cephes.stdtr not increasing?!

[jucor]

Wow!

t7> return cephes.stdtr(10, 10e150)
1
t7> return cephes.stdtr(10, 10e300)
0.5

[jucor]

And voila, created jucor/torch-cdflib :-) You were right Daniel, that's much cleaner that way.

Now there's the question of which test we should add to torch-cephes? Some tests that assert that the bug is here? That would probably sound the most coherent, but it feels weird to add tests that success for the presence of a bug.

[d11]

Nice one! Awesome job. Thanks for sorting out that doc I added too, I just realised I put it in the wrong place!

Hmm, yes, not too sure about the tests. I'm almost tempted to say we should just remove stdtr from torch-cephes, actually, unless there are cases where that implementation is better…?

Cheers
Dan

On 2 Dec 2013, at 22:14, Julien Cornebise [email protected] wrote:

And voila, created jucor/torch-cdflib :-) You were right Daniel, that's much cleaner that way.

Now there's the question of which test we should add to torch-cephes? Some tests that assert that the bug is here? That would probably sound the most coherent, but it feels weird to add tests that success for the presence of a bug.

—
Reply to this email directly or view it on GitHub.

[jucor]

Hi @d11
I'm going through our few open pull-requests. Can't remember what we decided for this one. Shall we discuss about bringing github.com/jucor/torch-cdflib into deepmind/ ?I do not think anyone is using it yet, but since @dpfau and @theoweber are up to getting torch-distributions a bit more efficient, it could be a good time to bring cdflib up to speed.
What do you think?

[jucor]

Ping @d11 @akfidjeland.