'Grouped' k-fold CV

A quick demo by Matt

In cross-validating, we'd like to drop out one well at a time. LeaveOneGroupOut is good for this:



In [1]:

    
import pandas as pd
training_data = pd.read_csv('../training_data.csv')

Isolate X and y:



In [2]:

    
X = training_data.drop(['Formation', 'Well Name', 'Depth','Facies'], axis=1).values
y = training_data['Facies'].values

We want the well names to use as groups in the k-fold analysis, so we'll get those too:



In [3]:

    
wells = training_data["Well Name"].values

Now we train as normal, but LeaveOneGroupOut gives us the approriate indices from X and y to test against one well at a time:



In [4]:

    
from sklearn.svm import SVC
from sklearn.model_selection import LeaveOneGroupOut

logo = LeaveOneGroupOut()

for train, test in logo.split(X, y, groups=wells):
    well_name = wells[test[0]]
    score = SVC().fit(X[train], y[train]).score(X[test], y[test])
    print("{:>20s}  {:.3f}".format(well_name, score))









    



     CHURCHMAN BIBLE  0.498
      CROSS H CATTLE  0.313
            LUKE G U  0.460
               NEWBY  0.294
               NOLAN  0.383
          Recruit F9  0.662
             SHANKLE  0.477
           SHRIMPLIN  0.476



In [ ]: