TPU, Listing Embeddings, Pyception

Weekly Reading List #2

Source

Issue #2: 2018/04/23 to 2018/04/29

This is an experimental series in which I briefly introduce the interesting data science stuffs I read, watched, or listened to during the week. Please give this post some claps if you’d like this series to be continued.

Hands-on with the Google TPUv2

It’s been a while since Google made TPU available on their cloud platform in beta in February. I’m curious if there are already some people sharing their experience using TPU on the Internet. So I did some Googling…

The article above then led me to…

Taken from RiseML blob

So it seems the released TPU (TPUv2) is a bit more cost effective then . However, the fact that TPU supports only mixed precision training may become an issue sometimes.

The downside is that there are a lot of hoops to jump through to be able to use TPU, according to the Paperspace blog post. And some of them are quite intimidating. Also, only Tensorflow supports TPU so far.

Update on 2018/4/26

This new post by RiseML showed that TPU might be even more cost-effective than we thought, and the top-1 accuracy (on the validation set) is bit better coming from TPU than from GPU.

Airbnb Listing Embeddings

This post by Airbnb describes how they embeds every listing on their platform to improve similar listing recommendations and later real-time search personalization. Its is well-written and easy to read. The model evaluation parts are particularly interesting. The methodology should be applicable to other similarity problems, too.

Pyception

A freaking hilarious and surprisingly educational parody by Anaconda.

A Nice Telegram Channel about Data Science

(I’m not affiliated with the channel.) I’ve found the links posted in this channel relevant and informative: