49     MLPTrain *net;  in mlp_init()  local
51     net = malloc(sizeof(*net));  in mlp_init()
52     net->topo = malloc(nbLayers*sizeof(net->topo[0]));  in mlp_init()
54         net->topo[i] = topo[i];  in mlp_init()
57     net->in_rate = malloc((inDim+1)*sizeof(net->in_rate[0]));  in mlp_init()
58     net->weights = malloc((nbLayers-1)*sizeof(net->weights));  in mlp_init()
59     net->best_weights = malloc((nbLayers-1)*sizeof(net->weights));  in mlp_init()
62         net->weights[i] = malloc((topo[i]+1)*topo[i+1]*sizeof(net->weights[0][0]));  in mlp_init()
63         net->best_weights[i] = malloc((topo[i]+1)*topo[i+1]*sizeof(net->weights[0][0]));  in mlp_init()
77         net->in_rate[1+j] = .5/(.0001+std);  in mlp_init()
83             net->weights[0][k*(topo[0]+1)+j+1] = randn(std);  in mlp_init()
85     net->in_rate[0] = 1;  in mlp_init()
90             sum += inMean[k]*net->weights[0][j*(topo[0]+1)+k+1];  in mlp_init()
91         net->weights[0][j*(topo[0]+1)] = -sum;  in mlp_init()
101         net->weights[nbLayers-2][j*(topo[nbLayers-2]+1)] = mean;  in mlp_init()
103             net->weights[nbLayers-2][j*(topo[nbLayers-2]+1)+k+1] = randn(std);  in mlp_init()
105     return net;  in mlp_init()
111 double compute_gradient(MLPTrain *net, float *inputs, float *outputs, int nbSamples, double *W0_gra…  in compute_gradient()  argument
124     topo = net->topo;  in compute_gradient()
125     inDim = net->topo[0];  in compute_gradient()
126     hiddenDim = net->topo[1];  in compute_gradient()
127     outDim = net->topo[2];  in compute_gradient()
130     W0 = net->weights[0];  in compute_gradient()
131     W1 = net->weights[1];  in compute_gradient()
197     MLPTrain *net;  member
211     int *topo = arg->net->topo;  in gradient_thread_process()
223 …arg->rms = compute_gradient(arg->net, arg->inputs, arg->outputs, arg->nbSamples, arg->W0_grad, arg…  in gradient_thread_process()
230 float mlp_train_backprop(MLPTrain *net, float *inputs, float *outputs, int nbSamples, int nbEpoch, …  in mlp_train_backprop()  argument
243     topo = net->topo;  in mlp_train_backprop()
252     topo = net->topo;  in mlp_train_backprop()
253     inDim = net->topo[0];  in mlp_train_backprop()
254     hiddenDim = net->topo[1];  in mlp_train_backprop()
255     outDim = net->topo[2];  in mlp_train_backprop()
256     W0 = net->weights[0];  in mlp_train_backprop()
257     W1 = net->weights[1];  in mlp_train_backprop()
258     best_W0 = net->best_weights[0];  in mlp_train_backprop()
259     best_W1 = net->best_weights[1];  in mlp_train_backprop()
276             W0_rate[i*(inDim+1)+j] = rate*net->in_rate[j];  in mlp_train_backprop()
282         args[i].net = net;  in mlp_train_backprop()
465     MLPTrain *net;  in main()  local
468     net = mlp_init(topo, 3, inputs, outputs, nbSamples);  in main()
469     rms = mlp_train_backprop(net, inputs, outputs, nbSamples, nbEpoch, 1);  in main()
479         printf ("%gf,", net->weights[0][i]);  in main()
488         printf ("%gf,", net->weights[1][i]);  in main()