Added wrapper for simple model initialization

Author: czaloom

Source/embeddedML.c

@@ -1,5 +1,7 @@
 /* 
  * EMBEDDEDML v1.3b
+ * Lightweight Version
+ * - Optimized for resource constrained platforms.
  */
 
 /*
@@ -198,3 +200,118 @@ float relu2_derivative(float x){
  else if(x > 1.0) return 0.1;
  return 1.0;
 }
+
+//----Wrapper Functions-----
+
+void set_model_memory(ANN *model, float *weights, float *dedw, float *bias, float *output){
+ model->weights = weights;
+ model->dedw = dedw;
+ model->bias = bias;
+ model->output = output;
+}
+
+void set_model_parameters(ANN *model, unsigned int *topology, unsigned int nlayers, char activation_function){
+ model->topology = topology;
+ model->n_layers = nlayers;
+
+ int i;
+ int nweights = 0, nbias = 0;
+ for(i = 1; i < nlayers; i++){
+ nweights += topology[i]*topology[i-1];
+ nbias += topology[i-1];
+ }
+
+ model->n_weights = nweights;
+ model->n_bias = nbias;
+
+ switch(activation_function){
+ case 'r':
+ model->output_activation_function = &relu;
+ model->hidden_activation_function = &relu;
+ break;
+ case 'R':
+ model->output_activation_function = &relu2;
+ model->hidden_activation_function = &relu2;
+ break;
+ case 's':
+ model->output_activation_function = &sigmoid;
+ model->hidden_activation_function = &sigmoid;
+ break;
+ case 't':
+ model->output_activation_function = &tanhf;
+ model->hidden_activation_function = &tanhf;
+ break;
+ default:
+ model->output_activation_function = &relu;
+ model->hidden_activation_function = &relu;
+ break;
+ }
+}
+
+void set_model_hyperparameters(ANN *model, float learning_rate, float bias_learning_rate, float momentum_factor){
+ model->eta = learning_rate;
+ model->beta = bias_learning_rate;
+ model->alpha = momentum_factor;
+}
+
+void set_learning_rate(ANN *model, float eta){
+ model->eta = eta;
+}
+
+void set_bias_learning_rate(ANN *model, float beta){
+ model->beta = beta;
+}
+
+void set_momentum_factor(ANN *model, float alpha){
+ model->alpha = alpha;
+}
+
+void set_output_actfunc(ANN *model, char func){
+ switch(func){
+ case 'r':
+ model->output_activation_function = &relu;
+ model->output_activation_derivative = &relu_derivative;
+ break;
+ case 'R':
+ model->output_activation_function = &relu2;
+ model->output_activation_derivative = &relu2_derivative;
+ break;
+ case 's':
+ model->output_activation_function = &sigmoid;
+ model->output_activation_derivative = &sigmoid_derivative;
+ break;
+ case 't':
+ model->output_activation_function = &tanhf;
+ model->output_activation_derivative = &tanhf_derivative;
+ break;
+ default:
+ model->output_activation_function = &relu;
+ model->output_activation_derivative = &relu_derivative;
+ break;
+ }
+}
+
+void set_hidden_actfunc(ANN *model, char func){
+ switch(func){
+ case 'r':
+ model->hidden_activation_function = &relu;
+ model->hidden_activation_derivative = &relu_derivative;
+ break;
+ case 'R':
+ model->hidden_activation_function = &relu2;
+ model->hidden_activation_derivative = &relu2_derivative;
+ break;
+ case 's':
+ model->hidden_activation_function = &sigmoid;
+ model->hidden_activation_derivative = &sigmoid_derivative;
+ break;
+ case 't':
+ model->hidden_activation_function = &tanhf;
+ model->hidden_activation_derivative = &tanhf_derivative;
+ break;
+ default:
+ model->hidden_activation_function = &relu;
+ model->hidden_activation_derivative = &relu_derivative;
+ break;
+ }
+}

Source/embeddedML.h

@@ -50,6 +50,16 @@ void run_ann(ANN *net, float *input);
 void init_ann(ANN *net);
 void init_pretrained_ann(ANN *net);
 
+void set_model_memory(ANN *model, float *weights, float *dedw, float *bias, float *output);
+void set_model_parameters(ANN *model, unsigned int *topology, unsigned int nlayers, char activation_function);
+void set_model_hyperparameters(ANN *model, float learning_rate, float bias_learning_rate, float momentum_factor);
+
+void set_learning_rate(ANN *model, float eta);
+void set_bias_learning_rate(ANN *model, float beta);
+void set_momentum_factor(ANN *model, float alpha);
+void set_output_actfunc(ANN *model, char func);
+void set_hidden_actfunc(ANN *model, char func);
+
 //-----Utility-----
 void fill_zeros(float *v, unsigned int size);
 void fill_number(float *v, unsigned int size, float number);