Allow empty sub-containers in reductions by majosm · Pull Request #129 · inducer/arraycontext

majosm · 2021-12-15T22:16:43Z

Allows some (but not all) of the sub-containers in an array container to be empty in reductions. This allows reductions to be used on single-species CV objects in mirgecom.

Depends on #128 (not in an essential way, just avoiding dealing with merge conflicts).

inducer · 2021-12-27T17:05:52Z

This has picked up some merge conflicts. Could you take a look? Also, ready to un-draft?

inducer

Thanks! LGTM aside from the all-empty handling, which I think ought to be quick to resolve.

inducer · 2021-12-27T19:09:26Z

+ if result is None:
+ raise ValueError("cannot reduce empty array container")


Personally, I think reducing over an empty container is well-defined: just return the neutral element. reduce_func should be trusted to do the right thing. What was the reasoning for putting in this error message? Basically, what I'm pushing for is to get rid of it and return reduce_func([]) instead.

If reduce_func worked for an empty list, I think this PR wouldn't be needed, right?
The motivation for this was for things like actx.np.max where reduce_func doesn't (currently) behave nicely for empty inputs. Should that maybe be fixed instead?

For reference, calling np.max on an empty list raises
>>> np.max([]) ValueError: zero-size array to reduction operation maximum which has no identity
and requires specifying an initial value with np.max([], initial=np.inf). We should probably do something like that in actx.np too.

The motivation for this was for things like actx.np.max where reduce_func doesn't (currently) behave nicely for empty inputs.
That's a great point. I had not thought far enough to realize that, but I agree now.
Should that maybe be fixed instead?
I think that's exactly what we should do.
We should probably do something like that in actx.np too.
I agree. While I wasn't loving numpy's interface of "no implicit neutral" element at first, I think it makes sense from the perspective of integers. I think we should aim to replicate this, both at the level of actx.np, and for the array container reductions under discussion here.
One subtlety is the ValueError for empty arrays. For symbolically-shaped pytato arrays, it may not be possible to do a perfect job.

Did some work towards that:
Introduce ReductionOperation class, accept 'initial' in reductions pytato#238
Match numpy interface for empty reductions #136
That latter PR needs some more work done to it. @majosm, could you take a look?

inducer · 2021-12-27T19:10:49Z

+ if result is None:
+ raise ValueError("cannot reduce empty array container")


inducer · 2021-12-28T14:39:48Z

@majosm, depending on how you feel, #136 might supersede this PR.

majosm · 2021-12-15T22:16:43Z

Allows some (but not all) of the sub-containers in an array container to be empty in reductions. This allows reductions to be used on single-species CV objects in mirgecom.

Depends on #128 (not in an essential way, just avoiding dealing with merge conflicts).

inducer · 2021-12-27T17:05:52Z

This has picked up some merge conflicts. Could you take a look? Also, ready to un-draft?

inducer

Thanks! LGTM aside from the all-empty handling, which I think ought to be quick to resolve.

inducer · 2021-12-27T19:09:26Z

+ if result is None:
+ raise ValueError("cannot reduce empty array container")


Personally, I think reducing over an empty container is well-defined: just return the neutral element. reduce_func should be trusted to do the right thing. What was the reasoning for putting in this error message? Basically, what I'm pushing for is to get rid of it and return reduce_func([]) instead.

If reduce_func worked for an empty list, I think this PR wouldn't be needed, right?
The motivation for this was for things like actx.np.max where reduce_func doesn't (currently) behave nicely for empty inputs. Should that maybe be fixed instead?

For reference, calling np.max on an empty list raises
>>> np.max([]) ValueError: zero-size array to reduction operation maximum which has no identity
and requires specifying an initial value with np.max([], initial=np.inf). We should probably do something like that in actx.np too.

The motivation for this was for things like actx.np.max where reduce_func doesn't (currently) behave nicely for empty inputs.
That's a great point. I had not thought far enough to realize that, but I agree now.
Should that maybe be fixed instead?
I think that's exactly what we should do.
We should probably do something like that in actx.np too.
I agree. While I wasn't loving numpy's interface of "no implicit neutral" element at first, I think it makes sense from the perspective of integers. I think we should aim to replicate this, both at the level of actx.np, and for the array container reductions under discussion here.
One subtlety is the ValueError for empty arrays. For symbolically-shaped pytato arrays, it may not be possible to do a perfect job.

