Test results (3.8)

201 tests   201 ✅  1m 3s ⏱️
  1 suites    0 💤
  1 files      0 ❌

Results for commit f56ca5c.

♻️ This comment has been updated with latest results.

Test results (3.9)

201 tests   201 ✅  1m 4s ⏱️
  1 suites    0 💤
  1 files      0 ❌

Results for commit f56ca5c.

♻️ This comment has been updated with latest results.

Test results (3.11)

201 tests   201 ✅  1m 5s ⏱️
  1 suites    0 💤
  1 files      0 ❌

Results for commit f56ca5c.

♻️ This comment has been updated with latest results.

Test results (3.10)

201 tests   201 ✅  1m 4s ⏱️
  1 suites    0 💤
  1 files      0 ❌

Results for commit f56ca5c.

♻️ This comment has been updated with latest results.

Test results (3.12)

201 tests   201 ✅  1m 33s ⏱️
  1 suites    0 💤
  1 files      0 ❌

Results for commit f56ca5c.

♻️ This comment has been updated with latest results.

Pull Request Test Coverage Report for Build 9098719025

Details

• 122 of 129 (94.57%) changed or added relevant lines in 6 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage decreased (-1.0%) to 98.932%

Totals
Change from base Build 8802553922:	-1.0%
Covered Lines:	1049
Relevant Lines:	1056

💛 - Coveralls

@ulysseherbach Simulating from a network parameter inferred by Hartree crash.
It happens inside the file bursty/base.py at line 66 with ValueError: scale < 0 or
at line 75 with ValueError: probabilities are not non-negative.

It seems kon and/or kon_bound returns negative value because I think the network parameter is not set correctly in hartree/base.py.

Was hartree returning normalized values before for a[0], a[1] ?
Do we need to multiply or divide them per degradation_rna ?

We might need to revert what we did for the coverage because we get often this WARNING: a <= 0 from infer_kinetics and the result of fit in notebook 2 seems wrong

@nseyler1 OK I see, this is an important problem. In fact I think there are 3 distinct problems:

1. infer_kinetics seems broken indeed, so maybe we need to revert to previous version, at least temporarily?
2. Hartree used to return normalized values (a[0] and a[1] by definition) which is not the case now (I've just "fixed" it)
3. In fact there is a more important problem with Hartree (and all inference methods) : they should not define a new parameter (which will have default values and break everything...) but rather use (and modify in place) the NetworkParameter object from the related NetworkModel instance during fit call. What would be the best implementation for that? Maybe passing an existing NetworkParameter object to the Inference.runmethod?