Conversation
Version 3.4 RC 1
Update short and long descriptions in setup.py
fix missing test_size argument
* Removed redundant specification of version number * Replaced pymc3 import in setup with regex function for getting version * Removed typo
Version 3.4 RC2
Increment version to 3.4
Increment version to 3.4.1
Fix for trace of subset of variables
* fix GARCH11: missing [] in tt.concatenate() * add test for GARCH11 and update docstring to current standards Closes #2938
* Added check_test_point function * Added change to RELEASE-NOTES * Fixed typo
* updated 'updating_priors' notebook * updated 'api_quickstart' notebook * updated 'AR' notebook * updated 'Bayes_factor' notebook * Added check_test_point method to Model (#2936) * Added check_test_point function * Added change to RELEASE-NOTES * Fixed typo * updated 'runing' typos to 'running' in these 4 notebooks
* Update arbitrary_stochastic.py Move custom logp outside of model block to avoide `AttributeError: Can't pickle local object 'build_model.<locals>.logp'` * Reduce divergence - Change scale of prior - increase tuning and target_accept Note that there are still ~5 divergences per 1000 samples after tunning, further reparameterization is necessary. * remove unnecessary comments * Change prior for model No more divergence with more informative priors. * Use more realistic prior to avoid divergence * Change init for sample avoid `ValueError: Mass matrix contains zeros on the diagonal.`
Currently, the default for InverseGamma(a, b) is the mean, but when a, b < 1 the mean is unspecify and the RV assign 0. as the default, which crash the sampler. This PR change the default to the mode.
* Optimized logpt computation I change the logpt computation in #2949 to fix #2948, however, it slows down the speed as some graph optimization is turned off (those optimization is originally cause the error in #2948). I am trying with a differen approach here. @ColCarroll * fix test
Replaced sigma argument with noise in MarginalSparse.marginal_likelihood
We see user having a hard time debugging their model when the error `Mass matrix contains zeros on the diagonal. Some derivatives might always be zero` arise. See eg https://discourse.pymc.io/t/unsupervised-clustering-mass-matrix-contains-zeros-on-the-diagonal/1222/. This PR prints out where the error is, so user can easier address the bug by eg changing the scale of those RVs
the index is after .ravel()
Better error message for Mass matrix contains zeros
This small correction is in response to issue Stochastic Volatility Example #2566. It concerns the two different examples of the Volatility Model used in the PyMC3 notebooks to introduce users to the wonder that is Bayesian modeling. In getting_started.ipynb, under: --Case Study 1: Stochastic volatility ---The Model the model specification uses `y_i` to represent the daily precentage returns. However, later on in the `with pm.Model() as sp500_model:` block the dummy variable `r` is used to represent the daily returns as well as the tensor variable name. The second correction also concerns the useage of `s_i` to represent the volatility process in daily returns. In both Volatility Process walkthroughs (there is a stand-alone notebook dedicated to it) the model specs treat `s_i` as the standard deviation of the StudentT distribution used to model the log-returns. In the PyMC3 API docs on the StudentT, the distribution is defined with `lambda` representing the precision. This is why the `volatility_process` variable is mapped from `pm.math.exp(-2 * s)` in the `pm.Model()` block. However, when the returns were defined in the model block the **kwarg was `lambda=1/volatility_process`. This has been fixed. Thanks to @twiecki and the OP for highlighting this error.
DOCS: 2973 Fix to doctring based on new behaviour in `distribution.de…
Add save_trace and load_trace
Fixing up the test and implementation Adding other draw_values Small test fix
* Fixes to Rice test. * make test case for Rice distribution pass * Improve numerical stability of rice distribution function for large values of nu and value Use ((x-b)**2)/2 + xb instead of (x**2 + b**2)/2 in the pdf for the rice distribution and include the np.exp(-xb) in the i0e to match the scipy implementation * Change test domain for sd of rice distribution to pass tests with float32
* Add save/load ndarray to release notes, small bug fix * Be careful about deleting things
So detecting transform name is more consistence. Might not be backward compatible if users are accessing the transformed RV of `cov_cholesky_cov_pack__` in custom code.
Also rerun notebook and add additional explanation regarding to changes of variables.
Chain transformation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
16 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.