Pandas数据框：将列转换为单列的行

Question

I have a dataframe that looks like

userId  feature1  feature2  feature3  ...
123456  0         0.45      0         ...
234567  0         0         0         ...
345678  0.6       0         0.2       ...
.
.

The features are mostly zeros but occasionally some of those would have non-zero values. A single row for a userId may have zero, one or more non-zero features.

I want to transform this into the following dataset:

userId  feature  value
123456  feature2 0.45
345678  feature1 0.6
345678  feature3 0.2

Essentially, we retain only the features that are non-zero for each userId. So, for userId 345678, we have 2 rows in the transformed dataset, one for feature1 and the other for feature3. userId 234567 is dropped since none of the features are non-zero.

Is this something that can be done using groupby or pivoting? If so, how?

Any other pandas-mic solutions?

Answer 1

Magic from melt

df.melt('userId').query('value!=0')
Out[459]: 
   userId  variable  value
2  345678  feature1   0.60
3  123456  feature2   0.45
8  345678  feature3   0.20

Notice using stack you need mask 0 to NaN

df.mask(df.eq(0)).set_index('userId').stack().reset_index()
Out[460]: 
   userId   level_1     0
0  123456  feature2  0.45
1  345678  feature1  0.60
2  345678  feature3  0.20