Live loss plot for training models in Keras (see: https://github.com/stared/livelossplot/ for a library)

live_loss_plot_keras.ipynb

Here's a version with loss and accuracy:

class PlotLearning(Callback):
    def on_train_begin(self, logs={}):
        self.i = 0
        self.x = []
        self.losses = []
        self.val_losses = []
        self.acc = []
        self.val_acc = []
        self.fig = plt.figure()
        
        self.logs = []

    def on_epoch_end(self, epoch, logs={}):
        
        self.logs.append(logs)
        self.x.append(self.i)
        self.losses.append(logs.get('loss'))
        self.val_losses.append(logs.get('val_loss'))
        self.acc.append(logs.get('acc'))
        self.val_acc.append(logs.get('val_acc'))
        self.i += 1
        f, (ax1, ax2) = plt.subplots(1, 2, sharex=True)
        
        clear_output(wait=True)
        
        ax1.set_yscale('log')
        ax1.plot(self.x, self.losses, label="loss")
        ax1.plot(self.x, self.val_losses, label="val_loss")
        ax1.legend()
        
        ax2.plot(self.x, self.acc, label="accuracy")
        ax2.plot(self.x, self.val_acc, label="validation accuracy")
        ax2.legend()
        
        plt.show();
        
plot = PlotLearning()

I also switched loss to a log plot since it tends to resolve that way

Works great, thanks!

Thanks for this, it's really nice!

Do you have a way to change the figure size? I'd like it to be larger but something like figsize=(20,10) doesn't work.

In the meantime, I wrote it for another project: https://github.com/deepsense-ai/intel-ai-webinar-neural-networks/blob/master/live_loss_plot.py (supporting multiple loss functions).

In action: https://github.com/deepsense-ai/intel-ai-webinar-neural-networks/blob/master/notebook_done.ipynb

UPDATE: I created a dedicated library: https://github.com/stared/livelossplot/ - contributions are welcomed! :)

You can install it with pip:

pip install livelossplot

Hi, I'd like to know if it is possible to plot loss curves with respect to iteration number in Keras? Thanks!

@ckolluru you can create the above using your own custom callback but in terms of granularity, it looks like Keras supports down to at most a batch level.

how do we add our own model to this?

Thanks for the great project! I was wondering why is my figure show blocking the training? it seems I should close the figure every iteration to let it run and show the updated results. The training does not continue unless I close the figure.

Thanks a lot

Thanks!

Thanks!

Thanks!

Thanks for the great project! I was wondering why is my figure show blocking the training? it seems I should close the figure every iteration to let it run and show the updated results. The training does not continue unless I close the figure.

was your problem solved? I have the same problem

This script is no longer being maintained (for the last year!).
Please use https://github.com/stared/livelossplot/ instead.

This script is no longer being maintained (for the last year!).
Please use https://github.com/stared/livelossplot/ instead.

I'm actually using that , but I have the same problem with that. the figure blocks the process and every epoch I have to close it for the build to continue ...

Thanks a lot it saved my day!
I simply made log-log scale to display th loss ... since I hope the loss decreasing by several orders of magnitude !

is it possible to get the live plot per batch instead of per epoch?

Hi, I have the same problem than @socca. Could somebody help please?

This is just fantastic. Thank you.

@ElaheMrz you need to use Jupyter for that.

Awesome work, it works very well!
Thanks for sharing :)