float16

[davek44]

I have imprecise floating points that I'd like to keep small. I made this change to the torch-hdf5 code, and it seems to work fine. Am I overlooking anything here?

[d11]

Hmm, I'm not sure actually. They will be read into a full-size float tensor, and I don't know without checking how safe the conversion of the small floats into large ones is, the way it is being done here. In the resulting float tensor, do you find the extra bits are set to zero? Or might they be uninitialized?
Also bear in mind that when writing the data out they will still be full floats, so this change prevents hdf5 -> Torch -> hdf5 being the identity map.
Apart from those concerns this seems fine though.

[googlebot]

Thanks for your pull request. It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

📝 Please visit https://cla.developers.google.com/ to sign.

Once you've signed, please reply here (e.g. I signed it!) and we'll verify. Thanks.

---

• If you've already signed a CLA, it's possible we don't have your GitHub username or you're using a different email address. Check your existing CLA data and verify that your email is set on your git commits.
• If you signed the CLA as a corporation, please let us know the company's name.