loss_val = floss(outputs_Sv.squeeze(),
targets_Sv) # Current loss
loss_val_mean += loss_val.item()
loss_val_mean = loss_val_mean / ValInstances
endtime = time.time()
print('Test time is %f' % (endtime - starttime))
log.write('Test time is %f\n' % (endtime - starttime))
#Observe results
if plot and ((epoch + 1) % plotT == 0 or epoch == 0):
[xtr, ytr, ptr, xval, yval, pval] = conter.torch2np(
[D[ii], S[ii], outputs_S, Dv[jj], Sv[jj], outputs_Sv],
formlist)
player.plotmat([
xtr[:, :, frame], ytr[:, :, frame], ptr[:, :, frame],
xval[:, :, frame], yval[:, :, frame], pval[:, :, frame]
],
tit=['xtr', 'ytr', 'ptr', 'xval', 'yval', 'pval'],
supt='Epoch{%d/%d}' % (epoch + 1, num_epochs))
plt.pause(0.1)
lossmean_vec[epoch] = loss_mean
lossmean_val_vec[epoch] = loss_val_mean
# Print loss
if (epoch + 1) % 1 == 0: # % 10
print('Epoch [%d/%d], Lossmean: %.5e, Validation lossmean: %.5e'\
%(epoch+1,num_epochs,loss_mean,loss_val_mean))
log.write('Epoch [%d/%d], Lossmean: %.5e, Validation lossmean: %.5e\n'\
%(epoch+1,num_epochs,loss_mean,loss_val_mean))
if loss.item() > 100:
log.write('saved at [iteration%d/%d,epoch%d/%d]\n'\
%(niter+1,int(TrainInstances/BatchSize),
epoch+1,num_epochs))
torch.save(net.state_dict(),
"Results/%s_%sCPU_Res3dC_Model_Tr%s_epoch%s_lr%.2e.pkl"\
%(ProjectName,prefix,TrainInstances,num_epochs,learning_rate))
minloss = min(loss_val_mean, minloss)
#Observe results
if plot and ((epoch + 1) % plotT == 0 or epoch == 0):
[xtr, ytr, ptr, xval, yval, pval] = conter.torch2np(
[D[ii], S[ii], outputs_S, Dv[jj], Sv[jj], outputs_Sv],
formlist)
fig=player.plotmat([xtr[:,:,frame],ytr[:,:,frame],ptr[:,:,frame],
xval[:,:,frame],yval[:,:,frame],pval[:,:,frame]],
tit=['xtr','ytr','ptr','xval','yval','pval'],
supt='{iteration%d/%d,epoch%d/%d}'\
%(niter+1,int(TrainInstances/BatchSize),epoch+1,num_epochs))
plt.pause(0.1)
inst = 0
loss_mean = 0
"""Save logs, prediction, loss figure, loss data, model and settings """
#Graphs
#Save the prediction figure
plt.savefig("Results/%s_%sCPU_Res3dC_Pred_Tr%s_epoch%s_lr%.2e.png"\
%(ProjectName,prefix,TrainInstances,
num_epochs,learning_rate))
#MSE
fig = plt.figure()
log.write('Test time is %f\n' % (endtime - starttime))
#Observe results
if plot and ((epoch + 1) % plotT == 0 or epoch == 0):
[xtr, ystr, pstr, xval, ysval, psval] = conter.torch2np(
[D[ii], S[ii], outputs_S, Dv[jj], Sv[jj], outputs_Sv],
formlist)
[yltr, pltr, ylval,
plval] = conter.torch2np([L[ii], outputs_L, Lv[jj], outputs_Lv],
formlist)
player.plotmat([
xtr[:, :, frame], ystr[:, :, frame], yltr[:, :, frame], None,
pstr[:, :, frame], pltr[:, :, frame], xval[:, :, frame],
ysval[:, :, frame], ylval[:, :, frame], None,
psval[:, :, frame], plval[:, :, frame]
],
tit=[
'xtr', 'ystr', 'yltr', None, 'pstr', 'pltr',
'xval', 'ysval', 'ylval', None, 'psval', 'plval'
],
supt='Epoch{%d/%d}' % (epoch + 1, num_epochs))
plt.pause(0.1)
lossmean_vec[epoch] = loss_mean
lossmean_val_vec[epoch] = loss_val_mean
exp_L, exp_S = net.getexp_LS()
exp_vec_L[epoch, :] = exp_L
exp_vec_S[epoch, :] = exp_S
# Print loss
if (epoch + 1) % 1 == 0: # % 10
%(ProjectName,prefix,ALPHA,TrainInstances,
num_epochs,learning_rate))
minloss=min(loss_val_mean,minloss)
#Observe results
if plot and ((epoch+1)%plotT==0 or epoch==0):
[xtr,ystr,pstr,xval,ysval,psval]=conter.torch2np([D[ii],S[ii],outputs_S,
Dv[jj],Sv[jj],outputs_Sv],
formlist)
[yltr,pltr,ylval,plval]=conter.torch2np([L[ii],outputs_L,
Lv[jj],outputs_Lv],
formlist)
player.plotmat([xtr[:,:,frame],ystr[:,:,frame],yltr[:,:,frame],
None ,pstr[:,:,frame],pltr[:,:,frame],
xval[:,:,frame],ysval[:,:,frame],ylval[:,:,frame],
None ,psval[:,:,frame],plval[:,:,frame]],
tit=['xtr','ystr','yltr',None,'pstr','pltr',
'xval','ysval','ylval',None,'psval','plval'],
supt='Epoch{%d/%d,%d/%d}'\
%(niter+1,int(TrainInstances/BatchSize),epoch+1,num_epochs))
plt.pause(0.1)
inst=0
loss_mean=0
"""Save logs, prediction, loss figure, loss data, model and settings """
#Graphs
#Save the prediction figure
plt.savefig("Results/%s_%sCPU_Unfolded_Pred_al%.2f_Tr%s_epoch%s_lr%.2e.png"\
%(ProjectName,prefix,ALPHA,TrainInstances,
num_epochs,learning_rate))