python - xarray 多索引连接的最佳实践

Question

我有一组 1000 个(2D)pd.Dataframe(比如，index:time, columns: run_id)，每个都有 3 个属性(比如温度、压力、位置) )。理想情况下，我希望将所有内容都放在 5 维的 xr.DataArray 中(或 4 维的 xr.Dataset 并将最后一个维度作为唯一数据变量)。

我创建了一个具有两个暗淡和 2+3 坐标的 DataArray，但 xr.concat 似乎不适用于多个维度。 (我遵循了这里提到的方法Add 'constant' dimension to xarray Dataset。)

示例:我从各个数据帧和属性列表构建 DataArray。

# Mock data:
data = {}
for i in np.arange(500):
    data[i] = pd.DataFrame(np.random.randn(1000, 8), index=pd.DatetimeIndex(start='01.01.2013',periods=1000,freq='h'),
                    columns=list('ABCDEFGH'))
df_catalogue = pd.DataFrame(np.random.choice(10,(500, 3)), columns=['temp','pre','zon'])

#Build DataArrays adding scalar coords
res_da = []
for i,v in df_catalogue.iterrows():
    i_df = data[i] # data is a dictionary of properly indexed dataframes

    da = xr.DataArray(i_df.values,
                   coords={'time':i_df.index.values,'runs':i_df.columns.values,
                           'temp':v['temp'], 'pre':v['pre'],'zon':v['zon']},
                   dims=['time','runs'])
    res_da.append(da)

但是当我尝试 all_da = xr.concat(res_da, dim=['temp','pre','zon']) 时，我得到了奇怪的结果。实现这样的目标的最佳方法是什么:

<xarray.DataArray (time: 8000, runs: 50, temp:8, pre:10, zon: 5)>
array([[[ 4545.453613,  4545.453613, ...,  4545.453613,  4545.453613],
        [ 4545.453613,  4545.453613, ...,  4545.453613,  4545.453613],
        ..., 
        [ 4177.425781,  4177.425781, ...,  4177.425781,  4177.425781]]], dtype=float32)
Coordinates:
  * runs  (runs) object 'A' 'B' ...
  * time  (time) datetime64[ns] 2013-12-31T23:00:00 2014-01-01 ...
  * zon   (zon) 'zon1', 'zon2', 'zon3', ......
  * temp  (temp)  'XX' 'YY', 'ZZ' .....
  * pre   (pre) 'AAA', 'BBB', 'CCC' ....

Answer 1

xarray.concat 仅支持沿单个维度串联。但我们可以通过连接、设置 MultiIndex 然后取消堆叠来解决这个问题。

我正在更改您的设置代码，因为这仅在您正在构建的新坐标的每个组合 (['temp','pre','zon']) 都是唯一的情况下才有效:

import numpy as np
import pandas as pd
import xarray as xr
import itertools

data = {}
for i in np.arange(500):
    data[i] = pd.DataFrame(np.random.randn(1000, 8),
                           index=pd.DatetimeIndex(start='01.01.2013',periods=1000,freq='h'),
                           columns=list('ABCDEFGH'))
cat_data = [(x, y, z)
            for x in range(20)
            for y in ['a', 'b', 'c', 'd', 'e']
            for z in ['A', 'B', 'C', 'D', 'E']]
df_catalogue = pd.DataFrame(cat_data, columns=['temp','pre','zon'])

#Build DataArrays adding scalar coords
res_da = []
for i,v in df_catalogue.iterrows():
    i_df = data[i] # data is a dictionary of properly indexed dataframes

    da = xr.DataArray(i_df.values,
                   coords={'time':i_df.index.values,'runs':i_df.columns.values,
                           'temp':v['temp'], 'pre':v['pre'],'zon':v['zon']},
                   dims=['time','runs'])
    res_da.append(da)

然后，我们可以简单地写:

xr.concat(res_da, dim='prop').set_index(prop=['temp', 'pre', 'zon']).unstack('prop')

这会产生您想要的 5D 数组:

<xarray.DataArray (time: 1000, runs: 8, temp: 20, pre: 5, zon: 5)>
array([[[[[-0.690557, ..., -1.526415],
          ...,
          [ 0.737887, ...,  1.585335]],

         ...,

         [[ 0.99557 , ...,  0.256517],
          ...,
          [ 0.179632, ..., -1.236502]]],


        ...,


        [[[ 0.234426, ..., -0.149901],
          ...,
          [ 1.492255, ..., -0.380909]],

         ...,

         [[-0.36111 , ..., -0.451571],
          ...,
          [ 0.10457 , ...,  0.722738]]]]])
Coordinates:
  * time     (time) datetime64[ns] 2013-01-01 2013-01-01T01:00:00 ...
  * runs     (runs) object 'A' 'B' 'C' 'D' 'E' 'F' 'G' 'H'
  * temp     (temp) int64 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
  * pre      (pre) object 'a' 'b' 'c' 'd' 'e'
  * zon      (zon) object 'A' 'B' 'C' 'D' 'E'