Did some work towards that:
Introduce ReductionOperation class, accept 'initial' in reductions pytato#238
Match numpy interface for empty reductions #136
That latter PR needs some more work done to it. @majosm, could you take a look?

inducer · 2021-12-27T19:10:49Z

+ if result is None:
+ raise ValueError("cannot reduce empty array container")


inducer · 2021-12-28T14:39:48Z

@majosm, depending on how you feel, #136 might supersede this PR.

majosm · 2021-12-15T22:16:43Z

Allows some (but not all) of the sub-containers in an array container to be empty in reductions. This allows reductions to be used on single-species CV objects in mirgecom.

Depends on #128 (not in an essential way, just avoiding dealing with merge conflicts).

inducer · 2021-12-27T17:05:52Z

This has picked up some merge conflicts. Could you take a look? Also, ready to un-draft?

inducer

Thanks! LGTM aside from the all-empty handling, which I think ought to be quick to resolve.

inducer · 2021-12-27T19:09:26Z

+ if result is None:
+ raise ValueError("cannot reduce empty array container")


Personally, I think reducing over an empty container is well-defined: just return the neutral element. reduce_func should be trusted to do the right thing. What was the reasoning for putting in this error message? Basically, what I'm pushing for is to get rid of it and return reduce_func([]) instead.

If reduce_func worked for an empty list, I think this PR wouldn't be needed, right?
The motivation for this was for things like actx.np.max where reduce_func doesn't (currently) behave nicely for empty inputs. Should that maybe be fixed instead?

For reference, calling np.max on an empty list raises
>>> np.max([]) ValueError: zero-size array to reduction operation maximum which has no identity
and requires specifying an initial value with np.max([], initial=np.inf). We should probably do something like that in actx.np too.

The motivation for this was for things like actx.np.max where reduce_func doesn't (currently) behave nicely for empty inputs.
That's a great point. I had not thought far enough to realize that, but I agree now.
Should that maybe be fixed instead?
I think that's exactly what we should do.
We should probably do something like that in actx.np too.
I agree. While I wasn't loving numpy's interface of "no implicit neutral" element at first, I think it makes sense from the perspective of integers. I think we should aim to replicate this, both at the level of actx.np, and for the array container reductions under discussion here.
One subtlety is the ValueError for empty arrays. For symbolically-shaped pytato arrays, it may not be possible to do a perfect job.

Did some work towards that:
Introduce ReductionOperation class, accept 'initial' in reductions pytato#238
Match numpy interface for empty reductions #136
That latter PR needs some more work done to it. @majosm, could you take a look?

inducer · 2021-12-27T19:10:49Z

+ if result is None:
+ raise ValueError("cannot reduce empty array container")


inducer · 2021-12-28T14:39:48Z

@majosm, depending on how you feel, #136 might supersede this PR.

majosm · 2021-12-15T22:16:43Z

Allows some (but not all) of the sub-containers in an array container to be empty in reductions. This allows reductions to be used on single-species CV objects in mirgecom.

Depends on #128 (not in an essential way, just avoiding dealing with merge conflicts).

inducer · 2021-12-27T17:05:52Z

This has picked up some merge conflicts. Could you take a look? Also, ready to un-draft?

inducer

Thanks! LGTM aside from the all-empty handling, which I think ought to be quick to resolve.

inducer · 2021-12-27T19:09:26Z

+ if result is None:
+ raise ValueError("cannot reduce empty array container")


Personally, I think reducing over an empty container is well-defined: just return the neutral element. reduce_func should be trusted to do the right thing. What was the reasoning for putting in this error message? Basically, what I'm pushing for is to get rid of it and return reduce_func([]) instead.

If reduce_func worked for an empty list, I think this PR wouldn't be needed, right?
The motivation for this was for things like actx.np.max where reduce_func doesn't (currently) behave nicely for empty inputs. Should that maybe be fixed instead?

