As mentioned by @bytesnake in #7, the reqwest crate is fairly heavy in its dependency requirements. To get the same functionality, I've replaced with a combination of curl and pbr (an ascii progress bar crate) for visualizing download progress.
Furthermore, since the standard BASE_URL used by the crate at http://yann.lecun.com/exdb/mnist/ has often ended up being down, I've added a new function to the Builder struct that allows passing a custom base url. I've tested this on cmoran.xyz (with the associated .gz filenames on the end), which seems to work fine. As a result, the datasets that have been traditionally included in this repository have been removed, and any data downloaded to data/* should be ignored by git.
Finally, since rulinalg is no longer maintained and the scientific computing community has seemed to center more around ndarray, I've replaced that dependency along with an example of how to convert the MNIST data into an Array3 data structure, like one might use for testing a neural network.
Happy to answer any questions or discuss changes which might make sense.
As mentioned by @bytesnake in #7, the
reqwest
crate is fairly heavy in its dependency requirements. To get the same functionality, I've replaced with a combination of curl and pbr (an ascii progress bar crate) for visualizing download progress.Furthermore, since the standard
BASE_URL
used by the crate at http://yann.lecun.com/exdb/mnist/ has often ended up being down, I've added a new function to the Builder struct that allows passing a custom base url. I've tested this on cmoran.xyz (with the associated .gz filenames on the end), which seems to work fine. As a result, the datasets that have been traditionally included in this repository have been removed, and any data downloaded todata/*
should be ignored by git.Finally, since rulinalg is no longer maintained and the scientific computing community has seemed to center more around
ndarray
, I've replaced that dependency along with an example of how to convert the MNIST data into anArray3
data structure, like one might use for testing a neural network.Happy to answer any questions or discuss changes which might make sense.