data_processing_code/preprocess.py

import pandas as pd
import argparse
import os
import numpy as np
from scipy.ndimage import gaussian_filter1d
from sklearn.preprocessing import MinMaxScaler



#这个文件是用来在原始文件的基础上删除一些Trial，进行特征工程，然后再打上标签
def angle_between_vectors(v1, v2):
    """Calculate the angle between two vectors."""
    v1 = v1 / np.linalg.norm(v1)
    v2 = v2 / np.linalg.norm(v2)
    u_minus_v = v1 - v2
    u_plus_v = v1 + v2
    angle = 2 * np.arctan2(np.linalg.norm(u_minus_v), np.linalg.norm(u_plus_v))
    return np.degrees(angle)
def smooth_velocity(velocity, sigma=5):
    return gaussian_filter1d(velocity, sigma=sigma)
def calculate_angular_velocity(forward_vectors, timestamps):
    """计算角速度"""
    angles = np.array(
        [angle_between_vectors(forward_vectors[i], forward_vectors[i + 1]) if i + 1 < len(forward_vectors) else 0 for i
         in range(len(forward_vectors) - 1)])  # 注意这里的变化，排除了最后一个向量的计算
    # 计算时间间隔
    times = np.diff(timestamps)/1000
    # 计算角速度，对于第一个时间点，将角度设置为0
    if len(times) > 0:  # 检查times是否非空，避免除以空数组
        angular_velocities = np.insert(angles / times, 0, 0)  # 在结果数组的开始插入0
    else:
        angular_velocities = np.array([0])  # 如果times为空，说明只有一个时间点，无法计算速度
    return angular_velocities
def calculate_linear_velocity(positions, timestamps):
    """计算线性速度"""
    distances = np.linalg.norm(np.diff(positions, axis=0), axis=1)
    times = np.diff(timestamps)/1000
    linear_velocities = np.insert(distances / times, 0,0)  # Insert 0 at the start as there's no velocity at the first timestamp
    return linear_velocities
def calculate_direction(start_position, current_position):
    """计算方向"""
    direction = current_position - start_position
    norm = np.linalg.norm(direction)
    return direction / norm if norm != 0 else np.zeros_like(direction)
def calculate_acceleration(velocities, timestamps):
    """根据速度计算加速度"""
    accels = [0]  # 第一个时间点的加速度假设为0
    for i in range(1, len(velocities)):
        accel = (velocities[i] - velocities[i-1]) / ((timestamps[i] - timestamps[i-1])/1000)
        accels.append(accel)
    return np.array(accels)

## 特征工程，先加上三个模态的角速度，和旋转角，再加上手的线性速度和移动距离
def process_single_sequence(df):
    """处理groupBy之后的单个序列DataFrame，计算所有模态的角速度、线性速度、方向和旋转轴"""
    # 提取时间戳
    timestamps = df['TimeStamp'].values
    # 定义所有需要计算的模态
    modalities = ['HMD', 'Hand', 'Leye']
    for modality in modalities:
        # 计算角速度和旋转轴
        if all(f'{modality}ForwardV{i}' in df.columns for i in ['X', 'Y', 'Z']):
            forward_vectors = df[[f'{modality}ForwardVX', f'{modality}ForwardVY', f'{modality}ForwardVZ']].values
            initial_forward_vector = forward_vectors[0]
            df[f'{modality}A'] = [(angle_between_vectors(initial_forward_vector, fv)) for fv in forward_vectors]
            angular_velocity = calculate_angular_velocity(forward_vectors, timestamps)
            smoothed_angular_velocity = smooth_velocity(angular_velocity)
            df[f'{modality}AV'] = smoothed_angular_velocity
            df[f'{modality}AAcc'] = calculate_acceleration(smoothed_angular_velocity, timestamps)
            # 计算旋转轴并进行标准化以仅保留方向信息
            if modality == 'Hand':
                initial_forward_vector = forward_vectors[0]
                rotation_axes = np.array([np.cross(initial_forward_vector, fv) for fv in forward_vectors])
                rotation_axes_normalized = np.array(
                    [axis / np.linalg.norm(axis) if np.linalg.norm(axis) != 0 else np.array([0.0, 0.0, 0.0]) for axis in
                     rotation_axes]
                )
                df[f'{modality}RotationAxis_X'] = rotation_axes_normalized[:, 0]
                df[f'{modality}RotationAxis_Y'] = rotation_axes_normalized[:, 1]
                df[f'{modality}RotationAxis_Z'] = rotation_axes_normalized[:, 2]

