Named dimensions in type-hint shapes - Githubissues

ramonhagenaars / nptyping

💡 Type hints for Numpy and Pandas

MIT License

588 stars 29 forks source link

Named dimensions in type-hint shapes #24

Closed petered closed 4 years ago

petered commented 4 years ago

I find it would be really nice to assign names to dimensions, e.g. NDArray[('H', 'W', 3), UInt8] ... and a more compact option: NDArray['H,W,3', UInt8]. This can serve two purposes:

Hint at the meaning of the dimensions

image: NDArray['H,W,3', UInt8] is clearly a (height x width) image with 3 color channels
input_data = NDArray['N,D', Float] is (using common conventions) an array of N D-dimensional data points.

Specify shape constraints across arguments:

def compute_losses(self, images: NDArray['N,H,W,3', UInt8], labels: NDArray['N', Int]) -> NDArray['N', Float] Would indicate that we expect that images.shape[0]==labels.shape[0]==return_value.shape[0]

ramonhagenaars commented 4 years ago

Hi @petered

This feature has been requested before and is "on my radar". Check out my 2nd comment in #12 which explains how I envision this ability. If you have anything to add or comment to that, please continue the discussion there.