python - groupby 并计算平均值但保留所有列

Question

我想使用 groupby 计算数字列的平均值，但保留所有列。来自 7 列的数据框示例如下:

tracking_id gene_id gene_short_name tss_id  locus   FPKM-1  FPKM-2ENSMUSG00000025902  ENSMUSG00000025902  Sox17   Tss1231 1:4490927-4496413   0.611985    232ENSMUSG00000025902  ENSMUSG00000025902  Sox17   Ts412   1:4490927-4496413   12  21ENSMUSG00000025902  ENSMUSG00000025902  Sox17   Ts56    1:4490927-4496413   2   213ENSMUSG00000025902  ENSMUSG00000025902  Sox17   TS512   1:4490927-4496413   0.611985    5ENSMUSG00000025902  ENSMUSG00000025902  Sox17   TS12241 1:4490927-4496413   0.611985    51ENSMUSG00000096126  ENSMUSG00000096126  Gm22307 TS124   1:4529016-4529123   35  1ENSMUSG00000096126  ENSMUSG00000096126  Gm22307 TS-1824 1:4529016-4529123   1   2ENSMUSG00000096126  ENSMUSG00000096126  Gm22307 TS1249082   1:4529016-4529123   2   5ENSMUSG00000088000  ENSMUSG00000088000  Gm25493 TS1290328   1:4723276-4723379   0   1ENSMUSG00000098104  ENSMUSG00000098104  Gm6085  TS01239-1   1:4687933-4689403   0.0743559   6ENSMUSG00000033845  ENSMUSG00000033845  Mrpl15  TSS31014,TSS82987,TSS82990,TSS86849 1:4773205-4785739   79.1154 7ENSMUSG00000093015  ENSMUSG00000093015  Gm22463 TSS79849    1:5644644-5644745   0   1ENSMUSG00000025905  ENSMUSG00000025905  Oprk1   TSS15316,TSS3878,TSS6226,TSS65522   1:5588492-5606131   0   6ENSMUSG00000033774  ENSMUSG00000033774  Npbwr1  TSS69693    1:5913706-5917398   0   8ENSMUSG00000033793  ENSMUSG00000033793  Atp6v1h TSS4651 1:5083172-5162549   24.2386 9ENSMUSG00000087247  ENSMUSG00000087247  Fam150a TSS42747    1:6359330-6394731   0.502804    1

I would like to group by the first 3 columns, and keep columns 4 and 5 in my output (best would be the first row of each repeated columns 1 to 3) and then calculate the mean of the numeric columns at the end. I have written this:

import pandas as pd
df = pd.read_table('grouping.txt')
grouped = df.groupby(list(df.columns[0:3]), sort=False).mean()

输出是:

tracking_id gene_id gene_short_name FPKM-1  FPKM-2ENSMUSG00000025902  ENSMUSG00000025902  Sox17   3.167191    104.4ENSMUSG00000096126  ENSMUSG00000096126  Gm22307 12.66666667 2.666666667ENSMUSG00000088000  ENSMUSG00000088000  Gm25493 0   1ENSMUSG00000098104  ENSMUSG00000098104  Gm6085  0.0743559   6ENSMUSG00000033845  ENSMUSG00000033845  Mrpl15  79.1154 7ENSMUSG00000093015  ENSMUSG00000093015  Gm22463 0   1ENSMUSG00000025905  ENSMUSG00000025905  Oprk1   0   6ENSMUSG00000033774  ENSMUSG00000033774  Npbwr1  0   8ENSMUSG00000033793  ENSMUSG00000033793  Atp6v1h 24.2386 9ENSMUSG00000087247  ENSMUSG00000087247  Fam150a 0.502804    1

以上是输出，但缺少输入文件的第 4 列(TSS)和第 5 列(基因座)。我怎样才能保留这两列(它们的值不同，因此不能成为 groupby 列的一部分。保留任何列的值对我来说都是可以的，只要有一个分组依据)。

Answer 1

您可以将 groupby() 聚合的结果合并回原始 DataFrame 的去重版本。也许是这样的:

# identify the columns we want to aggregate by; this could
# equivalently be defined as list(df.columns[0:3])
group_cols = ['tracking_id', 'gene_id', 'gene_short_name']
# identify the columns which we want to average; this could
# equivalently be defined as list(df.columns[4:])
metric_cols = ['FPKM-1', 'FPKM-2']

# create a new DataFrame with a MultiIndex consisting of the group_cols
# and a column for the mean of each column in metric_cols
aggs = df.groupby(group_cols)[metric_cols].mean()
# remove the metric_cols from df because we are going to replace them
# with the means in aggs
df.drop(metric_cols, axis=1, inplace=True)
# dedupe to leave only one row with each combination of group_cols
# in df
df.drop_duplicates(subset=group_cols, keep='last', inplace=True)
# add the mean columns from aggs into df
df = df.merge(right=aggs, right_index=True, left_on=group_cols, how='right')