融合前两行作为变量的数据框

Question

apologies but this has me stumped, I thought I could pass the following dataframe into a simple pd.melt using iloc to reference my varaibles but it wasn't working for me (i'll post the error in a moment)

sample df

Date,     0151,        0561,       0522,   0912
0,Date,     AVG Review,  AVG Review, Review, Review 
1,Date      NaN          NaN          NaN    NaN
2,01/01/18  2            2.5          4        5 

so as you can see, my ID as in the top row, the type of review is in the 2nd row, the date sits in the first column and the observations of the review are in rows on the date.

what I'm trying to do is melt this df to get the following

ID,   Date,     Review,        Score
0151, 01/01/18, Average Review 2

I thought I could be cheeky and just pass the following

pd.melt pd.melt(df,id_vars=[df.iloc[0]],value_vars=df.iloc[1] )

but this threw the error 'Series' objects are mutable, thus they cannot be hashed

I've had a look at similar answers to pd.melt and perhaps reshape or unpivot? but I'm lost on how I should proceed.

any help is much appreciated.

Edit for Nixon :

My first Row has my unique IDs

2nd row has my observation, which in this case is a type of review (average, normal)

3rd row onward has the variables assigned to the above observation - lets call this score.

1st column has my dates which have the score across by row.

Answer 1

An alternative to pd.melt is to set your rows as column levels of a multiindex and then stack them. Your metadata will be stored as an index rather than column though. Not sure if that matters.

df = pd.DataFrame([
    ['Date',     '0151',        '0561',       '0522',   '0912'],
    ['Date',     'AVG Review',  'AVG Review', 'Review', 'Review'],
    ['Date',     'NaN',         'NaN',        'NaN',    'NaN'],
    ['01/01/18', 2,             2.5,          4,        5],
])

df = df.set_index(0)
df.index.name = 'Date'
df.columns = pd.MultiIndex.from_arrays([df.iloc[0, :], df.iloc[1, :]], names=['ID', 'Review'])
df = df.drop(df.index[[0, 1, 2]])

df.stack('ID').stack('Review')

Output:

Date      ID    Review    
01/01/18  0151  AVG Review      2
          0522  Review          4
          0561  AVG Review    2.5
          0912  Review          5
dtype: object

You can easily revert index to columns with reset_index .