Parkinson‘s Freezing of Gait Prediction- baseline 学习

关注我的公众号YueTan进行交流探讨
欢迎关注数据比赛方案仓库 https://github.com/hongyingyue/Competition-solutions

base

time-series 🤝 tsflex - 0.237

def reader(f):
    try:
        df = pd.read_csv(f, index_col="Time", usecols=['Time', 'AccV', 'AccML', 'AccAP', 'StartHesitation', 'Turn' , 'Walking'])
        df['Id'] = f.split('/')[-1].split('.')[0]
        df = pd.merge(df, tasks[['Id','t_kmeans']], how='left', on='Id').fillna(-1)
        df = pd.merge(df, subjects[['Id','s_kmeans']], how='left', on='Id').fillna(-1)
        df_feats = fc.calculate(df, return_df=True, include_final_window=True, approve_sparsity=True, window_idx="begin").astype(np.float32)
        df = df.merge(df_feats, how="left", left_index=True, right_index=True)
        df.fillna(method="ffill", inplace=True)
        return df
    except: pass
train = pd.concat([reader(f) for f in tqdm(train)]).fillna(0); print(train.shape)
cols = [c for c in train.columns if c not in ['Id', 'Time', 'StartHesitation', 'Turn' , 'Walking', 'Valid', 'Task','Event']]
pcols = ['StartHesitation', 'Turn' , 'Walking']
scols = ['Id', 'StartHesitation', 'Turn' , 'Walking']

• 从tasks中增加kmeans类别
• 从subject中增加kmean类别
• 用tsflex增加时序特征【basic_feats, emg_feats】
• ensemble.ExtraTreesRegressor同时预测三个目标

base

GroupKfold Cross-Validation tsflex-0.246

• GroupKfold Cross Validation

def reader(f):
    try:
        df = pd.read_csv(f, index_col="Time", usecols=['Time', 'AccV', 'AccML', 'AccAP', 'StartHesitation', 'Turn' , 'Walking'])
        df['Id'] = f.split('/')[-1].split('.')[0]
        df['Module'] = pathlib.Path(f).parts[-2]
        df = pd.merge(df, tasks[['Id','t_kmeans']], how='left', on='Id').fillna(-1)
#         df = pd.merge(df, subjects[['Id','s_kmeans']], how='left', on='Id').fillna(-1)
        df = pd.merge(df, metadata_complex[['Id','Subject']+['Visit','Test','Medication','s_kmeans']], how='left', on='Id').fillna(-1)
        df_feats = fc.calculate(df, return_df=True, include_final_window=True, approve_sparsity=True, window_idx="begin").astype(np.float32)
        df = df.merge(df_feats, how="left", left_index=True, right_index=True)
        df.fillna(method="ffill", inplace=True)
        return df
    except: pass

• 保留采用subject聚类特征，而是采用了meta_data中关于subject的特征

base

Simple EDA on Time for targets-0.306

import pathlib
def reader(f):
    try:
        df = pd.read_csv(f, index_col="Time", usecols=['Time', 'AccV', 'AccML', 'AccAP', 'StartHesitation', 'Turn' , 'Walking'])
        
        df['Id'] = f.split('/')[-1].split('.')[0]
        df['Module'] = pathlib.Path(f).parts[-2]
        
        df['Time_frac']=(df.index/df.index.max()).values#currently the index of data is actually "Time"
        
        df = pd.merge(df, tasks[['Id','t_kmeans']], how='left', on='Id').fillna(-1)
#         df = pd.merge(df, subjects[['Id','s_kmeans']], how='left', on='Id').fillna(-1)
        df = pd.merge(df, metadata_complex[['Id','Subject']+['Visit','Test','Medication','s_kmeans']], how='left', on='Id').fillna(-1)
        df_feats = fc.calculate(df, return_df=True, include_final_window=True, approve_sparsity=True, window_idx="begin").astype(np.float32)
        df = df.merge(df_feats, how="left", left_index=True, right_index=True)
        df.fillna(method="ffill", inplace=True)
        return df
    except: pass
train = pd.concat([reader(f) for f in tqdm(train)]).fillna(0); print(train.shape)
cols = [c for c in train.columns if c not in ['Id','Subject','Module', 'Time', 'StartHesitation', 'Turn' , 'Walking', 'Valid', 'Task','Event']]
pcols = ['StartHesitation', 'Turn' , 'Walking']
scols = ['Id', 'StartHesitation', 'Turn' , 'Walking']

测试时

df['Time_frac']=(df.index/df.index.max()).values#currently the index of data is actually "Time"
    df = pd.merge(df, tasks[['Id','t_kmeans']], how='left', on='Id').fillna(-1)
#     df = pd.merge(df, subjects[['Id','s_kmeans']], how='left', on='Id').fillna(-1)
    df = pd.merge(df, metadata_complex[['Id','Subject']+['Visit','Test','Medication','s_kmeans']], how='left', on='Id').fillna(-1)
    df_feats = fc.calculate(df, return_df=True, include_final_window=True, approve_sparsity=True, window_idx="begin")

• 新特征
• 模型LGB

base

Gait Prediction-0.311/0.283

base-nn-torch

PyTorch FOG End-to-End Baseline [LB 0.254]

base-nn-tf

Parkinson FoG Pred Conv1D Separate TF Model-0.296

• 针对两种数据集分开建立模型训练，主要采用序列数据

_is_tdcs = basename(dirname(path)).startswith('tdcs')
df = pd.read_csv(path)  # 加载该序列数据

_cols = [*self.cfg.feature_list, *self.cfg.label_list, 'Valid', 'Task']  # 三个加速度，label 和Valid Task
self.valid_position = self.cfg.n_features + self.cfg.n_labels
self.task_position = self.valid_position + 1

# Pads past and future rows to dataframe values for indexing   # 选出需要的列，并padding
_values = df[cols].values.astype(np.float16)
return np.pad(_values, ((self.past_pad, self.future_pad),(0,0)), 'edge')

# 以上是单个——read
_values = [self._read(f) for f in df_paths]
        
self.mapping = []
_length = 0
for _value in _values:
    _shape = _value.shape[0]
    self.mapping.extend(range(_length+self.past_pad, _length+_shape-self.future_pad))
    _length += _shape

对于这个数据集，如何构建三维的关键在这里

• 对于y，根据batch的index选择即可
• 对于x, idx到过去的pad, idx到未来的pad, 中间还有间隔。也就是此baseline的关键

def _get_X_y(self, indices):
        _X = np.empty((len(indices), self.cfg.window_size, self.cfg.n_features), dtype=np.float16)
        for i, idx in enumerate(indices):
            _X[i] = self.values[idx-self.past_pad: idx+self.future_pad+1:self.cfg.wx, :self.cfg.n_features]
        return _X, self.values[indices, self.cfg.n_features:self.cfg.n_features+self.cfg.n_labels]