x_b=np.asarray(map(np.flipud,x)) pred = model.predictions(x,x_b,is_train) else: pred = model.predictions(x,is_train) # print("Prediction done....") if residual>0: if(minibatch_index==n_test_batches-1): pred = pred[0:(len(pred)-residual)] y=y[0:(len(y)-residual)] n_list=n_list[0:(len(n_list)-residual)] # du.write_predictions(params,pred,n_list) #u.write_pred(pred,minibatch_index,G_list,params) loss=np.nanmean(np.abs(pred -y))*2 (loss3d,l_list,s_list) =u.get_loss_bb(y,pred) #print(s_list) sq_loss_lst.append(s_list) loss_list=loss_list+l_list batch_loss += loss batch_loss3d += loss3d sq_loss_lst=np.nanmean(sq_loss_lst,axis=0) batch_loss/=n_test_batches batch_loss3d/=n_test_batches print "============================================================================" print sq_loss_lst s ='error %f, %f, %f,%f'%(batch_loss,batch_loss3d,n_test_batches,len(loss_list)) print (s) pu.plot_histograms(loss_list) pu.plot_error_frame(loss_list) #pu.plot_cumsum(loss_list)
for sp in range(seq_length): ls=full_lost[sp] #print len(ls) if(all==0): if(sp==0): final_loss[0:seq_length]=ls[0:seq_length] for index in range(seq_length,len(ls),seq_length): f_index=sp+index final_loss[f_index-7]=ls[index-7] else: print(sp) for index in range(0,len(ls)): f_index=sp+index final_loss[f_index]=final_loss[f_index]+ls[index] final_loss=final_loss[seq_length:len(final_loss)-seq_length] print len(final_loss) if(all==1): final_loss=[l/seq_length for l in final_loss] print len(final_loss) final_mean=np.mean(final_loss) s ='Final mean error %f'%(final_mean) print(s) pu.plot_histograms(final_loss) pu.plot_error_frame(final_loss) #pu.plot_cumsum(loss_list)
final_loss = [0] * (len(full_lost[0]) + seq_length) for sp in range(seq_length): ls = full_lost[sp] # print len(ls) if all == 0: if sp == 0: final_loss[0:seq_length] = ls[0:seq_length] for index in range(seq_length, len(ls), seq_length): f_index = sp + index final_loss[f_index - 7] = ls[index - 7] else: print (sp) for index in range(0, len(ls)): f_index = sp + index final_loss[f_index] = final_loss[f_index] + ls[index] final_loss = final_loss[seq_length : len(final_loss) - seq_length] print len(final_loss) if all == 1: final_loss = [l / seq_length for l in final_loss] print len(final_loss) final_mean = np.mean(final_loss) s = "Final mean error %f" % (final_mean) print (s) pu.plot_histograms(final_loss) pu.plot_error_frame(final_loss) # pu.plot_cumsum(loss_list)