[FIXED] Sklearn validation_curve : Which possibilities do I have for the "param_name" parameter

Issue

If I calculate the validation curve like follows:

PolynomialRegression(degree=2,**kwargs):
    return make_pipeline(PolynomialFeatures(degree),LinearRegression(**kwargs))

#...

degree=np.arange(0,21)

train_score,val_score=validation_curve(PolynomialRegression(),X,y,"polynomialfeatures__degree",degree,cv=7)

the function claims the parameter "param_name" which is the name of the parameter that will be evaluated ("param_range") In this case it is polynomialfeatures__degree. But:

1. From where do I get the string for "param_name" in general and from where do I get the string "polynomialfeatures__degree" in particular because I can't find this param_name in any documentation...Is the "function__param name" a common way to point on the param_name of a function??
2. For which estimators can I use the validation_curve function?

Solution

PART 1

Answer to your question # 1:

As param_name you can use one of the available parameters of the estimator of your choice.

For example: Let's say that you choose the estimator: LinearRegression. From this link you can see than the available parameters/arguments are fit_intercept=True, normalize=False, copy_X=True, n_jobs=1. So now you know what are the available parameters of the specific estimator.

Now, from this link we can see how to define the param_name and param_range.

Part of the example in this link:

param_range = np.logspace(-6, -1, 5)

train_scores, test_scores = validation_curve(
SVC(), X, y, param_name="gamma", param_range=param_range,
cv=10, scoring="accuracy", n_jobs=1)

EDIT 1:

For the "string" as you call it: if I understand right you want to do something like:

from sklearn.preprocessing import PolynomialFeatures

a= PolynomialFeatures(degree=2)
a.get_params()

result:

{'include_bias': True, 'interaction_only': False, 'degree': 2}

These are the available param_names for the PolynomialFeatures.

EDIT 2:

Another way to see the param_names is to go to the source code.

For example, here link you can see in lines 1149-1160 the strings of the parameters for the PolynomialFeatures estimator.

Finally, here link you can see in lines 421-444 the strings of the parameters for the LinearRegression estimator.

EDIT 3:

Other way to get and set the parameters of estimator in sklearn.

from sklearn.linear_model import LinearRegression

lr = LinearRegression()
lr.get_params(deep=True)

result:

{'copy_X': True, 'normalize': False, 'n_jobs': 1, 'fit_intercept': True}

Now that you know the parameters you create your parameters:

params = {  'copy_X': False,  'fit_intercept': True,  'normalize': True }

And then you can unpack them in the estimator:

lr_new = LinearRegression(**params)
lr_new

result:

LinearRegression(copy_X=True, fit_intercept=False, n_jobs=1, normalize=True)

PART 2:

Answer to your question # 2:

I think that in the case of sklearn module, you can use the validation_curve for all the estimators.

Hope this helps

Answered By - seralouk