For reference, calling np.max on an empty list raises
>>> np.max([]) ValueError: zero-size array to reduction operation maximum which has no identity
and requires specifying an initial value with np.max([], initial=np.inf). We should probably do something like that in actx.np too.

The motivation for this was for things like actx.np.max where reduce_func doesn't (currently) behave nicely for empty inputs.
That's a great point. I had not thought far enough to realize that, but I agree now.
Should that maybe be fixed instead?
I think that's exactly what we should do.
We should probably do something like that in actx.np too.
I agree. While I wasn't loving numpy's interface of "no implicit neutral" element at first, I think it makes sense from the perspective of integers. I think we should aim to replicate this, both at the level of actx.np, and for the array container reductions under discussion here.
One subtlety is the ValueError for empty arrays. For symbolically-shaped pytato arrays, it may not be possible to do a perfect job.

Did some work towards that:
Introduce ReductionOperation class, accept 'initial' in reductions pytato#238
Match numpy interface for empty reductions #136
That latter PR needs some more work done to it. @majosm, could you take a look?

inducer · 2021-12-27T19:10:49Z

+ if result is None:
+ raise ValueError("cannot reduce empty array container")


inducer · 2021-12-28T14:39:48Z

@majosm, depending on how you feel, #136 might supersede this PR.

majosm · 2021-12-15T22:16:43Z

Allows some (but not all) of the sub-containers in an array container to be empty in reductions. This allows reductions to be used on single-species CV objects in mirgecom.

Depends on #128 (not in an essential way, just avoiding dealing with merge conflicts).

inducer · 2021-12-27T17:05:52Z

This has picked up some merge conflicts. Could you take a look? Also, ready to un-draft?

inducer

Thanks! LGTM aside from the all-empty handling, which I think ought to be quick to resolve.

inducer · 2021-12-27T19:09:26Z

+ if result is None:
+ raise ValueError("cannot reduce empty array container")


Personally, I think reducing over an empty container is well-defined: just return the neutral element. reduce_func should be trusted to do the right thing. What was the reasoning for putting in this error message? Basically, what I'm pushing for is to get rid of it and return reduce_func([]) instead.

If reduce_func worked for an empty list, I think this PR wouldn't be needed, right?
The motivation for this was for things like actx.np.max where reduce_func doesn't (currently) behave nicely for empty inputs. Should that maybe be fixed instead?

For reference, calling np.max on an empty list raises
>>> np.max([]) ValueError: zero-size array to reduction operation maximum which has no identity
and requires specifying an initial value with np.max([], initial=np.inf). We should probably do something like that in actx.np too.

The motivation for this was for things like actx.np.max where reduce_func doesn't (currently) behave nicely for empty inputs.
That's a great point. I had not thought far enough to realize that, but I agree now.
Should that maybe be fixed instead?
I think that's exactly what we should do.
We should probably do something like that in actx.np too.
I agree. While I wasn't loving numpy's interface of "no implicit neutral" element at first, I think it makes sense from the perspective of integers. I think we should aim to replicate this, both at the level of actx.np, and for the array container reductions under discussion here.
One subtlety is the ValueError for empty arrays. For symbolically-shaped pytato arrays, it may not be possible to do a perfect job.

Did some work towards that:
Introduce ReductionOperation class, accept 'initial' in reductions pytato#238
Match numpy interface for empty reductions #136
That latter PR needs some more work done to it. @majosm, could you take a look?

inducer · 2021-12-27T19:10:49Z

+ if result is None:
+ raise ValueError("cannot reduce empty array container")


inducer · 2021-12-28T14:39:48Z

@majosm, depending on how you feel, #136 might supersede this PR.

majosm · 2021-12-15T22:16:43Z

Allows some (but not all) of the sub-containers in an array container to be empty in reductions. This allows reductions to be used on single-species CV objects in mirgecom.

Depends on #128 (not in an essential way, just avoiding dealing with merge conflicts).

inducer · 2021-12-27T17:05:52Z

This has picked up some merge conflicts. Could you take a look? Also, ready to un-draft?

inducer

Thanks! LGTM aside from the all-empty handling, which I think ought to be quick to resolve.

inducer · 2021-12-27T19:09:26Z

+ if result is None:
+ raise ValueError("cannot reduce empty array container")


Personally, I think reducing over an empty container is well-defined: just return the neutral element. reduce_func should be trusted to do the right thing. What was the reasoning for putting in this error message? Basically, what I'm pushing for is to get rid of it and return reduce_func([]) instead.

If reduce_func worked for an empty list, I think this PR wouldn't be needed, right?
The motivation for this was for things like actx.np.max where reduce_func doesn't (currently) behave nicely for empty inputs. Should that maybe be fixed instead?

For reference, calling np.max on an empty list raises
>>> np.max([]) ValueError: zero-size array to reduction operation maximum which has no identity
and requires specifying an initial value with np.max([], initial=np.inf). We should probably do something like that in actx.np too.

The motivation for this was for things like actx.np.max where reduce_func doesn't (currently) behave nicely for empty inputs.
That's a great point. I had not thought far enough to realize that, but I agree now.
Should that maybe be fixed instead?
I think that's exactly what we should do.
We should probably do something like that in actx.np too.
I agree. While I wasn't loving numpy's interface of "no implicit neutral" element at first, I think it makes sense from the perspective of integers. I think we should aim to replicate this, both at the level of actx.np, and for the array container reductions under discussion here.
One subtlety is the ValueError for empty arrays. For symbolically-shaped pytato arrays, it may not be possible to do a perfect job.

Did some work towards that:
Introduce ReductionOperation class, accept 'initial' in reductions pytato#238
Match numpy interface for empty reductions #136
That latter PR needs some more work done to it. @majosm, could you take a look?

inducer · 2021-12-27T19:10:49Z

+ if result is None:
+ raise ValueError("cannot reduce empty array container")


inducer · 2021-12-28T14:39:48Z

@majosm, depending on how you feel, #136 might supersede this PR.

majosm · 2021-12-15T22:16:43Z

Allows some (but not all) of the sub-containers in an array container to be empty in reductions. This allows reductions to be used on single-species CV objects in mirgecom.

Depends on #128 (not in an essential way, just avoiding dealing with merge conflicts).

inducer · 2021-12-27T17:05:52Z

This has picked up some merge conflicts. Could you take a look? Also, ready to un-draft?

inducer

Thanks! LGTM aside from the all-empty handling, which I think ought to be quick to resolve.

inducer · 2021-12-27T19:09:26Z

+ if result is None:
+ raise ValueError("cannot reduce empty array container")


Personally, I think reducing over an empty container is well-defined: just return the neutral element. reduce_func should be trusted to do the right thing. What was the reasoning for putting in this error message? Basically, what I'm pushing for is to get rid of it and return reduce_func([]) instead.

If reduce_func worked for an empty list, I think this PR wouldn't be needed, right?
The motivation for this was for things like actx.np.max where reduce_func doesn't (currently) behave nicely for empty inputs. Should that maybe be fixed instead?

For reference, calling np.max on an empty list raises
>>> np.max([]) ValueError: zero-size array to reduction operation maximum which has no identity
and requires specifying an initial value with np.max([], initial=np.inf). We should probably do something like that in actx.np too.

The motivation for this was for things like actx.np.max where reduce_func doesn't (currently) behave nicely for empty inputs.
That's a great point. I had not thought far enough to realize that, but I agree now.
Should that maybe be fixed instead?
I think that's exactly what we should do.
We should probably do something like that in actx.np too.
I agree. While I wasn't loving numpy's interface of "no implicit neutral" element at first, I think it makes sense from the perspective of integers. I think we should aim to replicate this, both at the level of actx.np, and for the array container reductions under discussion here.
One subtlety is the ValueError for empty arrays. For symbolically-shaped pytato arrays, it may not be possible to do a perfect job.

Did some work towards that:
Introduce ReductionOperation class, accept 'initial' in reductions pytato#238
Match numpy interface for empty reductions #136
That latter PR needs some more work done to it. @majosm, could you take a look?

inducer · 2021-12-27T19:10:49Z

+ if result is None:
+ raise ValueError("cannot reduce empty array container")


inducer · 2021-12-28T14:39:48Z

@majosm, depending on how you feel, #136 might supersede this PR.

majosm · 2021-12-15T22:16:43Z

Allows some (but not all) of the sub-containers in an array container to be empty in reductions. This allows reductions to be used on single-species CV objects in mirgecom.

