Utilizing the documentation in the correct way

Hey there,

Even though I have been in deep learning development for quite some time, I could not get used to the documentation as I should be I suppose since I am getting some hard times finding whether something I need exists in the library and so on… I assume one reason is that the usage addresses more REPL types of programmers and I am the opposite :slight_smile:. So, for instance, while I was dealing with creating a DataBlock for MNIST set, I found that Normalize could have mnist_stats by a total chance and could not coincide anything with that in the documentation. This kind of stuff also happens when I look for appropriate classes or functions for different kinds of tasks.

So, this is more like asking for an opinion and guidance rather than a question that firstly, whether the documentation fully covers everything that the library could do. Secondly, whether there are other ways to utilize the library’s potential than the documentation.

I’d like to mention that I am really amazed by how the library really makes constructing the steps needed in a regular PyTorch setup. Therefore, I would really appreciate any guidance. Thanks in advance!


Fast.ai can be confusing. First of all there are multiple versions out there, (depends on the version of the course you’re taking, or if you’re using fastbook), second the library is constantly evolving and the people doing the documentation keep up as best they can. But the real answer is a lot less satisfying. “Read the code Luke” If you want to know what something does use the ?? feature of jupyter notebook to find the module and read the code. If you find a problem with the documentation or wish to document some undocumented feature, you can look here for information on how to contribute.