arrays of datetime.time() objects don't convert into TimeSeries

[blais]

Create an array of datetime.time() objects as an index for a DataFrame, and plotting will be wrong, and the index is not of type TimeSeries. Convert that array (using combine()) into an array of datetime.datetime() objects, and it works.

[changhiskhan]

Thanks for the bug report Martin!
Currently pandas cannot create a DatetimeIndex out of array of datetime.time directly.
Suppose we had a convenience function time_range with a date keyword to do the combine operation, would that be sufficient for your use case? Do you need to manipulate the index as datetime.time objects?

[wesm]

It looks like the root issue is that matplotlib doesn't have a converter for datetime.time. I'm not sure why creating a Series indexed by datetime.time plots at all...bug?

import datetime
times = [(datetime.datetime(2012, 6, 26) + datetime.timedelta(minutes=i)).time() for i in range(100)]
ts = Series(np.random.randn(100), index=times)
ts.plot()

???

[changhiskhan]

it's plotting using linspace and then setting the xticklabels (this is in
the existing non-ts code).

need to convert to datetimes first

On Tue, Jun 26, 2012 at 6:09 PM, Wes McKinney <
reply@reply.github.com

wrote:

It looks like the root issue is that matplotlib doesn't have a converter
for datetime.time. I'm not sure why creating a Series indexed by
datetime.time plots at all...bug?

import datetime
times = [(datetime.datetime(2012, 6, 26) +
datetime.timedelta(minutes=i)).time() for i in range(100)]
ts = Series(np.random.randn(100), index=times)
ts.plot()

???

---

Reply to this email directly or view it on GitHub:
#1531 (comment)

Chang She
Lambda Foundry http://www.lambdafoundry.com

[wesm]

Well, strictly speaking if the index is datetime.time, it's not necessarily proper to assign them to a particular calendar date?

[changhiskhan]

Right, I agree it would be a hack.

I mean, we can try writing our own datetime.time converter for MPL. Can
probably mostly just reuse the existing datetime.datetime converter
perhaps. Maybe use a floating number to represent seconds since midnight as
the x-axis ordinals?

On Tue, Jun 26, 2012 at 6:30 PM, Wes McKinney <
reply@reply.github.com

wrote:

Well, strictly speaking if the index is datetime.time, it's not
necessarily proper to assign them to a particular calendar date?

---

Reply to this email directly or view it on GitHub:
#1531 (comment)

Chang She
Lambda Foundry http://www.lambdafoundry.com

[timmie]

@changhiskhan

I mean, we can try writing our own datetime.time converter for MPL.
would also be easier to fix #1300 then.

[changhiskhan]

How do you see the datetime.time converter help with that issue? I thought #1300 still pertained only to plotting datetime.datetime.

[wesm]

Merged #1531 and plotting with datetime.time index looking good now

[blais]

Dude you ruleeee

On Thu, Jun 28, 2012, at 13:04, Wes McKinney wrote:

Merged #1531 and plotting with datetime.time index looking good now

---

Reply to this email directly or view it on GitHub:
#1531 (comment)

[wesm]

It's all @changhiskhan on this one :) Glad it's sorted out