Alican Bozkurt
9b6441ecbc
Implement Multinomial distribution ( #4624 )
2018-01-13 11:26:14 +01:00
Neeraj Pradhan
736190fc78
Allow broadcasting of value x params in Categorical ( #4614 )
2018-01-12 12:16:19 +01:00
Fritz Obermeyer
71b1120ba8
Fix bug in Dirichlet.rsample(); add tests ( #4602 )
...
* Fix bug in Dirichlet.rsample(); add tests
* Address review comments
2018-01-11 12:29:10 -05:00
Fritz Obermeyer
8cff8e93d2
Add torch.distributions.utils._finfo for numerical stability ( #4572 )
...
* Add torch.distributions.utils.finfo
* Make _finfo private
* Address review comments
* Simplify _finfo() to key on Storage type
2018-01-10 21:42:47 -05:00
Fritz Obermeyer
3a335427b0
Start framework for kl_divergence(-,-) in torch.distributions ( #4525 )
2018-01-09 09:44:59 +01:00
Vishwak Srinivasan
5d6a5cf3a7
Implementation of Gumbel Distribution ( #4517 )
2018-01-08 23:21:27 +01:00
Neeraj Pradhan
8fe3d287b2
Fix return type for Bernoulli enumerate_support ( #4529 )
2018-01-08 23:17:43 +01:00
Alican Bozkurt
c9bc6c2bc3
Implement Student's t-distribution ( #4510 )
2018-01-08 10:23:48 +01:00
Neeraj Pradhan
408c84de7c
Supporting logits as parameters in Bernoulli and Categorical ( #4448 )
...
* Supporting logits as parameters in Bernoulli and Categorical
* address comments
* fix lint
* modify binary_cross_entropy_with_logits
* address comments
* add descriptor for lazy attributes
* address comments
2018-01-05 03:45:05 -05:00
Fritz Obermeyer
a3e91515de
Declare constraints for distribution parameters and support ( #4450 )
2018-01-04 23:58:26 +01:00
Vishwak Srinivasan
1e76ade9dc
Implementation of Pareto Distribution ( #4459 )
2018-01-04 22:57:47 +01:00
Fritz Obermeyer
43ab911182
Improve precision of dirichlet_grad() approximation ( #4421 )
2018-01-02 20:53:47 +01:00
Alican Bozkurt
02e7eba309
Implement Chi2 distribution ( #4425 )
...
* add chi2
* add tests for chi2
* add randomized test comments
2018-01-01 19:41:18 -05:00
Fritz Obermeyer
6185b27cc6
Improve precision of standard_gamma_grad() ( #4369 )
2017-12-29 12:11:04 +01:00
Neeraj Pradhan
fa8de6b4f3
Adding the Cauchy distribution to torch.distributions
2017-12-29 11:57:21 +01:00
Fritz Obermeyer
5c33400dd3
Implement OneHotCategorical distribution ( #4357 )
2017-12-28 16:54:55 +01:00
Edward Z. Yang
4453a5402f
allow_inf on test_beta_log_prob ( #4354 )
...
* allow_inf on test_beta_log_prob
* Support allow_inf on assertAlmostEqual
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
2017-12-27 09:25:12 -05:00
Neeraj Pradhan
1b608eea4e
Fix distribution tests due to merge order ( #4351 )
2017-12-26 16:37:15 -05:00
Neeraj Pradhan
ffa7fab67f
Minor changes to test utils to catch type errors ( #4270 )
2017-12-26 10:08:33 +01:00
Neeraj Pradhan
0c4b3f4271
Adding Uniform distribution to PyTorch ( #4328 )
2017-12-23 15:14:44 +01:00
gchanan
3304185c6c
Fix test_gamma_sample_grad. ( #4327 )
2017-12-22 22:01:04 -05:00
Will Feng
6c4e97220a
disable test_gamma_sample_grad until it's fixed ( #4324 )
...
* disable test_gamma_sample_grad until we have SciPy in xenial builds
* skip test_gamma_sample_grad until it's fixed
2017-12-22 17:11:15 -05:00
Alican Bozkurt
d7e6ede784
Implement Laplace distribution ( #4289 )
2017-12-21 17:03:03 +01:00
Fritz Obermeyer
54e11639f9
Fix broken test_beta_log_prob in Python 3.6 ( #4261 )
2017-12-20 19:41:03 -05:00
Fritz Obermeyer
0bc1505f34
Implement .entropy() methods for all distributions ( #4268 )
2017-12-20 14:06:01 +01:00
Fritz Obermeyer
69265ea5bc
Ensure gamma samples are positive ( #4262 )
2017-12-20 10:17:31 +01:00
Alican Bozkurt
94ff31f54d
Implement Exponential distribution ( #4234 )
...
* add exponential distribution
* add exponential tests
* fix default val of sample_shape
* lambd->rate
* updates per review
* remove notes, keep failure_rate same in exponential test
2017-12-18 16:44:35 -05:00
Fritz Obermeyer
ee98e7a82e
Implement Dirichlet and Beta distributions ( #4117 )
2017-12-18 19:11:37 +01:00
Neeraj Pradhan
fac711c238
Provide full support for distribution shapes ( #4193 )
2017-12-15 12:41:08 +01:00
Alican Bozkurt
7f25fff2fe
add reparameterization, combine sample and sample_n ( #4142 )
2017-12-15 00:25:39 +01:00
Neeraj Pradhan
4f4e0df68f
Allow for broadcasting of distribution parameters ( #4140 )
2017-12-14 09:37:03 +01:00
Fritz Obermeyer
0ab68b8db4
Implement .enumerate_support() for Bernoulli, Categorical distributions ( #4129 )
2017-12-13 13:01:05 +01:00
Fritz Obermeyer
05ebd21a36
Implement reparameterized gradient for Gamma sampler ( #3978 )
2017-12-11 03:32:15 -08:00
ngimel
0d68ce9383
Use integer division to fix failing test
2017-12-04 21:16:09 -08:00
Fritz Obermeyer
165d0897e4
Implement distributions.Gamma ( #3841 )
2017-12-02 01:10:08 +01:00
Fritz Obermeyer
1f64c2ef91
Rename pyro.distributions.Multinomial -> .Categorical ( #3766 )
...
* Rename distributions.Multinomial -> distributions.Categorical
* Rename Multinomial -> Categorical
* Update docs
* Update variable.py
* Update distributions.py
* Update variable.py
2017-11-18 16:10:07 -05:00
Geoffrey Roeder
dce525ab6b
adds sample_n function ( #3249 )
...
* adds sample_n function
* fixes style issues
* uses more efficient api calls
* fix bug where transpose applied to 1 dimension
2017-10-31 09:04:05 -04:00
Sam Gross
d9b89a352c
Replace StochasticFunctions v2 ( #3165 )
...
This removes the StochasticFunctions for bernoulli, multinomial, and
normal and replaces them with classes in the torch.distributions
package. Each distribution supports the differentiable log_prob function
that returns the log of the pdf/pmf of the samples.
The current StochasticFunction implementation has a few problems: it can
be painful to use when there are multiple stochastic outputs which need
to be back-propagated through. It also requires that we store grad_fns
on Variables that have requires_grad=False in order to find stochastic
nodes.
2017-10-19 15:05:07 -04:00
