cfg.TRAIN.BATCH_SIZE,
thresh=thresh,
visualize=use_visdom)
for i in range(len(imdb.classes)):
cls_name = imdb.classes[i]
cur_ap = aps[i]
logger.scalar_summary('{}_AP'.format(cls_name), cur_ap, step)
vis.line(X=np.array([step]),
Y=np.array([np.mean(aps)]),
win=vis_map_window,
update='append')
# tensorboard + weights + grads
if step % 2000 == 0 and step > 0:
for name, weights in net.named_parameters():
tag = name.replace('.', '/')
logger.histo_summary(tag, weights.data.cpu().numpy(), step)
logger.histo_summary(tag + "/gradients",
weights.grad.data.cpu().numpy(), step)
#TODO: Perform all visualizations here
#You can define other interval variable if you want (this is just an
#example)
#The intervals for different things are defined in the handout
if visualize and step % vis_interval == 0:
#TODO: Create required visualizations
if use_tensorboard:
print('Logging to Tensorboard')
data_log.scalar_summary(tag='train/loss', value=loss, step=step)
if use_visdom:
else:
vis.line(X=torch.ones((1)).cpu() * step,
Y=torch.Tensor([loss]).cpu(),
win=loss_window,
update='append')
#print(hist_interval)
if step % hist_interval == 0:
#Get Histograms here
print("Getting Histogram")
if use_tensorboard:
state_dict = net.state_dict()
for k, v in state_dict.items():
if ('weight' in k):
writer.add_histogram('Weights_' + str(k), v, step)
for name, param in net.named_parameters():
if (param.requires_grad and param.grad is not None):
writer.add_histogram('Gradients_' + str(name),
param.grad, step)
'''for name,param in net.named_parameters():
#pdb.set_trace()
if(param.requires_grad):
#pdb.set_trace()
writer.add_histogram(name, param.detach().cpu(), step)'''
# Save model occasionally
if (step % cfg.TRAIN.SNAPSHOT_ITERS == 0) and step > 0:
save_name = os.path.join(
output_dir, '{}_{}.h5'.format(cfg.TRAIN.SNAPSHOT_PREFIX, step))
network.save_net(save_name, net)
print('Saved model to {}'.format(save_name))
#TODO: Perform all visualizations here
#You can define other interval variable if you want (this is just an
#example)
#The intervals for different things are defined in the handout
if visualize and step % vis_interval == 0:
#TODO: Create required visualizations
# logger.scalar_summary("train/loss",loss.data[0],step)
if use_tensorboard:
print('Logging to Tensorboard')
logger.scalar_summary("train/loss", loss.data[0], step)
if use_visdom:
print('Logging to visdom')
if visualize and step % grad_interval == 0:
for tag, params in net.named_parameters():
if params.grad is None:
continue
tag = tag.replace('.', '/')
weights = params.data
gradients = params.grad.data
logger.hist_summary(tag, weights, step)
logger.hist_summary(tag + '/grad', gradients, step)
# Save model occasionally
if (step % cfg.TRAIN.SNAPSHOT_ITERS == 0) and step > 0:
save_name = os.path.join(
output_dir, '{}_{}.h5'.format(cfg.TRAIN.SNAPSHOT_PREFIX, step))
network.save_net(save_name, net)
print('Saved model to {}'.format(save_name))