python中使用dataframe二维数据转换为三维数据的几种方法

import pandas as pd
import numpy as np

# 创建示例DataFrame
df = pd.DataFrame({
    'A': range(12),
    'B': range(12, 24),
    'C': range(24, 36)
})

# 转换为三维数组 (3, 4, 3) - 3个时间步长，4个样本，3个特征
array_3d = df.values.reshape(3, 4, 3)
print(array_3d.shape)  # (3, 4, 3)

# 假设每4行代表一个时间序列段
n_timesteps = 3
n_samples = 4
n_features = 3

# 确保DataFrame行数匹配
assert len(df) == n_timesteps * n_samples

# 重塑为 (时间步长, 样本数, 特征数)
reshaped = df.values.reshape(n_timesteps, n_samples, n_features)

# 创建包含分组列的DataFrame
df_grouped = pd.DataFrame({
    'group': ['A', 'A', 'A', 'B', 'B', 'B', 'C', 'C', 'C'],
    'feature1': range(9),
    'feature2': range(10, 19)
})

# 按照分组转换为三维
groups = df_grouped['group'].unique()
n_groups = len(groups)
n_rows_per_group = 3
n_features = 2

array_3d = np.zeros((n_groups, n_rows_per_group, n_features))

for i, group in enumerate(groups):
    group_data = df_grouped[df_grouped['group'] == group][['feature1', 'feature2']].values
    array_3d[i] = group_data

# 创建多层索引DataFrame
index = pd.MultiIndex.from_product([
    ['Group1', 'Group2'],  # 第一层
    ['Time1', 'Time2', 'Time3'],  # 第二层
    ['Feature1', 'Feature2']  # 第三层
], names=['Group', 'Time', 'Feature'])

df_multi = pd.DataFrame({'Value': np.random.randn(12)}, index=index)

# 转换为三维数组
unstacked = df_multi.unstack(level=['Group', 'Time'])
array_3d = unstacked.values.reshape(2, 3, 2)

def create_sequences(data, seq_length):
    """将时间序列数据转换为监督学习格式"""
    sequences = []
    for i in range(len(data) - seq_length + 1):
        seq = data.iloc[i:i + seq_length].values
        sequences.append(seq)
    return np.array(sequences)

# 创建序列数据
sequence_length = 3
X_3d = create_sequences(df, sequence_length)
print(X_3d.shape)  # (10, 3, 3) - 10个序列，每个3个时间步，每个时间步3个特征

# TensorFlow
import tensorflow as tf
tensor_3d = tf.reshape(tf.constant(df.values), (3, 4, 3))

# PyTorch
import torch
tensor_3d = torch.tensor(df.values).reshape(3, 4, 3)

import xarray as xr
# 创建三维DataArray
da = xr.DataArray(
    df.values.reshape(3, 4, 3),
    dims=['time', 'sample', 'feature'],
    coords={
        'time': pd.date_range('2023-01-01', periods=3, freq='D'),
        'sample': range(4),
        'feature': ['A', 'B', 'C']
    }
)