Depends on #128 (not in an essential way, just avoiding dealing with merge conflicts).

inducer · 2021-12-27T17:05:52Z

This has picked up some merge conflicts. Could you take a look? Also, ready to un-draft?

inducer

Thanks! LGTM aside from the all-empty handling, which I think ought to be quick to resolve.

inducer · 2021-12-27T19:09:26Z

+ if result is None:
+ raise ValueError("cannot reduce empty array container")


Personally, I think reducing over an empty container is well-defined: just return the neutral element. reduce_func should be trusted to do the right thing. What was the reasoning for putting in this error message? Basically, what I'm pushing for is to get rid of it and return reduce_func([]) instead.

If reduce_func worked for an empty list, I think this PR wouldn't be needed, right?
The motivation for this was for things like actx.np.max where reduce_func doesn't (currently) behave nicely for empty inputs. Should that maybe be fixed instead?

For reference, calling np.max on an empty list raises
>>> np.max([]) ValueError: zero-size array to reduction operation maximum which has no identity
and requires specifying an initial value with np.max([], initial=np.inf). We should probably do something like that in actx.np too.

The motivation for this was for things like actx.np.max where reduce_func doesn't (currently) behave nicely for empty inputs.
That's a great point. I had not thought far enough to realize that, but I agree now.
Should that maybe be fixed instead?
I think that's exactly what we should do.
We should probably do something like that in actx.np too.
I agree. While I wasn't loving numpy's interface of "no implicit neutral" element at first, I think it makes sense from the perspective of integers. I think we should aim to replicate this, both at the level of actx.np, and for the array container reductions under discussion here.
One subtlety is the ValueError for empty arrays. For symbolically-shaped pytato arrays, it may not be possible to do a perfect job.

Did some work towards that:
Introduce ReductionOperation class, accept 'initial' in reductions pytato#238
Match numpy interface for empty reductions #136
That latter PR needs some more work done to it. @majosm, could you take a look?

inducer · 2021-12-27T19:10:49Z

+ if result is None:
+ raise ValueError("cannot reduce empty array container")


inducer · 2021-12-28T14:39:48Z

@majosm, depending on how you feel, #136 might supersede this PR.

This was referenced Dec 15, 2021
Use arraycontext reductions in reductions inducer/grudge#199
Draft
More simplifications after grudge array container support illinois-ceesd/mirgecom#579
Draft

majosm force-pushed the empty-subcontainers branch from 974bdfd to 0ff5f70Compare December 15, 2021 22:42

majosm mentioned this pull request Dec 16, 2021
CI failing in mypy #130
Closed

majosm force-pushed the empty-subcontainers branch 2 times, most recently from 4ab1b0b to 8bd3252Compare December 16, 2021 21:35

allow empty sub-containers in reductions
cb8f3e7

majosm force-pushed the empty-subcontainers branch from 8bd3252 to cb8f3e7Compare December 27, 2021 18:18

majosm marked this pull request as ready for review December 27, 2021 18:56

majosm requested a review from inducer December 27, 2021 18:56

inducer reviewed Dec 27, 2021
View reviewed changes

inducer mentioned this pull request Dec 28, 2021
Introduce ReductionOperation class, accept 'initial' in reductions inducer/pytato#238
Merged

inducer mentioned this pull request Jun 16, 2022
Do not fail for empty reductions inducer/grudge#266
Open

