I have used Keras regressor to fit a regression to data. I used Scikit-learn wrapper and Pipeline to first standardize the data and then fit it on Keras regressor. Sort of like this:
from sklearn.grid_search import GridSearchCV
from keras.models import Sequential
from keras.layers import Dense
from keras.wrappers.scikit_learn import KerasRegressor
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.cross_validation import cross_val_score
from sklearn.cross_validation import KFold
from sklearn.externals import joblib
import cPickle
import pandas as pd
import os
from create_model import *
estimators = []
estimators.append(('standardize', StandardScaler()))
estimators.append(('mlp', KerasRegressor(build_fn=create_model, nb_epoch=50, batch_size=5, verbose=0, neurons = 1)))
pipeline = Pipeline(estimators)
I then Grid Search for best fit through GridSearchCv and get a best fit in a variable:
batch_size = [60, 80, 100, 200]
epochs = [2, 4, 6, 8, 10, 50]
neurons = np.arange(3,10,1)
optimizer = ['sgd', 'adam', 'rmsprom']
activation = ['relu', 'tanh']
lr = [0.001, 0.01, 0.1]
param_grid = dict(mlp__neurons = neurons, mlp__batch_size = batch_size, mlp__nb_epoch = epochs, mlp__optimizer = optimizer, mlp__activation = activation, mlp__learn_rate = lr)
grid = GridSearchCV(estimator=pipeline, param_grid=param_grid, cv = kfold,scoring='mean_squared_error')
grid_result = grid.fit(X, Y)
clf = []
clf = grid_result.best_estimator_
The clf variable has 2 process as defined in pipeline. My question is how do I extract weights and biases of keras regressor for best fit (clf), through get_params function?:
clf.get_params()
I couldn't find good documentation for this.
