Sklearn library and algorithms

AI Discussions
,
1

I was exploring this link

In particular I was trying the LinearSVR() to improve the R^2 value of the California Housing dataset model.

I initially tried these lines of code
regression_pipeline = Pipeline([

('scaler', StandardScaler()),

('LinearStandardVectorRegressor', LinearSVR())

])

y_pred = regression_pipeline.predict(X_test)

r2_score( y_test, y_pred)

At first I got this warning ‘ConvergenceWarning: Liblinear failed to converge, increase the number of iterations.’

To resolve it I changed the regression_pipeline to
regression_pipeline = Pipeline([

('scaler', StandardScaler()),

('LinearStandardVectorRegressor', LinearSVR(max_iter=10000))

])
So I couldn’t quite figure out why updating max_iter value to above a 1000 made it go away

The other thing was with a simple linear regression the R^2 squared value was 0.575787706032451 and with this LinearSVR it didn’t seem to improve and was 0.5542804765083618.

Any insight would be helpful

Any ideas

2

Likely this is a very slow convergence, due to relying too much on default parameter values. Try tweaking the input parameters.