This was referenced Dec 15, 2021
Use arraycontext reductions in reductions inducer/grudge#199
Draft
More simplifications after grudge array container support illinois-ceesd/mirgecom#579
Draft

majosm force-pushed the empty-subcontainers branch from 974bdfd to 0ff5f70Compare December 15, 2021 22:42

majosm mentioned this pull request Dec 16, 2021
CI failing in mypy #130
Closed

majosm force-pushed the empty-subcontainers branch 2 times, most recently from 4ab1b0b to 8bd3252Compare December 16, 2021 21:35

allow empty sub-containers in reductions
cb8f3e7

majosm force-pushed the empty-subcontainers branch from 8bd3252 to cb8f3e7Compare December 27, 2021 18:18

majosm marked this pull request as ready for review December 27, 2021 18:56

majosm requested a review from inducer December 27, 2021 18:56

inducer reviewed Dec 27, 2021
View reviewed changes

inducer mentioned this pull request Dec 28, 2021
Introduce ReductionOperation class, accept 'initial' in reductions inducer/pytato#238
Merged

inducer mentioned this pull request Jun 16, 2022
Do not fail for empty reductions inducer/grudge#266
Open

This was referenced Dec 15, 2021
Use arraycontext reductions in reductions inducer/grudge#199
Draft
More simplifications after grudge array container support illinois-ceesd/mirgecom#579
Draft

majosm force-pushed the empty-subcontainers branch from 974bdfd to 0ff5f70Compare December 15, 2021 22:42

majosm mentioned this pull request Dec 16, 2021
CI failing in mypy #130
Closed

majosm force-pushed the empty-subcontainers branch 2 times, most recently from 4ab1b0b to 8bd3252Compare December 16, 2021 21:35

allow empty sub-containers in reductions
cb8f3e7

majosm force-pushed the empty-subcontainers branch from 8bd3252 to cb8f3e7Compare December 27, 2021 18:18

majosm marked this pull request as ready for review December 27, 2021 18:56

majosm requested a review from inducer December 27, 2021 18:56

inducer reviewed Dec 27, 2021
View reviewed changes

inducer mentioned this pull request Dec 28, 2021
Introduce ReductionOperation class, accept 'initial' in reductions inducer/pytato#238
Merged

inducer mentioned this pull request Jun 16, 2022
Do not fail for empty reductions inducer/grudge#266
Open

This was referenced Dec 15, 2021
Use arraycontext reductions in reductions inducer/grudge#199
Draft
More simplifications after grudge array container support illinois-ceesd/mirgecom#579
Draft

majosm force-pushed the empty-subcontainers branch from 974bdfd to 0ff5f70Compare December 15, 2021 22:42

majosm mentioned this pull request Dec 16, 2021
CI failing in mypy #130
Closed

majosm force-pushed the empty-subcontainers branch 2 times, most recently from 4ab1b0b to 8bd3252Compare December 16, 2021 21:35

allow empty sub-containers in reductions
cb8f3e7

majosm force-pushed the empty-subcontainers branch from 8bd3252 to cb8f3e7Compare December 27, 2021 18:18

majosm marked this pull request as ready for review December 27, 2021 18:56

majosm requested a review from inducer December 27, 2021 18:56

inducer reviewed Dec 27, 2021
View reviewed changes

inducer mentioned this pull request Dec 28, 2021
Introduce ReductionOperation class, accept 'initial' in reductions inducer/pytato#238
Merged

inducer mentioned this pull request Jun 16, 2022
Do not fail for empty reductions inducer/grudge#266
Open

This was referenced Dec 15, 2021
Use arraycontext reductions in reductions inducer/grudge#199
Draft
More simplifications after grudge array container support illinois-ceesd/mirgecom#579
Draft

majosm force-pushed the empty-subcontainers branch from 974bdfd to 0ff5f70Compare December 15, 2021 22:42

majosm mentioned this pull request Dec 16, 2021
CI failing in mypy #130
Closed

majosm force-pushed the empty-subcontainers branch 2 times, most recently from 4ab1b0b to 8bd3252Compare December 16, 2021 21:35

allow empty sub-containers in reductions
cb8f3e7

majosm force-pushed the empty-subcontainers branch from 8bd3252 to cb8f3e7Compare December 27, 2021 18:18

majosm marked this pull request as ready for review December 27, 2021 18:56

majosm requested a review from inducer December 27, 2021 18:56

inducer reviewed Dec 27, 2021
View reviewed changes

inducer mentioned this pull request Dec 28, 2021
Introduce ReductionOperation class, accept 'initial' in reductions inducer/pytato#238
Merged

inducer mentioned this pull request Jun 16, 2022
Do not fail for empty reductions inducer/grudge#266
Open

