模型进度 Scikit Learn

Question

Is there an easy way to obtain predictions for any of the models in Scikit learn before they converge? I'm aiming to make animations of models as they train to observe their behavior.

Ideally, this would be similar to how one can extract results after an epoch of training for an ANN in Keras.

Answer 1

You may want to use the learning curve class for evaluating intermediate results.

Among the parameters, there's one called train_sizes that lets you specify the number of training examples that will be used to generate the curve. It can be both absolute, or relative to the available samples.

For instance, using linear regression:

from sklearn.linear_model import LinearRegression
from sklearn.model_selection import learning_curve
TRAIN_SIZES = [1, 100, 500, 2000, 5000]
train_sizes, train_scores, validation_scores = learning_curve(
                                                  estimator=LinearRegression(),
                                                  X=myX,
                                                  y=myY,
                                                  train_sizes=TRAIN_SIZES
                                               )