This is a new version of #15648 based on the latest master branch.
Unlike the previous PR where I fixed a lot of the doctests in addition to integrating xdoctest, I'm going to reduce the scope here. I'm simply going to integrate xdoctest, and then I'm going to mark all of the failing tests as "SKIP". This will let xdoctest run on the dashboards, provide some value, and still let the dashboards pass. I'll leave fixing the doctests themselves to another PR.
In my initial commit, I do the bare minimum to get something running with failing dashboards. The few tests that I marked as skip are causing segfaults. Running xdoctest results in 293 failed, 201 passed tests. The next commits will be to disable those tests. (unfortunately I don't have a tool that will insert the `#xdoctest: +SKIP` directive over every failing test, so I'm going to do this mostly manually.)
Fixes https://github.com/pytorch/pytorch/issues/71105
@ezyang
Pull Request resolved: https://github.com/pytorch/pytorch/pull/82797
Approved by: https://github.com/ezyang
This PR fixes#69466 and introduces some other minor changes. Tests are somewhat more involved because a reference implementation in `scipy` is not available; tests proceed differently for discrete and continuous distributions.
For continuous distributions, we evaluate the gradient of the `log_prob` at the mode. Tests pass if the gradient is zero OR (the mode is at the boundary of the support of the distribution AND the `log_prob` decreases as we move away from the boundary to the interior of the support).
For discrete distributions, the notion of a gradient is not well defined. We thus "look" ahead and behind one step (e.g. if the mode of a Poisson distribution is 9, we consider 8 and 10). If the step ahead/behind is still within the support of the distribution, we assert that the `log_prob` is smaller than at the mode.
For one-hot encoded distributions (currently just `OneHotCategorical`), we evaluate the underlying mode (i.e. encoded as an integral tensor), "advance" by one label to get another sample that should have lower probability using `other = (mode + 1) % event_size` and re-encode as one-hot. The resultant `other` sample should have lower probability than the mode.
Furthermore, Gamma, half Cauchy, and half normal distributions have their support changed from positive to nonnegative. This change is necessary because the mode of the "half" distributions is zero, and the mode of the gamma distribution is zero for `concentration <= 1`.
cc @fritzo
Pull Request resolved: https://github.com/pytorch/pytorch/pull/76690
Approved by: https://github.com/neerajprad
Summary:
I have experienced that sometimes both were in `__dict__`, but it chose to copy `probs` which loses precision over `logits`. This is especially important when training (bayesian) neural networks or doing other type of optimization, since the loss is heavily affected.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/18614
Differential Revision: D14793486
Pulled By: ezyang
fbshipit-source-id: d4ff5e34fbb4021ea9de9f58af09a7de00d80a63
Summary:
This PR removes the usage of _finfo defined in torch.distributions.utils and changes the call sites
to use torch.finfo instead
Differential Revision: D13451936
Pulled By: soumith
fbshipit-source-id: 6dbda3a6179d9407bc3396bf1a2baf3e85bc4cf2
Summary:
This adds tests in tests/test_distributions.py to ensure that all methods of `Distribution` objects are jittable.
I've replaced a few samplers with jittable versions:
- `.uniform_()` -> `torch.rand()`
- `.exponential_()` -> `-(-torch.rand()).log1p()`
- `.normal_()` -> `torch.normal(torch.zeros(...), torch.ones(...), ...)`
Some jit failures remain, and are marked in test_distributions.py
- `Cauchy` and `HalfCauchy` do not support sampling due to missing `.cauchy_()`
- `Binomial` does not support `.enumerate_support()` due to `arange` ignoring its first arg.
- `MultivariateNormal`, `LowRankMultivariateNormal` do not support `.mean`, `.entropy`
- [x] Currently some tests fail (I've skipped those) due to unavailability of `aten::uniform` and `aten::cauchy` in the jit. Can someone suggest how to add these? I tried to add declarations to `torch/csrc/ir.cpp` and `torch/csrc/passes/shape_analysis.cpp`, but that resulted in "Couldn't find operator" errors.
