Tensorboard fails to plot model weights for all epochs during training

TensorFlow Special Interest Groups TensorBoard
1

I am trying to plot the progression of my model’s weights during training using add_scalar, however, the tensorboard plot only shows the weights for any one epoch.

What I mean by this is, when I load tensorboard, I only see “epoch 0” in the scalars section even if I have run my model for 10 epochs. However, I dont have this issue while plotting histograms in the same code.

My code is as follows:

for epoch in total_epochs:
          train model
          calculate error
          optimizer.step()
          calculate average loss
          
          for name,weight in model.named_parameters():
              SummaryWriter.add_histogram(name, weight, epoch)
              SummaryWriter.add_scalar(str(name), weight, epoch)

Here is an example of what I mean. I had run the model for 10 epochs, the graph only shows epoch 0 and 1. However, the histogram (not pictured) contains the progression of all 10 epochs.

3

For custom training loops and TensorBoard, have you tried the method described in Using TensorBoard with other methods (tf.summary)?