129 - What are Callbacks, Checkpoints and Early Stopping in deep learning (Keras and TensorFlow)

[@alankrit_mishra]

03:54 - Model checkpoint
06:35 - Early Stopping
08:00 - CSV_logger

[@debatradas9268]

thanks bro

[@sahartaheri1032]

Many thanks for all the valuable video, I am glad finally I found someone who can explain everything from scratch.

[@g.barios5805]

here working on my final thesis of my masters, you helped me! thanksssss🙏🏼🙏🏼

[@bhaskartripathi]

I used checkpoints in SDXL and came back to your video

[@achrafkmout9398]

your series are GOLD

[@muhammadhollandi2586]

Your video really helps my brother, I hope you always get blessings from God, keep up the spirit!

[@DigitalSreeni]

Glad I could help

[@Kyrios_X]

Hi, great video ! Using EarlyStopping or Modelcheckpoint on training causes an "ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all() " in line "if self.monitor_op(current - self.min_delta, self.best):". My input and output are tf.dataset with batch_size=14. I tried to modify on_epoch_end function of early stopping/modelcheckpoint code but I still have the same error. Do you have any idea to suggest please ?

[@shashireddy7371]

Thanks Sreeni

[@burakkahveci4123]

Thank you for this quite helpful video. :)

[@DigitalSreeni]

You're very welcome!

[@ekremguzelyel]

Is there a way to continue training after saving a checkpoint? One problem I have with online editors like Google Colab is that the sessions frequently go out of time in the middle of training. Can I just load the model from a checkpoint, and re-run training from that point? Thanks!

[@DigitalSreeni]

Just load the last saved model and continue by doing model. fit.
When you load a saved model all weights get loaded which makes it the starting point for future training.

[@ekremguzelyel]

@@DigitalSreeni Thank you. Then I assume the best way is to stop your training periodically, save your model, and continue. Do you think it's feasible?

[@rore3801]

if anyone gets an error in the file replace val_acc with val_accuracy in the file and Checkpoint.

[@DigitalSreeni]

Thanks. It depends on how 'history' stores the metrics.

[@lakshmireddy8551]

Very clear explanation... Thank you!!!

[@kenanmorani9204]

If you don't finish training but you finish one epoch of training, how would you resume training from where it stopped or at least from last full epoch after runtime error? Thanks

[@DigitalSreeni]

Please watch my video 131 (ruclips.net/video/4umFSRPx-94/видео.html)

[@mums2109]

Thank you for this video!

[@my_master55]

Thank you! Very clear explanation 😊

[@afriquemodel2375]

Save model parameters on each checkpoint how i can do that???

[@gursimransekhon688]

Hist= model.fit()
Hist.history["loss"] For model loss
Hist.history["accuracy' ] For model accuracy
Try this

[@briskminded9020]

great description

[@DigitalSreeni]

Thanks!

[@lendixful7932]

What GPU do u use ?

[@haraldtopfer5732]

Nvidia Quadro 4000 I think

[@aomo5293]

Please have y made before a video whare you showed how to work with local images.

[@debatradas9268]

thank you so much

[@lendixful7932]

Nice video 👍🏻👍🏻

[@jairotorregrosa]

Thank you!

[@DigitalSreeni]

Welcome!

[@zirghamilyas4537]

You didn't tell, how to use checkpoints later.

[@DigitalSreeni]

I explained the process of using checkpoint starting 3:59 in the video.