FAI-917: Add error message capturing within Python models (#137)

* Added error catching wrapper to model

* linting and black

* fixed typo within predict_fun message

* improved error log formatting

* lazy logging

Author: RobGeada

pyproject.toml

@@ -64,6 +64,7 @@ build-backend = "setuptools.build_meta"
 package-dir = { "" = "src" }
 
 [tool.pytest.ini_options]
+log_cli = true
 addopts = '-m="not block_plots"'
 markers = [
     "block_plots: Test plots will block execution of subsequent tests until closed"

src/trustyai/model/__init__.py

@@ -2,6 +2,8 @@
 # pylint: disable = unused-import, wrong-import-order
 # pylint: disable = consider-using-f-string
 """General model classes"""
+import logging
+import traceback
 import uuid as _uuid
 from typing import List, Optional, Union, Callable
 import pandas as pd
@@ -322,7 +324,8 @@ def __init__(self, predict_fun, **kwargs):
                 transfer between Java and Python. If false, Arrow will be automatically used in
                 situations where it is advantageous to do so.
         """
-        self.predict_fun = predict_fun
+
+        self.predict_fun = self._error_catcher(predict_fun)
         self.kwargs = kwargs
 
         self.prediction_provider_arrow = None
@@ -332,6 +335,24 @@ def __init__(self, predict_fun, **kwargs):
         # set model to use non-arrow by default, as this requires no dataset information
         self._set_nonarrow()
 
+    def _error_catcher(self, predict_fun):
+        """Wrapper for predict function to capture errors to Python logger before the JVM dies"""
+
+        def wrapper(x):
+            try:
+                return predict_fun(x)
+            except Exception as e:
+                logging.error(
+                    " Fatal runtime error within the `predict_fun` supplied to trustyai.Model"
+                )
+                logging.error(
+                    " The error message has been captured and reproduced below:"
+                )
+                logging.error(" %s", traceback.format_exc())
+                raise e
+
+        return wrapper
+
     @property
     def dataframe_input(self):
         """Get dataframe_input kwarg value"""
@@ -483,7 +504,7 @@ def __enter__(self):
                 self.previous_model_state = self.model.prediction_provider
                 self.model._set_arrow(self.paradigm_input)
 
-        def __exit__(self, exit_type, value, traceback):
+        def __exit__(self, exit_type, value, tb):
             if self.model_is_python:
                 self.model.prediction_provider = self.previous_model_state
 
@@ -502,7 +523,7 @@ def __enter__(self):
                 self.previous_model_state = self.model.prediction_provider
                 self.model._set_nonarrow()
 
-        def __exit__(self, exit_type, value, traceback):
+        def __exit__(self, exit_type, value, tb):
             if self.model_is_python:
                 self.model.prediction_provider = self.previous_model_state
 

tests/general/test_model.py

@@ -1,5 +1,6 @@
 # pylint: disable=import-error, wrong-import-position, wrong-import-order, invalid-name
 """Test model provider interface"""
+from trustyai.explainers import LimeExplainer
 
 from common import *
 from trustyai.model import Model, Dataset, feature
@@ -46,3 +47,14 @@ def test_cast_output_arrow():
         output_val = m.predictAsync(pis).get()
         assert len(output_val) == 25
 
+
+def test_error_model(caplog):
+    """test that a broken model spits out useful debugging info"""
+    m = Model(lambda x: str(x) - str(x))
+    try:
+        LimeExplainer().explain(0, 0, m)
+    except Exception:
+        pass
+
+    assert "Fatal runtime error" in caplog.text
+    assert "TypeError: unsupported operand type(s) for -: 'str' and 'str'" in caplog.text