I am trying to visualize neural network weights by adding weights to the histograms of tensorboard.
for name, weight in self.net.named_parameters():
self.writer.add_histogram(name, weight, global_step=step_id)
code itself runs without any errors, but in tensorboard I do not see the histograms(i don’t see any log).
I printed the name and weights to confirm they are not empty tensor.
Do you use flush()? Generally, file size increase would indicate success on pytorch side. On browser side, chrome-based browser may be needed, and density plots are on separate page.
from torch import nn
from torch.utils.tensorboard import SummaryWriter
net = nn.Linear(8,8)
writer = SummaryWriter(DIR) #,flush_secs=120
for name, weight in net.named_parameters():
writer.add_histogram(name, weight, global_step=1)
writer.flush() #to avoid delayed disk writes when inside a training loop