Use of numpy.isclose method

[jeking04]

Currently, to use the numpy.isclose method we need to install numpy from pip rather than apt. This is a bit cumbersome. Can we remove this dependency?

[cdellin]

Since the implementation [0] is only 23 lines, maybe we can just maintain our own version and hot-patch it in when we import numpy if it isn't there?

[0] https://github.com/numpy/numpy/blob/v1.9.1/numpy/core/numeric.py#L2238

[psigen]

Is this an issue in 14.04 as well or only in 12.04?

On Fri, Mar 6, 2015 at 3:06 PM, cdellin notifications@github.com wrote:

Since the implementation [0] is only 23 lines, maybe we can just maintain
our own version and hot-patch it in when we import numpy if it isn't there?

[0] https://github.com/numpy/numpy/blob/v1.9.1/numpy/core/numeric.py#L2238

—
Reply to this email directly or view it on GitHub
#63 (comment)
.

[mkoval]

I believe the version of NumPy shipped with 14.04 has is_close.

More importantly, I'm not sure if we want to use is_close in this context
anyway. Aren't the default tolerances much tigher than what we care about?

On Fri, Mar 6, 2015 at 3:08 PM, Pras Velagapudi notifications@github.com
wrote:

Is this an issue in 14.04 as well or only in 12.04?

On Fri, Mar 6, 2015 at 3:06 PM, cdellin notifications@github.com wrote:

Since the implementation [0] is only 23 lines, maybe we can just maintain
our own version and hot-patch it in when we import numpy if it isn't
there?

[0]
https://github.com/numpy/numpy/blob/v1.9.1/numpy/core/numeric.py#L2238

—
Reply to this email directly or view it on GitHub
<
https://github.com/personalrobotics/prpy/issues/63#issuecomment-77626241>

.

—
Reply to this email directly or view it on GitHub
#63 (comment)
.

[siddhss5]

OK, how is one supposed to programmatically decide if a configuration is close enough to another?
Is it with: all(abs(a-b) < robot.GetActiveDOFResolutions())?

If so, or if it's something slightly different, it'd be nice if that were in prpy.util, no?

[mkoval]

I see two places where numpy.isclose is used:

1. In ComputeJointVelocityFromTwist:

dq_bounds = [(0, max) if (numpy.isclose(q_curr[i], q_min[i])) else
             (min, 0) if (numpy.isclose(q_curr[i], q_max[i])) else
             (min, max) for i, (min, max) in enumerate(bounds)]

Here, I think it makes sense to use GetActiveDOFResolution like you suggested.

1. In PlanWorkspacePath:

while not numpy.isclose(t, traj.GetDuration()):

Here, I think this should be replaced with t < traj.GetDuration().

@siddhss5 Thoughts?

[psigen]

@siddhss5, @mkoval: Has this been addressed/fixed already?

I think @mkoval's suggestions make sense, and while I am fine with using numpy.isclose() in general, I don't think it's necessary in our codebase right now since there aren't too many cases where an exact comparison is actually what we want.