NIFTy merge requestshttps://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests2020-03-25T13:32:31Zhttps://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/430metric_gaussian_kl.py: Fix samples property2020-03-25T13:32:31ZGordian Edenhofermetric_gaussian_kl.py: Fix samples propertyFix a bug which was introduced in eded7903fb making it impossible to
access the sample property of a KL with mirrored samples as tuples
can not be added to lists.Fix a bug which was introduced in eded7903fb making it impossible to
access the sample property of a KL with mirrored samples as tuples
can not be added to lists.https://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/429Fix KL2020-03-25T14:42:20ZMartin ReineckeFix KLI think I made a stupid mistake when I merged the MPI KL with the normal one...
@parras, @reimar, could you please double check?
(The new KL does not store the mirrored samples to save space, and in the MPI case, a mirrored sample ...I think I made a stupid mistake when I merged the MPI KL with the normal one...
@parras, @reimar, could you please double check?
(The new KL does not store the mirrored samples to save space, and in the MPI case, a mirrored sample pair always sits on exactly one task.)
Unfortunately, this is not the bug which caused the recent breakage ...https://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/428Switch to new numpy random generators2020-03-24T11:00:16ZMartin ReineckeSwitch to new numpy random generatorsAfter the change some test fail. I'm not absolutely sure whether they fail because they rely on a specific seed. If so, we should change them so that this is no longer the case.
@reimar The MPI KL still needs a fix to work consistentl...After the change some test fail. I'm not absolutely sure whether they fail because they rely on a specific seed. If so, we should change them so that this is no longer the case.
@reimar The MPI KL still needs a fix to work consistently even when the number of tasks is changed. I added a FIXME comment. If it's easy for you to implement, please go ahead, otherwise I'll try to do it at some point.
Still TODO: docstrings in `nifty6.random`.
@parras, @reimar: What do you think about the interface?https://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/427Linearization logic fixups2020-03-24T11:00:31ZPhilipp Arrasparras@mpa-garching.mpg.deLinearization logic fixupsMartin ReineckeMartin Reineckehttps://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/426Be more paranoid about initializing the RNG2020-03-24T11:25:14ZMartin ReineckeBe more paranoid about initializing the RNGThis is an attempt to enforce reproducible runs. Whenever `from_random` is called without a preceding `random.init()`, an exception will be raised.
@parras, does this look OK in principle?
I also realized that there are still some ...This is an attempt to enforce reproducible runs. Whenever `from_random` is called without a preceding `random.init()`, an exception will be raised.
@parras, does this look OK in principle?
I also realized that there are still some direct calls to `np.random` functionality within `nifty6`, which I think is bad. We should replace this soon.Martin ReineckeMartin Reineckehttps://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/425Fixups2020-03-17T07:44:56ZPhilipp Arrasparras@mpa-garching.mpg.deFixups@lerou is that fine with you?@lerou is that fine with you?https://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/424Add Wiener process and integrated Wiener process2020-03-16T17:44:52ZPhilipp Arrasparras@mpa-garching.mpg.deAdd Wiener process and integrated Wiener processhttps://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/423Metric and jacobian fixups2020-03-15T08:26:04ZPhilipp Arrasparras@mpa-garching.mpg.deMetric and jacobian fixups@mtr, ready to be merged@mtr, ready to be mergedhttps://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/422Add forgotten test2020-03-11T22:39:14ZPhilipp Arrasparras@mpa-garching.mpg.deAdd forgotten testhttps://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/421added the one_over method to MutliField2020-03-11T21:45:59ZReimar H Leikeadded the one_over method to MutliFieldall methods should be consistent between Field and `MultiField`.
The `one_over` method is also necessary for our eht-imaging code.all methods should be consistent between Field and `MultiField`.
The `one_over` method is also necessary for our eht-imaging code.Philipp Arrasparras@mpa-garching.mpg.dePhilipp Arrasparras@mpa-garching.mpg.dehttps://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/420Update year in copyright2020-03-15T13:46:07ZPhilipp Arrasparras@mpa-garching.mpg.deUpdate year in copyrighthttps://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/419Uniform operator2020-03-15T08:39:59ZRouven LemmerzUniform operatorThis implements a `UniformOperator`, which generates a Uniform distribution.
Also, this fixes the `_InterpolationOperator` to allow for different `table_funcs` than logarithm, which lead to a wrong jacobian beforehand.This implements a `UniformOperator`, which generates a Uniform distribution.
Also, this fixes the `_InterpolationOperator` to allow for different `table_funcs` than logarithm, which lead to a wrong jacobian beforehand.https://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/418New apply2020-03-11T23:04:47ZPhilipp Arrasparras@mpa-garching.mpg.deNew applyhttps://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/417Performance pa2020-03-27T13:03:57ZPhilipp Arrasparras@mpa-garching.mpg.dePerformance pahttps://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/416Inverse gamma2020-03-09T12:18:23ZPhilipp Arrasparras@mpa-garching.mpg.deInverse gammaMartin ReineckeMartin Reineckehttps://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/415CorrelatedFieldMaker: offset parametrization in one place2020-04-01T15:32:22ZLukas PlatzCorrelatedFieldMaker: offset parametrization in one placeHi @parras, @pfrank, @phaim,
in the `CorrelatedFieldMaker` as it stands at the moment, the offset from zero is parametrized in two separate places.
I propose to unify the offset parametrization into `CorrelatedFieldMaker.make` via the ...Hi @parras, @pfrank, @phaim,
in the `CorrelatedFieldMaker` as it stands at the moment, the offset from zero is parametrized in two separate places.
I propose to unify the offset parametrization into `CorrelatedFieldMaker.make` via the parameters `offset_mean`, `offset_std_mean`, `offset_std_std`.
While at it, modified the `getting_started_3` demo (where the CF model is introduced first) to pass the parameters by name and with comments, giving the reader a chance to understand what is happening and providing a nice template for copying.
Also, I renamed a few function parameters in `CorrelatedFieldMaker` in hope to make them more descriptive, but you might disagree with my judgement there.
As the creators of the model, what do you think about these changes?
Cheers,
Lukashttps://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/414Mightier operators2020-03-08T10:52:01ZReimar H LeikeMightier operatorsadd more methods to operator such that it works more like Linearizationadd more methods to operator such that it works more like LinearizationPhilipp Arrasparras@mpa-garching.mpg.dePhilipp Arrasparras@mpa-garching.mpg.dehttps://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/413Fixup gig energy2020-03-06T14:24:32ZPhilipp Arrasparras@mpa-garching.mpg.deFixup gig energyhttps://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/412Gig energy cleanup pa2020-03-06T12:05:47ZPhilipp Arrasparras@mpa-garching.mpg.deGig energy cleanup pahttps://gitlab.mpcdf.mpg.de/ift/nifty/-/merge_requests/411Gig energy2020-03-06T12:02:32ZMartin ReineckeGig energyAndrija KosticAndrija Kostic