edflow
edflow copied to clipboard
pesser
Reduce boilerplate code for your ML projects. TensorFlow and PyTorch.
I wanted to use the csv files generated for use with the `Stochastic Pair class` https://github.com/pesser/eddata/blob/8d18996a6d7d82ed6c6ffbe6c89b3d6d1c816f53/eddata/stochastic_pair/__init__.py#L18 But only use the much more simple `CsvDataset`. The problem is that the interface...
I realized that some modules might not be needed anymore, like https://github.com/pesser/edflow/tree/dev/edflow/nn We should just remove them.
Implement a tool, that lets a user see, what takes long and what doesn't in a given training cycle. This will allow us to use resources far more efficiently.
It is often beneficial to play around with a model, without the overhead of writing a loader script for some data you just want to see your model evaluated on....
Make it easy for edflow users to define, if they want to use gpus and if so which. This project should also include an easy to use resource detector, that...
It might be good to have an easy way to see if the data loading is a bottleneck of the GPU is a bottleneck. In a second stage, it might...
the related project pytorch-lightning implements an easy way to just scale the project onto multiple GPUs. I think we should adapt it somehow, at least when working on a single...
- missed a couple of lines which still talk about previous commandline args: https://github.com/pesser/edflow/blob/9100ca7f698e75457ab89fd40aacee19cda54cac/edflow/eval/pipeline.py#L62 https://github.com/pesser/edflow/blob/9100ca7f698e75457ab89fd40aacee19cda54cac/edflow/eval/pipeline.py#L90 https://github.com/pesser/edflow/blob/9100ca7f698e75457ab89fd40aacee19cda54cac/edflow/eval/pipeline.py#L102 https://github.com/pesser/edflow/blob/9100ca7f698e75457ab89fd40aacee19cda54cac/edflow/eval/pipeline.py#L377 - should document properly what can be returned in eval op, how it...
The example in the DatasetMixin docstrings does not show that labels are appended under 'labels_'
**Is your feature request related to a problem? Please describe.** Sometimes only one datset ist required. The implementation in the dev branch however requires a separate definition of a train...