search

buriburisuri / sugartensor

A slim tensorflow wrapper that provides syntactic sugar for tensor variables. This library will be helpful for practical deep learning researchers not beginners.
MIT License
372 stars 63 forks source link

[log of emails]How to use placeholder and feed_dict way to train #26

Closed ryx19th closed 7 years ago

ryx19th commented 7 years ago

This is a copy of a conversation via emails. I guess it can benefit a little more to put it here as a log.

Hello Namju,

I'm a programmer from GitHub and developing some DL models based on your sugartensor. It has brought me much convenience and thank you a lot for it.

But there seems to be only one way to input data, that is putting them in queues. I know it could be faster compared to the traditional placeholder and feed_dict stuff. However, it would be embarrassing when dealing with really large data, because TensorFlow has set the limit of a single tensor as 2GB. I want to process a large pile of images now but all of them cannot be loaded at the same time.

So may I ask if it any possible to use the traditional placeholder and feed_dict way to input data in sugartensor? Or what should I modify to create such functions? Thank you.

Looking forward to your reply.

Exon

ryx19th commented 7 years ago

And the author replied as following.

You can use tf feed_dic as usual as it is. ( You can use any tensorflow functionality as it is with SugarTensor… ) Or you can use tensorflow queue as normal way. ( See, https://www.tensorflow.org/programmers_guide/threading_and_queues ) See my example : https://github.com/buriburisuri/speech-to-text-wavenet/blob/master/data.py

Thank you.