        # 对于HMD和Hand，还需计算线性速度和方向还有移动距离
        if modality in ['HMD', 'Hand'] and all(f'{modality}Position{i}' in df.columns for i in ['X', 'Y', 'Z']):
            positions = df[[f'{modality}PositionX', f'{modality}PositionY', f'{modality}PositionZ']].values
            initial_position = positions[0]
            linear_velocity = calculate_linear_velocity(positions, timestamps)
            df[f'{modality}L'] = [np.linalg.norm(pos-initial_position) for pos in positions]
            smoothed_velocity = smooth_velocity(linear_velocity)
            df[f'{modality}LV'] = smoothed_velocity
            df[f'{modality}LAcc'] = calculate_acceleration(smoothed_velocity, timestamps)  # 新特性：加速度
            # 计算方向
            if modality == 'Hand':
                start_position = positions[0]
                directions = np.array([calculate_direction(start_position, pos) for pos in positions])
                df[f'{modality}Direction_X'] = directions[:, 0]
                df[f'{modality}Direction_Y'] = directions[:, 1]
                df[f'{modality}Direction_Z'] = directions[:, 2]

    return df


def label_trials_with_motion_metrics(df):
    # Define a helper function for labeling each group
    def label_group(group):
        # For the 'HandLinearDistance' and 'HandAngularDistance', take the last value in the group as it represents the total
        total_linear_distance = group['HandL'].iloc[-1]
        total_angular_distance = group['HandA'].iloc[-1]
        # Assign these totals to new columns for every row in the group
        group['LLabel'] = total_linear_distance
        group['ALabel'] = total_angular_distance
        return group
    # Apply the labeling function to each trial group
    df_labeled = df.groupby(['ParticipantID', 'BlockID', 'TrialID'], group_keys=False).apply(label_group)
    return df_labeled


def split_data_by_theta_grouped(df):
    grouped = df.groupby(['ParticipantID', 'BlockID', 'TrialID'])
    theta_groups = {}
    # 将每个组添加到对应Theta值的列表中
    for _, group in grouped:
        theta_value = group['Theta'].iloc[0]
        if theta_value not in theta_groups:
            theta_groups[theta_value] = []
        theta_groups[theta_value].append(group)
    train_dfs = []
    test_dfs = []
    for theta_value, groups in theta_groups.items():
        # 计算训练集大小
        n_train = int(len(groups) * 0.8)
        # 随机选择训练集序列
        np.random.seed(1)  # 确保可重复性
        train_indices = np.random.choice(len(groups), size=n_train, replace=False)
        train_groups = [groups[i] for i in train_indices]
        # 选择测试集序列
        test_groups = [groups[i] for i in range(len(groups)) if i not in train_indices]
        # 将训练集和测试集序列合并
        train_df = pd.concat(train_groups)
        test_df = pd.concat(test_groups)
        train_dfs.append(train_df)
        test_dfs.append(test_df)
    # 合并所有训练集和测试集的DataFrame
    final_train_df = pd.concat(train_dfs).sort_index()
    final_test_df = pd.concat(test_dfs).sort_index()
    return final_train_df, final_test_df


def split_data_by_theta_grouped(df):
    grouped = df.groupby(['ParticipantID', 'BlockID', 'TrialID'])
    theta_groups = {}
    # 将每个组添加到对应Theta值的列表中
    for _, group in grouped:
        theta_value = group['Theta'].iloc[0]
        if theta_value not in theta_groups:
            theta_groups[theta_value] = []
        theta_groups[theta_value].append(group)
    train_dfs = []
    test_dfs = []
    for theta_value, groups in theta_groups.items():
        # 计算训练集大小
        n_train = int(len(groups) * 0.8)
        # 随机选择训练集序列
        np.random.seed(1)  # 确保可重复性
        train_indices = np.random.choice(len(groups), size=n_train, replace=False)
        train_groups = [groups[i] for i in train_indices]
        # 选择测试集序列
        test_groups = [groups[i] for i in range(len(groups)) if i not in train_indices]
        # 将训练集和测试集序列合并
        train_df = pd.concat(train_groups)
        test_df = pd.concat(test_groups)
        train_dfs.append(train_df)
        test_dfs.append(test_df)
    # 合并所有训练集和测试集的DataFrame
    final_train_df = pd.concat(train_dfs).sort_index()
    final_test_df = pd.concat(test_dfs).sort_index()
    return final_train_df, final_test_df


