torchbox.summary package

Submodules

torchbox.summary.loss_log module

class torchbox.summary.loss_log.LossLog(plotdir=None, xlabel='Epoch', ylabel='Loss', title=None, filename=None, logdict=None, lom='min')

Bases: object

add(key, value)

assign(key, value)

get(key=None)

judge(key, n1=50, n2=10)

judge how to save weights

|____n1____|__n2__||

current epoch

If the average loss of the last n2 epochs is better than the average of the previous n1 epochs and the loss value of the current epoch is the best among the n2 epochs, then save the weights of current epoch with 'Average' flag. If the loss of current epoch is the best of all previous epochs, then save the weights of current epoch with 'Single' flag.

Parameters:

• key (str) – the key string of loss for judge.

• n1 (int, optional) – the number of latest epochs, by default 50

• n2 (int, optional) – the number of , by default 10

Returns:

• bool – The current is the best?

• str – 'Single' or 'Average'

plot(x=None, offset=0)

updir(plotdir=None)

Module contents