Tried to infer the onnx version of the RF model but suffered fiasco

README.md

@@ -92,7 +92,14 @@ If you want to infer a previously trained model, make sure you've placed the che
 poetry run python3 commands.py infer --config_name [config_name_without_extension]
 ```
 
-### Deployment
+### Deployment with MLflow
+
+**Warning! This feature works stably only with the CatBoost model.** Predictions of the
+onnx version of the Random Forest differ from the original one (see
+[this](https://github.com/onnx/sklearn-onnx/issues/1047#issuecomment-1851837537)).
+Moreover, I was not able to infer the onnx version with MLflow (although everything worked
+fine with the `mlflow.sklearn` flavour as you can see in the `hw2` version of the
+repository).
 
 In order to deploy a trained model, run:
 
@@ -102,8 +109,10 @@ poetry run mlflow models serve -p 5001 -m checkpoints/mlflow_[model_type]_ckpt/
 
 where `[model_type]` is `cb` or `rf`.
 
-After this, it is possible to send requests to the model. I've created a script for the
-correct json generation with the first example from the training set:
+After this, it is possible to send requests to the model. I've created a script to
+generate the correct json containing the first example from the training set, but the json
+itself is in the repository, so you can skip this step. If you want to generate the json
+by yourself, run:
 
 ```
 poetry run python3 create_example_request.py create_example_request

create_example_request.py

@@ -11,9 +11,9 @@ def create_example_request() -> None:
                 "season": "spring",
                 "month": 1,
                 "hour": 0,
-                "holiday": False,
+                "holiday": 0,
                 "weekday": 6,
-                "workingday": False,
+                "workingday": 0,
                 "weather": "clear",
                 "temp": 9.84,
                 "feel_temp": 14.395,

example_request.json

@@ -14,6 +14,6 @@
       "humidity",
       "windspeed"
     ],
-    "data": [["spring", 1, 0, false, 6, false, "clear", 9.84, 14.395, 0.81, 0.0]]
+    "data": [["spring", 1, 0, 0, 6, 0, "clear", 9.84, 14.395, 0.81, 0.0]]
   }
 }

mlopscourse/train.py

@@ -65,10 +65,10 @@ def train(self) -> None:
                 )
                 model.log_fis_and_metrics(exp_id, X_train.columns)
             else:
-                model_onnx = to_onnx(model.model, X=X_train.iloc[:1], verbose=1)
+                model_onnx = to_onnx(model.model, X=X_train[:1], verbose=1)
                 mlflow.onnx.save_model(
                     model_onnx,
-                    f"checkpoints/mlflow_{self.cfg.model.name}_onnx_ckpt/",
+                    f"checkpoints/mlflow_{self.cfg.model.name}_ckpt/",
                     signature=signature,
                 )
                 model.log_fis_and_metrics(exp_id, X_train, y_train)

pyproject.toml

@@ -15,7 +15,7 @@ catboost = "^1.2.2"
 fire = "^0.5.0"
 hydra-core = "^1.3.2"
 mlflow = "^2.8.1"
-skl2onnx = "^1.15.0"
+skl2onnx = "^1.16.0"
 onnxruntime = "^1.16.3"
 
 [tool.poetry.group.dev.dependencies]