Question

我有一个完整阶乘实验的数据。例如，对于每个N样本，我有J个测量类型和K个测量位点。例如，我以长格式收到此数据，

import numpy as np
import pandas as pd
import itertools
from numpy.random import normal as rnorm

# [[N], [J], [K]]
levels = [[1,2,3,4], ['start', 'stop'], ['gene1', 'gene2', 'gene3']]

# fully crossed
exp_design = list(itertools.product(*levels))

df = pd.DataFrame(exp_design, columns=["sample", "mode", "gene"])

# some fake data
df['x'] = rnorm(size=len(exp_design))

这将得出24个观测值（x），其中包含三个因素中每个因素的列。

> df.head()
    sample  mode    gene    x
0   1       start   gene1   -1.229370
1   1       start   gene2   1.129773
2   1       start   gene3   -1.155202
3   1       stop    gene1   -0.757551
4   1       stop    gene2   -0.166129

我想将这些观察值转换为相应的(N,J,K)形张量（numpy数组）。我当时正在考虑使用MultiIndex转向宽格式，然后提取值将生成正确的张量，但它只是作为列向量而出现：

> df.pivot_table(values='x', index=['sample', 'mode', 'gene']).values
array([[-1.22936989],
       [ 1.12977346],
       [-1.15520216],
       ...,
       [-0.1031641 ],
       [ 1.1296491 ],
       [ 1.31113584]])

是否可以从长格式pandas.DataFrame中获取张量格式的数据？

Answer 1

尝试

df.agg('nunique')

Out[69]: 
sample     4
mode       2
gene       3
x         24
dtype: int64
s=df.agg('nunique')
df.x.values.reshape(s['sample'],s['mode'],s['gene'])
Out[71]: 
array([[[-2.78133759e-01, -1.42234420e+00,  5.42439121e-01],
        [ 2.15359867e+00,  6.55837886e-01, -1.01293568e+00]],
       [[ 7.92306679e-01, -1.62539763e-01, -6.13120335e-01],
        [-2.91567999e-01, -4.01257702e-01,  7.96422763e-01]],
       [[ 1.05088264e-01, -7.23400925e-02,  2.78515041e-01],
        [ 2.63088568e-01,  1.47477886e+00, -2.10735619e+00]],
       [[-1.71756374e+00,  6.12224005e-04, -3.11562798e-02],
        [ 5.26028807e-01, -1.18502045e+00,  1.88633760e+00]]])

使用形状的因子水平将pandas.DataFrame转换为numpy张量

1 个答案: