_ = [clone(self.estimator) for _ in range(y.shape[1])]\n        \n        for i, estimator in enumerate(self.estimators_):\n            if eval_set:\n                fit_params['eval_set'] = [(eval_set[0], eval_set[1][:, i])]\n            estimator.fit(X, y[:, i], **fit_params)\n        \n        return self","metadata":{"execution":{"iopub.status.busy":"2023-03-29T13:00:31.369926Z","iopub.execute_input":"2023-03-29T13:00:31.370245Z","iopub.status.idle":"2023-03-29T13:00:31.388673Z","shell.execute_reply.started":"2023-03-29T13:00:31.370215Z","shell.execute_reply":"2023-03-29T13:00:31.38766Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from sklearn.model_selection import GroupKFold\n\nN_FOLDS=5\nkfold = GroupKFold(N_FOLDS)\ngroup_var = train.Subject\ngroups=kfold.split(train, groups=group_var)\nregs=[]\ncvs=[]\nfor fold, (tr_idx,te_idx ) in enumerate(tqdm(groups, total=N_FOLDS, desc=\"Folds\")):\n    tr_idx=pd.Series(tr_idx).sample(n=2000000,random_state=42).values #2000000\n    \n    # Create a base XGBoost regressor with the common parameters\n    base_regressor = lgb.LGBMRegressor(**best_params_)\n\n    # Wrap the base regressor with the MultiOutputRegressor\n    multioutput_regressor = LGBMMultiOutputRegressor(base_regressor)\n\n    x_tr,y_tr=train.loc[tr_idx,cols].to_numpy(),train.loc[tr_idx,pcols].to_numpy()\n    x_te,y_te=train.loc[te_idx,cols].to_numpy(),train.loc[te_idx,pcols].to_numpy()\n\n    multioutput_regressor.fit(\n    x_tr,y_tr,\n    eval_set=(x_te,y_te),\n    eval_metric=custom_average_precision,\n    early_stopping_rounds=25\n    )\n    regs.append(multioutput_regressor)\n    cv=metrics.average_precision_score(y_te, multioutput_regressor.predict(x_te).clip(0.0,1.0))\n    cvs.append(cv)\nprint(cvs)","metadata":{"execution":{"iopub.status.busy":"2023-03-29T13:00:31.390401Z","iopub.execute_input":"2023-03-29T13:00:31.390712Z","iopub.status.idle":"2023-03-29T13:00:32.452086Z","shell.execute_reply.started":"2023-03-29T13:00:31.390682Z","shell.execute_reply":"2023-03-29T13:00:32.449609Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Predict for test","metadata":{}},{"cell_type":"code","source":"sub['t'] = 0\nsubmission = []\nfor f in test:\n    df = pd.read_csv(f)\n    df.set_index('Time', drop=True, inplace=True)\n    \n    #df[\"mean_acc_v\"] = df[\"AccV\"].mean()\n    #df[\"mean_acc_ml\"] = df[\"AccML\"].mean()\n    #df[\"mean_acc_ap\"] = df[\"AccAP\"].mean()\n    #df[\"min_acc_v\"] = df[\"AccV\"].mean()\n    #df[\"min_acc_ml\"] = df[\"AccML\"].mean()\n    #df[\"min_acc_ap\"] = df[\"AccAP\"].mean()\n    #df[\"max_acc_v\"] = df[\"AccV\"].mean()\n    #df[\"max_acc_ml\"] = df[\"AccML\"].mean()\n    #df[\"max_acc_ap\"] = df[\"AccAP\"].mean()\n    #df[\"median_acc_v\"] = df[\"AccV\"].mean()\n    #df[\"median_acc_ml\"] = df[\"AccML\"].mean()\n    #df[\"median_acc_ap\"] = df[\"AccAP\"].mean()\n    #df[\"std_acc_v\"] = df[\"AccV\"].mean()\n    #df[\"std_acc_ml\"] = df[\"AccML\"].mean()\n    #df[\"std_acc_ap\"] = df[\"AccAP\"].mean()\n    #df[\"rms_acc_v\"] = np.sqrt(np.mean(np.square(df[\"AccV\"])))\n    #df[\"rms_acc_ml\"] = np.sqrt(np.mean(np.square(df[\"AccML\"])))\n    #df[\"rms_acc_ap\"] = np.sqrt(np.mean(np.square(df[\"AccAP\"])))\n    #df[\"zcr_acc_v\"] = np.mean(np.diff(np.sign(df[\"AccV\"])) != 0)\n    #df[\"zcr_acc_ml\"] = np.mean(np.diff(np.sign(df[\"AccML\"])) != 0)\n    #df[\"zcr_acc_ap\"] = np.mean(np.diff(np.sign(df[\"AccAP\"])) != 0)\n    \n    #lags = [1,2,3]\n    #for lag in lags:\n        # Moving average\n    #    df[f'ma_{lag}_acc_v'] = df[\"AccV\"].rolling(lag).mean()\n    #    df[f'ma_{lag}_acc_ml'] = df[\"AccML\"].rolling(lag).mean()\n    #    df[f'ma_{lag}_acc_ap'] = df[\"AccAP\"].rolling(lag).mean()\n        # Moving std\n    #    df[f'std_{lag}_acc_v'] = df[\"AccV\"].rolling(lag).std()\n    #    df[f'std_{lag}_acc_ml'] = df[\"AccML\"].rolling(lag).std()\n    #    df[f'std_{lag}_acc_ap'] = df[\"AccAP\"].rolling(lag).std()\n            \n    \n    df['Id'] = f.split('/')[-1].split('.')[0]\n#     df = df.fillna(0).reset_index(drop=True)\n    df = pd.merge(df, tasks[['Id','t_kmeans']], how='left', on='Id').fillna(-1)\n#     df = pd.merge(df, subjects[['Id','s_kmeans']], how='left', on='Id').fillna(-1)\n    df = pd.merge(df, metadata_complex[['Id','Subject']+['Visit','Test','Medication','s_kmeans']], how='left', on='Id').fillna(-1)\n    df_feats = fc.calculate(df, return_df=True, include_final_window=True, approve_sparsity=True, window_idx=\"begin\")\n    df = df.merge(df_feats, how=\"left\", left_index=True, right_index=True)\n    df.fillna(method=\"ffill\", inplace=True)\n#     res = pd.DataFrame(np.round(reg.predict(df[cols]).clip(0.0,1.0),3), columns=pcols)\n    \n    res_vals=[]\n    for i_fold in range(N_FOLDS):\n        res_val=np.round(regs[i_fold].predict(df[cols]).clip(0.0,1.0),3)\n        res_vals.append(np.expand_dims(res_val,axis=2))\n    res_vals=np.mean(np.concatenate(res_vals,axis=2),axis=2)\n    res = pd.DataFrame(res_vals, columns=pcols)\n    \n    df = pd.concat([df,res], axis=1)\n    df['Id'] = df['Id'].astype(str) + '_' + df.index.astype(str)\n    submission.append(df[scols])\nsubmission = pd.concat(submission)\nsubmission = pd.merge(sub[['Id']], submission, how='left', on='Id').fillna(0.0)\nsubmission[scols].to_csv('submission.csv', index=False)","metadata":{"execution":{"iopub.status.busy":"2023-03-29T13:00:32.45349Z","iopub.status.idle":"2023-03-29T13:00:32.453932Z","shell.execute_reply.started":"2023-03-29T13:00:32.453703Z","shell.execute_reply":"2023-03-29T13:00:32.453725Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"submission","metadata":{"execution":{"iopub.status.busy":"2023-03-29T13:00:32.455205Z","iopub.status.idle":"2023-03-29T13:00:32.455613Z","shell.execute_reply.started":"2023-03-29T13:00:32.455416Z","shell.execute_reply":"2023-03-29T13:00:32.455438Z"},"trusted":true},"execution_count":null,"outputs":[]}]}