[REVIEW] Add float64 warning when loading LightGBM model (#2874)

* Add float64 warning when loading LightGBM model * Update changelog
rapidsai · Sep 26, 2020 · fae1cba · fae1cba
1 parent 17b55cb
commit fae1cba
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 0 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -7,6 +7,7 @@
 - PR #2789: Dask LabelEncoder
 - PR #2152: add FIL C++ benchmark
 - PR #2638: Improve cython build with custom `build_ext`
+- PR #2874: Issue warning for degraded accuracy with float64 models in Treelite
 
 ## Improvements
 - PR #2873: Remove empty marker kernel code for NVTX markers

diff --git a/python/cuml/fil/fil.pyx b/python/cuml/fil/fil.pyx
@@ -133,6 +133,9 @@ cdef class TreeliteModel():
                 err = TreeliteGetLastError()
                 raise RuntimeError("Failed to load %s (%s)" % (filename, err))
         elif model_type == "lightgbm":
+            logger.warn("Treelite currently does not support float64 model"
+                        " parameters. Accuracy may degrade relative to"
+                        " native LightGBM invocation.")
             res = TreeliteLoadLightGBMModel(filename_bytes, &handle)
             if res < 0:
                 err = TreeliteGetLastError()