Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.
And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.
Faster Tabular Dataloader
See original GitHub issueFeature request
What is the expected behavior?
Load data more faster than use def __getitem__(self, index): x, y = self.x[index], self.y[index] return x, y
What is the motivation or use case for adding/changing the behavior? Fix the CPU to GPU data-loading bottleneck.
How should this be implemented in your opinion? Already have an implemented. https://github.com/hcarlens/pytorch-tabular/blob/master/fast_tensor_data_loader.py
Are you willing to work on this yourself? No
Issue Analytics
- State:
- Created 3 years ago
- Comments:7
Top Results From Across the Web
Top Related Medium Post
Top Related StackOverflow Question
Troubleshoot Live Code
Top Related Reddit Thread
Top Related Hackernoon Post
Top Related Tweet
Top Related Dev.to Post
Top Related Hashnode Post
I didn’t test with the Tabnet, but with some small models, it’s indeed can speed up 10~20 times. Some other people’s benchmark: https://towardsdatascience.com/better-data-loading-20x-pytorch-speed-up-for-tabular-data-e264b9e34352
@Optimox Just curious, what is the size of the dataset being tested? Is it possible that the size is too small so that we can’t see the improvement?