Merge pull request #49 from predict-idlab/dev-jonas

💨 updating dependencies - python 3.12 compatible

Author: JarneVerhaeghe

.github/workflows/test.yml

@@ -27,7 +27,7 @@ jobs:
       fail-fast: false
       matrix:
         os: ['ubuntu-latest']  # TODO - do we add other OSes?
-        python-version: ["3.7", "3.8", "3.9", "3.10"]
+        python-version: ["3.9", "3.10", "3.11", "3.12"]
     defaults:
       run:
         shell: bash

poetry.lock

@@ -1,5 +1,6 @@
 __author__ = "Jarne Verhaeghe, Jeroen Van Der Donckt"
 
+import gc
 import warnings
 from abc import ABC
 from copy import copy
@@ -12,8 +13,6 @@
 from sklearn.model_selection import train_test_split
 from tqdm.auto import tqdm
 
-import gc
-
 
 class ShapExplainer(ABC):
     """Interface class for a (POWERshap explainer class."""
@@ -172,7 +171,8 @@ def explain(
             Shap_values = np.abs(Shap_values)
 
             if len(np.shape(Shap_values)) > 2:
-                Shap_values = np.max(Shap_values, axis=0)
+                # Shap_values = np.max(Shap_values, axis=0)
+                Shap_values = np.max(Shap_values, axis=0).T
 
             # TODO: consider to convert to even float16?
             Shap_values = np.mean(Shap_values, axis=0).astype("float32")
@@ -185,7 +185,6 @@ def explain(
             # on every iteration.
             gc.collect()
 
-
         shaps = np.array(shaps)
 
         return pd.DataFrame(data=shaps, columns=X.columns.values)
@@ -350,7 +349,7 @@ def supports_model(model) -> bool:
     def _fit_get_shap(self, X_train, Y_train, X_val, Y_val, random_seed, **kwargs) -> np.array:
         import tensorflow as tf
 
-        tf.compat.v1.disable_v2_behavior()  # https://github.com/slundberg/shap/issues/2189
+        # tf.compat.v1.disable_v2_behavior()  # https://github.com/slundberg/shap/issues/2189
 
         # Fit the model
         PowerShap_model = tf.keras.models.clone_model(self.model)
@@ -359,6 +358,7 @@ def _fit_get_shap(self, X_train, Y_train, X_val, Y_val, random_seed, **kwargs) -
             loss=kwargs["loss"],
             optimizer=kwargs["optimizer"],
             metrics=metrics if metrics is None else [metrics],
+            # run_eagerly=True,
         )
         _ = PowerShap_model.fit(
             X_train,

powershap/shap_wrappers/shap_explainer.py

@@ -9,18 +9,20 @@ repository = "https://github.com/predict-idlab/powershap"
 keywords = ["feature selection", "shap", "data-science", "machine learning"]
 
 [tool.poetry.dependencies]
-python = ">=3.7.1,<3.11"
+python = ">=3.9,<=3.12"  
 catboost = "^1.0.5"
 statsmodels = "^0.13.2"
-numpy = "^1.21"
 pandas = ">=1.3"
-shap = "^0.40"
+shap = "^0.45.0"
 scikit-learn = "*"
 
 [tool.poetry.dev-dependencies]
 pytest = "^7.1.1"
 pytest-cov = "^3.0.0"
-tensorflow = "^2.8.0"
+tensorflow = [
+    { version = "<2.16.0", python = "<3.12"}
+    # support 3.12 when shap support tf 2.16 - https://github.com/shap/shap/issues/3562
+]
 lightgbm = "^3.3.2"
 xgboost = "^1.6.0"
 black = "^22.12.0"
@@ -54,5 +56,5 @@ multi_line_output = 3
 include_trailing_comma = true
 force_grid_wrap = 0
 combine_as_imports = true
-color_output = true
+color_output = false
 skip = "tests/toml_test.py"

pyproject.toml

@@ -6,6 +6,7 @@
 
 from .conftest import dummy_classification, dummy_regression
 
+# tf.compat.v1.enable_eager_execution()
 
 def test_deep_learning_class_powershap(dummy_classification):
     X, y = dummy_classification

tests/test_deep_learning_powershap.py

@@ -1,5 +1,7 @@
 __author__ = "Jeroen Van Der Donckt"
 
+import hashlib
+
 import numpy as np
 import pandas as pd
 from catboost import CatBoostClassifier, CatBoostRegressor
@@ -8,8 +10,6 @@
 
 from .conftest import dummy_classification, dummy_regression
 
-import hashlib
-
 ### DEFAULT MODEL & AUTOMATIC MODE
 
 
@@ -251,8 +251,7 @@ def test_powershap_cv_groupshufflesplit(dummy_classification):
 
 
 def test_no_mutate_df(dummy_classification):
-    """Ensure that powershap fit doesn't mutate an input pandas dataframe.
-    """
+    """Ensure that powershap fit doesn't mutate an input pandas dataframe."""
     X, y = dummy_classification
     n_informative = sum([c.startswith("informative") for c in X.columns])
     assert n_informative > 0, "No informative columns in the dummy data!"
@@ -277,8 +276,7 @@ def test_no_mutate_df(dummy_classification):
 
 
 def test_no_mutate_numpy(dummy_classification):
-    """Ensure that powershap fit doesn't mutate an input numpy array.
-    """
+    """Ensure that powershap fit doesn't mutate an input numpy array."""
     X, y = dummy_classification
 
     X = X.to_numpy()