1

I am now using FANN to do incremental learning. Would anyone let me know whether my program is wrong or not? Thank you.

I have a data set to train. But in future I will get some new data set. I wanna incrementally train the current nn with the new data set, called "incremental learning".

I firstly create and train a nn with old data "old.data". I also set the training algorithm.

struct fann *ann = fann_create_standard(num_layers, num_input, num_neurons_hidden, num_output); fann_set_activation_function_hidden(ann, FANN_SIGMOID_SYMMETRIC); fann_set_activation_function_output(ann, FANN_SIGMOID_SYMMETRIC);
fann_set_training_algorithm(ann, FANN_TRAIN_INCREMENTAL);
fann_train_on_file(ann, "old.data", max_epochs, epochs_between_reports, desired_error); fann_save(ann, "mynn.net");
fann_destroy(ann);

Then when I have new data set "new.data", I think I can program like this:

struct fann *ann = fann_create_from_file("mynn.net");
fann_train_on_file(ann, "new.data", max_epochs, epochs_between_reports, desired_error);

Is my program correct?

tshepang
  • 12,111
  • 21
  • 91
  • 136
user2807986
  • 11
  • 3

1 Answers1

0

No

if you train with new data maybe lost old.data's experience

Birol Kuyumcu
  • 1,195
  • 7
  • 17