get model params while training

[anshdavid]

Description

I am using SGDRegressor for training a model. Is there any way i can store the Bias and the Norm values for each iteration in the training loop ?

Using verbose i can see the output , I just want to store the parameters in a list. I was trying to run the regressor in a for loop as it returns an instance of itself, but that did'nt work

clf = SGDRegressor(max_iter=1, alpha=alpha)                                                                                              
for counter in range (0,epochs):
    clf.fit(X, Y)
    print("slope {0}, intercept {1}".format(clf.coef_, clf.intercept_))

Can someone help me out ?
thanks....

[glemaitre]

This could be linked to a callback mechanism which was already a feature request which we don't have implemented for the moment. A potential discussion can be found in #10973

[anshdavid]

Thank you for the reply @glemaitre . I figured out a way of doing it . The result is off by a few decimal places though ...

clf = SGDRegressor(max_iter=1, alpha=0.01)
coef=0
intercpt=0
for counter in range (0,100):

    clf.fit(X, Y, coef_init=coef, intercept_init=intercpt )
    coef = clf.coef_
    intercpt = clf.intercept_
    intercpts.append(intercpt)
    slopes.append(coef)

[amueller]

you should be using partial_fit

[glemaitre]

you should be using partial_fit

I was not fast enough to mention that. Basically, max_iter=1 in this case and you don't need to pass the coef_init and intercept_init.

[anshdavid]

@amueller , @glemaitre thank you . got it