- [x] There are still lots of `TracerWarning`s that something doesn't match something. I'm not sure whether these are real.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/11560
Differential Revision: D9816327
Pulled By: apaszke
fbshipit-source-id: 72ec998ea13fc4c76d1ed003d9502e0fbaf728b8
Summary:
This adds a `.expand` method for distributions that is akin to the `torch.Tensor.expand` method for tensors. It returns a new distribution instance with batch dimensions expanded to the desired `batch_shape`. Since this calls `torch.Tensor.expand` on the distribution's parameters, it does not allocate new memory for the expanded distribution instance's parameters.
e.g.
```python
>>> d = dist.Normal(torch.zeros(100, 1), torch.ones(100, 1))
>>> d.sample().shape
torch.Size([100, 1])
>>> d.expand([100, 10]).sample().shape
torch.Size([100, 10])
```
We have already been using the `.expand` method in Pyro in our [patch](https://github.com/uber/pyro/blob/dev/pyro/distributions/torch.py#L10) of `torch.distributions`. We use this in our models to enable dynamic broadcasting. This has also been requested by a few users on the distributions slack, and we believe will be useful to the larger community.
Note that currently, there is no convenient and efficient way to expand distribution instances:
- Many distributions use `TransformedDistribution` (or wrap over another distribution instance. e.g. `OneHotCategorical` uses a `Categorical` instance) under the hood, or have lazy parameters. This makes it difficult to collect all the relevant parameters, broadcast them and construct new instances.
- In the few cases where this is even possible, the resulting implementation would be inefficient since we will go through a lot of broadcasting and args validation logic in `__init__.py` that can be avoided.
The `.expand` method allows for a safe and efficient way to expand distribution instances. Additionally, this bypasses `__init__.py` (using `__new__` and populating relevant attributes) since we do not need to do any broadcasting or args validation (which was already done when the instance was first created). This can result in significant savings as compared to constructing new instances via `__init__` (that said, the `sample` and `log_prob` methods will probably be the rate determining steps in many applications).
e.g.
```python
>>> a = dist.Bernoulli(torch.ones([10000, 1]), validate_args=True)
>>> %timeit a.expand([10000, 100])
15.2 µs ± 224 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)
>>> %timeit dist.Bernoulli(torch.ones([10000, 100]), validate_args=True)
11.8 ms ± 153 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
```
cc. fritzo, apaszke, vishwakftw, alicanb
Pull Request resolved: https://github.com/pytorch/pytorch/pull/11341
Differential Revision: D9728485
Pulled By: soumith
fbshipit-source-id: 3b94c23bc6a43ee704389e6287aa83d1e278d52f
Summary:
`__repr__` currently fails for distributions with lazy attributes in PyTorch master, throwing a `KeyError`. This fixes the issue.
**Additionally:**
- Added `logits` to `arg_constraints` for distributions that accept either `probs` or `logits`. This is both to have `__repr__` display the `logits` param when available, and to be able to do validation checks (e.g. NaN checks) when the logit parametrization is used. fritzo, alicanb - I think there were reasons why we had not done so in the first place, but I am unable to recall now. It passes all the tests, but let me know if there is something that I am missing at the moment.
- There are certain distributions, e.g. `OneHotCategorical` which won't show any parameters because it uses a `categorical` instance under the hood and neither `logits` / `probs` in `arg_constraints` are present in the instance's `__dict__`. This isn't addressed in this PR.
cc. vishwakftw, fritzo, nadavbh12, apaszke
Pull Request resolved: https://github.com/pytorch/pytorch/pull/11263
Differential Revision: D9654959
Pulled By: apaszke
fbshipit-source-id: 16f5b20243fe8e2c13e9c528050d4df0b8ea6e45