# 定义一个函数，将特征缩放到指定的范围内
def preprocess_features(train_df, test_df):
    # 定义包含负值的特征列和其他特征列
    negative_value_features = ['HandAAcc', 'HMDAAcc',"LeyeAAcc","HandLAcc","HMDLAcc"]
    other_features = ["HMDA", "HMDAV", "HandA", "HandAV", "LeyeA", "LeyeAV","HMDL", "HMDLV","HandL", "HandLV"]
    # 为包含负值的特征和其他特征创建两个不同的缩放器
    scaler_negatives = MinMaxScaler(feature_range=(-1, 1))
    scaler_others = MinMaxScaler(feature_range=(0, 1))
    # 对训练集应用fit_transform，对测试集应用transform
    train_df[negative_value_features] = scaler_negatives.fit_transform(train_df[negative_value_features])
    test_df[negative_value_features] = scaler_negatives.transform(test_df[negative_value_features])
    train_df[other_features] = scaler_others.fit_transform(train_df[other_features])
    test_df[other_features] = scaler_others.transform(test_df[other_features])
    return train_df, test_df

 #这个function用于修改Evaluation的数据集
def main(Participant_ID):
    # 读入数据
    input_file_path = f'../Data/Study2Evaluation/{Participant_ID}_Trajectory.csv'
    output_train_path = f'../Data/Study2Evaluation/Preprocessed/{Participant_ID}_train_data_preprocessed_evaluation.csv'
    output_test_path = f'../Data/Study2Evaluation/Preprocessed/{Participant_ID}_test_data_preprocessed_evaluation.csv'

    data=pd.read_csv(input_file_path)
    data_cleaned = data.loc[:, ~data.columns.str.contains('^Unnamed')]
    data_no_error = data_cleaned[data_cleaned['isError'] == False]
    final_data = data_no_error[(data_no_error['TrialID'] != 0) & (data_no_error['TrialID'] != 6) & (data_no_error['TrialID'] != 12)
                               & (data_no_error['TrialID'] != 18) & (data_no_error['TrialID'] != 24) & (data_no_error['TrialID'] != 30) & (data_no_error['TrialID'] != 36)]
    # 特征工程
    processed_groups = final_data.groupby(['ParticipantID','BlockID', 'TrialID']).apply(process_single_sequence)
    processed_df = processed_groups.reset_index(drop=True)

    # 为处理完的数据添加标签
    df_with_features_labelled = label_trials_with_motion_metrics(processed_df.copy())
    final_train_df, final_test_df = split_data_by_theta_grouped(df_with_features_labelled)
    # 特征预处理
    final_train_df_preprocessed, final_test_df_preprocessed = preprocess_features(final_train_df.copy(), final_test_df.copy())
    # 保存处理后的数据集到CSV文件
    final_train_df_preprocessed.to_csv(output_train_path, index=False)
    final_test_df_preprocessed.to_csv(output_test_path, index=False)


# def main(Participant_ID):
#     # 读入数据
#     input_file_path = f'../Data/{Participant_ID}_Trajectory.csv'
#     output_train_path = f'../Data/Study1AllUSers/Preprocessed/{Participant_ID}_train_data_preprocessed.csv'
#     output_test_path = f'../Data/Study1AllUSers/Preprocessed/{Participant_ID}_test_data_preprocessed.csv'
#     data = pd. read_csv(input_file_path)
#     participant_id = int(os.path.basename(input_file_path).split('_')[0])
#     data.insert(0, 'ParticipantID', participant_id)
#     data_cleaned = data.loc[:, ~data.columns.str.contains('^Unnamed')]
#     data_no_error = data_cleaned[data_cleaned['isError'] == False]
#     cleaned_data_path = '../Data/cleaned_data_trimmed.xlsx'
#     cleaned_data = pd.read_excel(cleaned_data_path)
#     filtered_data = pd.merge(data_no_error, cleaned_data, on=['ParticipantID', 'BlockID', 'TrialID'], how='inner')
#     final_data = filtered_data[(filtered_data['TrialID'] != 0) & (filtered_data['TrialID'] != 8) & (filtered_data['TrialID'] != 16) & (filtered_data['TrialID'] != 24)]
#     # 特征工程
#     processed_groups = final_data.groupby(['ParticipantID','BlockID', 'TrialID']).apply(process_single_sequence)
#     processed_df = processed_groups.reset_index(drop=True)
#     # 为处理完的数据添加标签
#     df_with_features_labelled = label_trials_with_motion_metrics(processed_df.copy())
#     final_train_df, final_test_df = split_data_by_theta_grouped(df_with_features_labelled)
#     # 特征预处理
#     final_train_df_preprocessed, final_test_df_preprocessed = preprocess_features(final_train_df.copy(), final_test_df.copy())
#     # 保存处理后的数据集到CSV文件
#     final_train_df_preprocessed.to_csv(output_train_path, index=False)
#     final_test_df_preprocessed.to_csv(output_test_path, index=False)

if __name__ == '__main__':
    for i in range(79, 80):
        main(str(i))