Update readme and fix lgbm bug, no code modifications

Author: SimeonStoykovQC

benchmarks/README.md

@@ -0,0 +1,29 @@
+## Running the benchmarks
+
+Create a conda environment and activate it:
+```commandline
+conda env create -f environment.yml
+conda activate lleaves
+```
+
+Modify [setup.py](../setup.py) to not exclude the `benchmark` package:
+```python
+packages=find_packages(exclude=[])  # Used to be: exclude=["benchmark"]
+```
+
+Install the packages in the environment (optionally, in development mode with `-e`):
+```commandline
+python -m pip install --no-build-isolation -e .
+```
+
+Generate the test data and train the necessary models:
+```commandline
+benchmarks/data/setup_data.sh
+python benchmarks/train_NYC_model.py
+```
+
+Finally, run the benchmarks from within the benchmarks folder:
+```commandline
+cd benchmarks
+python benchmark.py
+```

environment.yml

@@ -10,7 +10,7 @@ dependencies:
   # testing
   - pytest
   - hypothesis
-  - lightgbm
+  - lightgbm<4  # Pin needed until https://github.com/onnx/onnxmltools/issues/645 is fixed
   - pre-commit
   - pyarrow
   - pandas