Getting progress updates from GridSearchCV with scikit-learn

Question

I am currently implementing a Support Vector Regression in Python, where I am estimating the parameters C and gamma through the GridSearchCV. I am initially searching from approximately 400 combinations of C and gamma. This is a very exhaustive search which has now been running for over an hour on my computer.

What I would like is to receive status updates e.g. how many combinations are left to test or similar, since at the moment it is hard to tell if the program is working or if it has just frozen or whatever is going on.

From what I have read on the sci-kit learn documentation I cannot seem to find any help with this. Is there a wrap-around?

Answer 1

GridSearchCV has a verbose= keyword. Try setting it to e.g. 100.

If you are using sklearn.cross_validation.cross_val_score to evaluate your model, you can also set its verbose= to a high level.

If you need more detail, there is also the possibility to "hack" the scoring object you want to use to make it print the score to the screen or to a file every time it is called, for example:

from sklearn.metrics.scorer import accuracy_scorer

def my_accuracy_scorer(*args):
    score = accuracy_scorer(*args)
    print('score is {}'.format(score))
    return score

Use this function as the scoring keyword in cross_val_score or GridSearchCV, by putting scoring=my_accuracy_scorer