Minjie Xu

Hi @dmhacker , kudos for putting all these together and for sharing your great work 👍 You've also got me started with my first ever Alexa programming experience :-) I've...

And btw, even after I've already started my own (free Dynos) Heroku server and modified it to download "short" videos only ([here](https://github.com/chokkyvista/dmhacker-youtube/commit/98e57efb071e85d21c8e7ad209e2803a01e80264#diff-b0347600e98ffad207d1311fb72f4f9cR23)), it's still quite common (almost always actually) for...

> Fetch the audio as a link directly from YouTube or some third-party provided. Just fyi, it seems like [this](https://github.com/ndg63276/alexa-youtube/blob/aa9d9aa15553de3206b6233117d9ffc475d024fd/lambda_function.py#L474) is doing exactly that using [pytube](https://github.com/nficano/pytube)?

(no code-change in the above [force-push](https://github.com/bastings/interpretable_predictions/compare/1307d75f3b76e9b0f44f737e9b0af36348d4ce80..85e416631c61337a2aa413ce9ed7795d0d976951), only switching `user.email` to match my github account)

I'm also thinking maybe a smarter initialization, e.g. letting the HardKuma layer output the expected selection rate right from the beginning, would help with stabilize the training as well? Did...

I also find myself unable to replicate https://cloud.google.com/tpu/docs/profile-tpu-vm#profile_tab in order to capture profiles on **TPU VMs** (TPU nodes work fine as @lackhole noted). In my case, the Tensorboard web UI...

As it turns out, the `localhost:6000 returned Unavailable: failed to connect to all addresses` error above was due to me forgetting to start the TF profiler server, which can be...