Question About Methods in Predictive Imputer

[oattah1]

• Predictive Imputer version:
• Python version:
• Operating System:

Description

Hi,
Thank you so much for implementing missForest in python; it has been really helpful. I was wondering why self, instead of new_imputed similar to the output of the pseudocode in the missForest paper(Algorithm 1; https://academic.oup.com/bioinformatics/article/28/1/112/219101)? Also in test_predictive_imputer.py, I see that in the test_predictive_imputer method after you called the predictive imputer class with imputer = predictive_imputer.PredictiveImputer() then you call imputer.fit(X).transform(X.copy())and assign that to be X_trans. I looked at the transform method and I noticed that the initial code to the transformed method is similar to the beginning of the fit method. I was wondering why is the transform method called after the model is fitted? Also could you explain more about the transform method?

What I Did

I copy the code in the Predictive Imputer class and edited the code in the fit method to print new_Imputed and and print X-trans = imputer.fit(X).transform(X.copy()) here is the result:
New Imputed From Fit Alone: [[-0.75511697 -0.56000831 -0.17324395 ... -0.28197847 -0.75023925
   1.57648847]
 [ 1.57128414  2.01646626 -0.4169594  ... -0.98846385  0.74134789
  -1.37396475]
 [-0.07996254  1.47008046 -0.70736089 ... -0.32923566  2.85404741
  -0.74426667]
 ...
 [ 1.31779515  1.58276785 -0.75654577 ... -0.72755881  1.08314364
  -0.58445367]
 [-0.13867505 -1.01695036 -0.97534895 ...  0.69337525  1.10940039
   0.32357511]
 [-1.29000647 -1.51285039 -0.37142956 ...  1.61670097 -1.3070204
   1.06002442]]
62.30102516297484
363.3446210619877
X-trans: [[-0.75511697 -0.56000831 -0.0921431  ... -0.28197847 -0.75023925
   1.61018726]
 [ 1.60153873  2.01646626 -0.44672337 ... -0.98846385  0.74134789
  -1.37396475]
 [-0.07996254  1.47008046 -0.70736089 ... -0.29503304  2.85404741
  -0.74426667]
 ...
 [ 0.00804308  1.40770558 -0.80077326 ... -0.48625397  1.08314364
  -0.58445367]
 [-0.13867505 -1.01695036 -0.80506119 ...  0.69337525  1.10940039
   0.32357511]
 [-1.19984082 -1.51285039 -0.37142956 ...  1.61670097 -1.3070204
   1.06002442]]
Some elements are similar but obviously some elements vary which was why I was wondering if you could explain the transform method

Thank you so much for doing this
Ochiba

[oattah1]

Also in the pseudocode of missForest paper, they created a vector of sorted indices of columns in X with respect to increasing amount of missing values; however your variable most_by_nan sort with respect to decreasing amount of missing values, could you explain why you did.
Thank you so much
Ochiba

[oattah1]

Also could the n_estimators be changed to 100 instead of 50 like the default in missForest