Time is not a float. Note timings are sometimes float, sometimes np.float64

[craffel]

Time is not a float. Note timings are sometimes float, sometimes np.float64

[craffel]

Well, technically a np.float64 is a float, in the isinstance sense:

>>> a = np.float64(3.3)
>>> type(a)
numpy.float64
>>> isinstance(a, float)
True

In order to stay sane I usually make the assumption that np.float64 can be used interchangeably with float. Code which doesn't do make this assumption bums me out but does exist (Theano is a good example). I don't think it would do any harm to be more explicit about casting things to float, if that would fix the issue you're experiencing. But, I'd also gently suggest that the other code you're using which considers np.float64 and float different should maybe be changed too.

[douglaseck]

I can fix this on my side by treating note timings as c++ doubles not
floats.

That said, the documentation error stands :-) The time is documented as a
float. Then you explicitly raise an exception if a float is passed in.

On Wed, Apr 13, 2016 at 3:38 PM, Colin Raffel [email protected]
wrote:

Well, technically a np.float64 is a float, in the isinstance sense:

a = np.float64(3.3)>>> type(a)
numpy.float64>>> isinstance(a, float)True

In order to stay sane I usually make the assumption that np.float64 can
be used interchangeably with float. Code which doesn't do make this
assumption bums me out but does exist (Theano is a good example). I don't
think it would do any harm to be more explicit about casting things to
float, if that would fix the issue you're experiencing. But, I'd also
gently suggest that the other code you're using which considers np.float64
and float different should maybe be changed too.

—
You are receiving this because you authored the thread.
Reply to this email directly or view it on GitHub
#63 (comment)

[craffel]

That said, the documentation error stands :-) The time is documented as a
float.

Well, my policy is usually that the specified type in the docstring implies that the actual type will satisfy isinstance(thing, type).

Then you explicitly raise an exception if a float is passed in.

Which exception do you mean? If it's raising an Exception when a float is passed in, that's definitely a bug!

[douglaseck]

I misunderstood something about floats. I thought an np.float32 was a
python float. It turns out it's not. Only 64-bit floats evaluate to true
for isinstance(x, float). So your code is fine. I just need to switch my
code to use c++ doubles, not c++ floats.

Sorry for the confusion.

On Wed, Apr 13, 2016 at 4:58 PM, Colin Raffel [email protected]
wrote:

That said, the documentation error stands :-) The time is documented as a
float.

Well, my policy is usually that the specified type in the docstring
implies that the actual type will satisfy isinstance(thing, type).

Then you explicitly raise an exception if a float is passed in.

Which exception do you mean? If it's raising an Exception when a float is
passed in, that's definitely a bug!

—
You are receiving this because you authored the thread.
Reply to this email directly or view it on GitHub
#63 (comment)

[craffel]

I thought an np.float32 was a python float. It turns out it's not.

That's interesting, I also didn't realize isinstance(np.float32(3.1), float) evaluates to False. I guess it makes sense. I'll close this issue, feel free to reopen if you find any other problems here.