WIP notebooks cleanup #1465
WIP notebooks cleanup #1465
Conversation
|
The Cox survival notebook looks pretty sad next to Austin's survival notebook. I may have a go at moving whatever nuggets might be in the Cox notebook over to the newer one, and remove |
|
Ohh there's an error. However all sounds good to me. |
* More robust transformations and scaling adjustments * Fixed bad name for scaling bound * Tests sd rather than tau (or both) in normal logp * Added default scaling_bound for adjust_precision
* Added warnings for non-positive support in positive parameters for continuous distributions * Improved message from assert_nonpos_support
Generalize find_MAP to accept more scipy.optimize functions
Ignore SQLite file generated by tests
Bugfix multinomial vector
* Fix parallel sampling * Everything is good again
Changed check for non-positive support to check for negative support
|
I notice that the Cox model notebook does not work. The ADVI estimates do not help the sampler at all, and the NUTS sampling is painfully slow. Further comparison shows that the NUTS sampler gives different (wrong) values compared to Metropolis. |
|
Does anyone want to debug the Cox model? @AustinRochford maybe |
|
I can take a look at the Cox model, but not for a few days, as I will be traveling this weekend. |
|
This seems to have commits I made to master, does this need a rebase? |
|
Those appeared after I rebased to master. Dunno. |
|
Maybe you need to edit manually your rebase - or something like that. I've had similar problems in the past no idea why they occur. |
|
Closing in favor of #1535 |
5 participants
Making sure all notebooks can run, and deleting poorly-maintained models.
Status:
discrete_find_MAP.ipynbis broken, related to find_MAP fails when starting values passed #1466cox_modelnot generating correct resultsNUTS_scaling_using_ADVI.ipynbusing current master. Broken for NUTS.On my setup, the rest of the notebooks run as advertised. I've cleaned up a couple of them a little.