implement parameters (for now ignored): verbose, nesterovs_momentum
(!) Initialize weights by random values near zero
(!) Momentum
(!) Divide the gradient by n_samples
(!) stop training after an excessive number of epochs
(!) shuffle the training patterns.
(!) Try a number of random starting configurations (e.g. 10 or more training runs or trials).
(!) Check the learning curve
(!) table and plots
(!) Regularization
implemented parameters: hidden_layer_sizes, hidden_activation, output_activation, alpha, batch_size, max_iter, shuffle, warm_start, momentum, loss, solver, random_state, learning_rate, learning_rate_init, power_t, tol, n_iter_no_change, early_stopping, validation_fraction
ignored parameters: beta_1, beta_2, epsilon, max_fun