[FEA] Add predict_proba() to XGBoost-style models in FIL C++ #2894
Conversation
|
Please update the changelog in order to start CI tests. View the gpuCI docs here. |
levsnv
changed the title
levsnv
changed the title
|
To copy the Slack response: I definitely agree that there can be a neat single template for softmax, and I will implement the suggestions above |
levsnv
added
the
4 - Waiting on Reviewer
| using BlockReduceHost = | ||
| typename cub::BlockReduce<vec<NITEMS, float>, FIL_TPB, | ||
| cub::BLOCK_REDUCE_WARP_REDUCTIONS, 1, 1, 600>; | ||
| size_t block_reduce_footprint_host() { |
There was a problem hiding this comment.
I was changing half the cases where this was used, as well as deleting the device-side template using statements. This meant the host ones were left alone and did not need to define class separately from footprint. This allowed to implement smem footprint additions for GROVE_PER_CLASS_* in a much more readable way (see below).
levsnv
added
4 - Waiting on Author
levsnv
changed the title
|
now conflicts with #3088 for combo ML::fil::output_t combo values (but a simple update to resolve merge conflict) |
This reverts commit 8581222.
levsnv
added
4 - Waiting on Author
|
depends on #3582 |
| # FIL doesn't yet support predict_proba() for multi-class | ||
| # TODO: Add a test for predict_proba() when it's supported | ||
| gbm_preds = bst.predict(X) | ||
| gbm_preds = gbm_preds.argmax(axis=1) |
There was a problem hiding this comment.
since lightgbm doesn't support probabilities without the sklearn API, using it here for both predictions
levsnv
added
3 - Ready for Review
|
rerun tests |
levsnv
added
4 - Waiting on Reviewer
|
@JohnZed PR looks good to me, ready to merge and has addressed all feedback, when you have a sec could you take a look? |
|
@gpucibot merge |
levsnv
added
4 - Waiting on Author
Codecov Report
@@ Coverage Diff @@
## branch-0.19 #2894 +/- ##
===============================================
+ Coverage 80.70% 80.74% +0.03%
===============================================
Files 227 227
Lines 17615 17737 +122
===============================================
+ Hits 14217 14322 +105
- Misses 3398 3415 +17
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
No description provided.