Metadata-Version: 2.1
Name: edutorch
Version: 0.0.1
Summary: Rewritten PyTorch framework designed to help you learn AI/ML
Home-page: https://github.com/tyleryep/edutorch
Author: Tyler Yep
Author-email: tyep10@gmail.com
License: UNKNOWN
Description: # edutorch
        Rewritten PyTorch framework designed to help you learn AI/ML
        
        
        
        # Dynamic Shape-Checking / Type-Checking (Typorch)
        I'm interested in shape-checking for tensors. It doesnt need to be static type checking,
        having fancy asserts would be good enough.
        
        > (B, C, H, W)
        
        > (B, 1, H, W)
        
        > (B, H, W)
        
        If you run a program with the shape-checker, it automatically inserts assert statements that the left side of the variable assignment must have that shape. Letters are tracked throughout (e.g. a new letter introduces a new variable), and a number asserts that that dimension must match exactly.
        
        Tuple shapes maybe, to distinguish a shape comment from a regular comment.
        
        Using this mode, the code is compiled uniquely and increases runtime.
        
        Once you are confident with your shapes, you can simply run your program normally.
        
        # Goals
        1. Readability. Everything should make it immediately obvious how the layer or mmodel works on its own.
        2.
        
        
        No autograd - if you want a simple autograd implementation, check out Karpathy's micrograd repo.
        
        TODO:
        
        - Convert to dataclasses?
Platform: UNKNOWN
Classifier: Programming Language :: Python :: 3
Classifier: License :: OSI Approved :: MIT License
Classifier: Operating System :: OS Independent
Requires-Python: >=3.7
Description-Content-Type: text/markdown