This was referenced Dec 15, 2021
Use arraycontext reductions in reductions inducer/grudge#199
Draft
More simplifications after grudge array container support illinois-ceesd/mirgecom#579
Draft

majosm force-pushed the empty-subcontainers branch from 974bdfd to 0ff5f70Compare December 15, 2021 22:42

majosm mentioned this pull request Dec 16, 2021
CI failing in mypy #130
Closed

majosm force-pushed the empty-subcontainers branch 2 times, most recently from 4ab1b0b to 8bd3252Compare December 16, 2021 21:35

allow empty sub-containers in reductions
cb8f3e7

majosm force-pushed the empty-subcontainers branch from 8bd3252 to cb8f3e7Compare December 27, 2021 18:18

majosm marked this pull request as ready for review December 27, 2021 18:56

majosm requested a review from inducer December 27, 2021 18:56

inducer reviewed Dec 27, 2021
View reviewed changes

inducer mentioned this pull request Dec 28, 2021
Introduce ReductionOperation class, accept 'initial' in reductions inducer/pytato#238
Merged

inducer mentioned this pull request Jun 16, 2022
Do not fail for empty reductions inducer/grudge#266
Open

This was referenced Dec 15, 2021
Use arraycontext reductions in reductions inducer/grudge#199
Draft
More simplifications after grudge array container support illinois-ceesd/mirgecom#579
Draft

majosm force-pushed the empty-subcontainers branch from 974bdfd to 0ff5f70Compare December 15, 2021 22:42

majosm mentioned this pull request Dec 16, 2021
CI failing in mypy #130
Closed

majosm force-pushed the empty-subcontainers branch 2 times, most recently from 4ab1b0b to 8bd3252Compare December 16, 2021 21:35

allow empty sub-containers in reductions
cb8f3e7

majosm force-pushed the empty-subcontainers branch from 8bd3252 to cb8f3e7Compare December 27, 2021 18:18

majosm marked this pull request as ready for review December 27, 2021 18:56

majosm requested a review from inducer December 27, 2021 18:56

inducer reviewed Dec 27, 2021
View reviewed changes

inducer mentioned this pull request Dec 28, 2021
Introduce ReductionOperation class, accept 'initial' in reductions inducer/pytato#238
Merged

inducer mentioned this pull request Jun 16, 2022
Do not fail for empty reductions inducer/grudge#266
Open

This was referenced Dec 15, 2021
Use arraycontext reductions in reductions inducer/grudge#199
Draft
More simplifications after grudge array container support illinois-ceesd/mirgecom#579
Draft

majosm force-pushed the empty-subcontainers branch from 974bdfd to 0ff5f70Compare December 15, 2021 22:42

majosm mentioned this pull request Dec 16, 2021
CI failing in mypy #130
Closed

majosm force-pushed the empty-subcontainers branch 2 times, most recently from 4ab1b0b to 8bd3252Compare December 16, 2021 21:35

allow empty sub-containers in reductions
cb8f3e7

majosm force-pushed the empty-subcontainers branch from 8bd3252 to cb8f3e7Compare December 27, 2021 18:18

majosm marked this pull request as ready for review December 27, 2021 18:56

majosm requested a review from inducer December 27, 2021 18:56

inducer reviewed Dec 27, 2021
View reviewed changes

inducer mentioned this pull request Dec 28, 2021
Introduce ReductionOperation class, accept 'initial' in reductions inducer/pytato#238
Merged

inducer mentioned this pull request Jun 16, 2022
Do not fail for empty reductions inducer/grudge#266
Open

		if result is None:
		raise ValueError("cannot reduce empty array container")

Conversation

majosm commented Dec 15, 2021

Uh oh!

inducer commented Dec 27, 2021

Uh oh!

inducer left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

inducer commented Dec 28, 2021

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

majosm commented Dec 15, 2021

Uh oh!

inducer commented Dec 27, 2021

Uh oh!

inducer left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

inducer commented Dec 28, 2021

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

majosm commented Dec 15, 2021

Uh oh!

inducer commented Dec 27, 2021

Uh oh!

inducer left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!