Nerlworker #286
Nerlworker #286
Conversation
… of layerSizesStr
…for highest dim found
There was a problem hiding this comment.
@Orisadek @GuyPerets106
You have ruined something - we can't merge this!
F1 Score is 0 it means that you messed up with basic perceptron!
Please check the cause:
I can blame dataset changes, train and predict functions in opennNNnif.cpp
| inputs_dimensions.setValues({num_of_samples, inputs_number}); | ||
|
|
||
| *calculate_res = neural_network->calculate_outputs(PredictNNptr->data->data(), inputs_dimensions); | ||
| if(neural_network->has_convolutional_layer()) |
There was a problem hiding this comment.
@Orisadek Why this code goes here? what its purpose?
There is post_predict and pre_predict for this
There was a problem hiding this comment.
@Orisadek Why this code goes here? what its purpose? There is post_predict and pre_predict for this
In "calculate outputs" the function change according to the rank of the input.
in CNN we need to enter rank == 4 to get result.
neural_network->has_convolutional_layer() check if there is convolutional layer in the NN
I don't know what is pre-predict, but I saw the post predict.
| std::shared_ptr<NerlLayer> prev_layer = curr_layer->get_prev_layer_ptr(); | ||
| int prev_layer_size = prev_layer->get_dim_size(DIM_X_IDX); | ||
| int prev_layer_size; | ||
| if(neural_network_ptr->get_layer_pointer(neural_network_ptr->get_layers_number()-1)->get_type_string() == "Flatten"){ |
There was a problem hiding this comment.
@Orisadek
You must check if this code somehow influences the full flow test
| @@ -16,30 +17,119 @@ namespace nerlnet | |||
| generate_opennn_neural_network(); | |||
| _training_strategy_ptr = std::make_shared<opennn::TrainingStrategy>(); | |||
| generate_training_strategy(); | |||
| //TODO Ori and Nadav - implement training strategy (loss method, optimizer type, epochs, only sgd and adam) | |||
| _ae_red_ptr = std::make_shared<AeRed>(); | |||
No description provided.