Performance on the MEKA wrapper

[necrosource-bot]

Mirrored from scikit-multilearn/scikit-multilearn#302

I am using the MEKA wrapper. It's OK, but when I print the variable output_ (print(meka.output_)), the performance measurements are mostly zero.

== Predictive Performance

Number of test instances (N)   
Accuracy                       0    
Jaccard index                  0    
Hamming score                  0.668
Exact match                    0    
Jaccard distance               1    
Hamming loss                   0.332
ZeroOne loss                   1    
Harmonic score                 NaN
One error                      1    
Rank loss                      0    
Avg precision                  1    
Log Loss (lim. L)              0.394
Log Loss (lim. D)              0.394
Micro Precision                0    
Micro Recall                   0    
Macro Precision                0    
Macro Recall                   0    
F1 (micro averaged)            0    
F1 (macro averaged by example) 0    
F1 (macro averaged by label)   0    
AUPRC (macro averaged)         NaN
AUROC (macro averaged)         NaN
Curve Data                     
Macro Curve Data               
Micro Curve Data               
Label indices                  [     0     1     2     3     4     5 ]
Accuracy (per label)           [ 1,000 1,000 0,005 0,010 1,000 0,995 ]
Harmonic (per label)           [   NaN   NaN   NaN   NaN   NaN   NaN ]
Precision (per label)          [ 0,000 0,000 0,000 0,000 0,000 0,000 ]
Recall (per label)             [ 0,000 0,000 0,000 0,000 0,000 0,000 ]
Empty labelvectors (predicted) 0    
Label cardinality (predicted)  1.99 
Levenshtein distance           0.332
Label cardinality (difference) -1.99
avg. relevance (test set)      [ 0,000 0,000 0,000 0,000 0,000 0,000 ]
avg. relevance (predicted)     [ 0,000 0,000 0,995 0,990 0,000 0,005 ]
avg. relevance (difference)    [ 0,000 0,000 -0,995 -0,990 0,000 -0,005 ]

What could you do to ensure that all the values are filled in?

I could calculate the measures via Python, but some need the scores (probability) from the classification (like Ranking Loss and Coverage, for example). In this case, how do I obtain the classification scores in the MEKA wrapper?