ericmjl / dl-workshop Goto Github PK

jupyter labextension install @jupyter-widgets/jupyterlab-manager and related blocks are unclear as to which is needed for which use-case, jupyter notebook vs jupyter-lab, and so may accidentally be run twice by tutorial attendees.

tl;dr version to copy/paste into a corrected version of a notebook:

def vmapped_func(array):
    result = []
    for element in array:
        result.append(func(element))
    result = np.stack(result)
    return result

As per title

I could not find a license for these materials. Did I miss it? If not, would you mind adding a license to know how we can (re)use the tutorial materials? Thanks for writing this!

As per title. Either decide to set it as an exercise or set it as an example.

For the DP-GMM model, we should be able to avoid singularities by choosing an Inverse Gamma prior on the sigma parameter, I think.

In particular, this figure? https://github.com/ericmjl/dl-workshop/blob/master/docs/figures/infographic.pdf

As a part of my thesis' introduction section I've included a background on neural networks and CNNs, and having a figure that explains the architecture in three different ways would be helpful to the reader.

As per title.

From Michael Tullius (@mvtullius), who came by on SciPy 2021.

Hi Eric, Thank you for your presentation yesterday and for making all the materials available online. I would like to offer my feedback, for what its worth. I felt like I followed your presentation fairly well, but I only was successful with the easier examples. I even spent a couple hours afterwards going through the material on my own, but the more complex examples eluded me. I have had a little prior exposure to this type of programming, but it is still a very large shift in thinking that my brain really hasn't transitioned to. So my suggestions would be to maybe have some intermediate level examples in between the easier and harder examples, and probably more importantly, to have more demonstration of the process you use in mapping code that uses for loops to code that uses lax.scan and/or vmap. It was pretty easy for me to translate the simpler examples from for loops to code without loops, but I was stuck on the harder examples. I even tried making the dataframes in the examples much simpler and looking at the output of various operations to try to visualize what I was doing, but in the end I still struggled. Anyway, I may just need to spend MUCH more time with these concepts if I truly want to be competent with them, but I wanted to give some feedback that might be helpful for you for future presentations. Best Regards, Michael

Key actions that can help for the future:

1. Introduce illustrations of how to map from loopy code to non-loopy code. (Some repetition may be needed here.)
2. Introduce exercises of intermediate complexity.

This would be a great extension to the workshop, as per comment from Andy Long.