[MRG+2] DOC examples added to the rest of sklearn/covariance classes#11732
[MRG+2] DOC examples added to the rest of sklearn/covariance classes#11732jnothman merged 85 commits intoscikit-learn:masterfrom
Conversation
|
So apparently the results depend on something other than the random seed. The output is deterministic, as long as we don't change the setup. But my personal computer, a server I used to test, and the setup travis uses, do not agree with each other and each give a different output. In this PR the values are the ones reported by travis, and that's why the travis tests pass. But users may get slightly different results based on their setup. I suspect it may depend on lower lever libraries used (blas, lapack, etc), but not sure. |
4099adb to
a2fede3
Compare
|
What kind of differences are we talking about? |
|
@jnothman |
|
ping me about this a while after release and we will try understand if it's
a bug... i suppose it's probably not
|
sklearn/covariance/graph_lasso_.py
Outdated
| ... largest_coef=.7, | ||
| ... random_state=0) | ||
| >>> cov = linalg.inv(prec) | ||
| >>> d = np.sqrt(np.diag(cov)) |
There was a problem hiding this comment.
I think this normalisation deserves a comment, and ideally would be g performed with some existing scipy function
There was a problem hiding this comment.
I found other mistakes in the example. They'd make the example kinda too long, and our docs already show this example. So changed it to be more like the other covariance examples we have, with a manually set sparse covariance matrix, if that's okay.
9aa882c to
a60ea6e
Compare
* Added tips for reading the code base * Put it in contributing.rst * Added bullet point about inheritance
* [scipy-dev] manually scramble the indices * [scipy-dev] review comment
… examples/graphlasso
…cikit-learn#11732)" This reverts commit be36a44.
…cikit-learn#11732)" This reverts commit be36a44.
See #3846
Examples added to GraphicalLasso and GraphicalLassoCV in
sklearn/covariance/graph_lasso_.py