FAI-952: Change dependencies to align with ODH workbench images (#143)

* Change dependencies to align with ODH workbench images
* Deprecate Python 3.7
* Replace Bokeh with jupyter_bokeh
* Replace XGBoost model with scikit-learn
* Use correct dev dependencies in CI
* Replace pytest-xdist with pytest-forked (Not all xdist features needed, just fork)

Author: ruivieira

.github/workflows/workflow.yml

@@ -7,7 +7,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        python-version: [ 3.7, 3.8, 3.9 ]
+        python-version: [ 3.8, 3.9 ]
         java-version: [ 11, 17 ]
         maven-version: [ '3.8.6' ]
     steps:
@@ -25,8 +25,8 @@ jobs:
         uses: ./.github/actions/build-core
       - name: Install TrustyAI Python package
         run: |
-          pip install -r requirements-dev.txt
           pip install .
+          pip install ".[dev]"
       - name: Lint
         run: |
           pylint --ignore-imports=yes $(find src/trustyai -type f -name "*.py")

pyproject.toml

@@ -5,7 +5,7 @@ description = "Python bindings to the TrustyAI explainability library."
 authors = [{ name = "Rui Vieira", email = "[email protected]" }]
 license = { file = "LICENSE" }
 readme = "README.md"
-requires-python = ">=3.7"
+requires-python = ">=3.8"
 
 keywords = ["trustyai", "xai", "explainability", "ml"]
 
@@ -23,32 +23,30 @@ classifiers = [
 dependencies = [
     "Jpype1==1.4.1",
     "pyarrow==7.0.0",
-    "matplotlib==3.5.1",
-    "pandas==1.2.5",
-    "numpy",
-    "bokeh==2.4.3"
+    "matplotlib~=3.6.3",
+    "pandas~=1.5.3",
+    "numpy~=1.24.1",
+    "jupyter-bokeh~=3.0.5"
 ]
 
 [project.optional-dependencies]
 dev = [
     "JPype1==1.4.1",
-    "black==21.6b0",
+    "black~=22.12.0",
     "click==8.0.4",
-    "joblib==1.0.1",
-    "jupyter==1.0.0",
-    "scikit-learn==0.24.2",
+    "joblib~=1.2.0",
+    "jupyterlab~=3.5.3",
+    "numpydoc==1.5.0",
+    "pyarrow==7.0.0",
     "pylint==2.15.6",
-    "pytest==6.2.4",
+    "pytest~=7.2.1",
+    "pytest-benchmark==4.0.0",
+    "pytest-forked~=1.6.0",
+    "scikit-learn~=1.2.1",
     "setuptools",
-    "wheel",
-    "xgboost==1.4.2",
     "twine==3.4.2",
-    "pyarrow==7.0.0",
-    "numpydoc",
-    "matplotlib==3.5.1",
-    "pandas==1.2.5",
-    "pytest-xdist",
-    "pytest-benchmark"
+    "wheel~=0.38.4",
+    "xgboost==1.4.2"
 ]
 
 [project.urls]

requirements-dev.txt

@@ -191,7 +191,7 @@ def _get_bokeh_plot(self, output_name) -> bokeh.models.Plot:
         )
         source = ColumnDataSource(lime_data_source)
         htool = HoverTool(
-            names=["bars"],
+            name="bars",
             tooltips="<h3>LIME</h3> {} saliency to {}: @saliency_colored".format(
                 feature_html("@feature"), output_html(output_name)
             ),

src/trustyai/explainers/lime.py

@@ -295,21 +295,21 @@ def _get_bokeh_plot(self, output_name):
 
         # create hovertools
         htool_fnull = HoverTool(
-            names=["fnull"],
+            name="fnull",
             tooltips=("<h3>SHAP</h3>Baseline {}: {}").format(
                 output_name, output_html("{:.2f}".format(fnull))
             ),
             line_policy="interp",
         )
         htool_pred = HoverTool(
-            names=["pred"],
+            name="pred",
             tooltips=("<h3>SHAP</h3>Predicted {}: {}").format(
                 output_name, output_html("{:.2f}".format(prediction))
             ),
             line_policy="interp",
         )
         htool_bars = HoverTool(
-            names=["bars"],
+            name="bars",
             tooltips="<h3>SHAP</h3> {} contributions to {}: @saliency_text".format(
                 feature_html("@feature"), output_html(output_name)
             ),

src/trustyai/explainers/shap.py

@@ -14,7 +14,7 @@
     LinearColorMapper,
     ColorBar,
     Tabs,
-    Panel,
+    TabPanel,
     Div,
     GridBox,
 )
@@ -324,17 +324,17 @@ def _get_plots(self):
             title = str(k)
             lime_tabbed = Tabs(
                 tabs=[
-                    Panel(child=lime_figures[k], title="LIME"),
-                    Panel(
+                    TabPanel(child=lime_figures[k], title="LIME"),
+                    TabPanel(
                         child=Div(text=LIME_TEXT.format(output_html(title))),
                         title="About LIME",
                     ),
                 ]
             )
             shap_tabbed = Tabs(
                 tabs=[
-                    Panel(child=shap_figures[k], title="SHAP"),
-                    Panel(
+                    TabPanel(child=shap_figures[k], title="SHAP"),
+                    TabPanel(
                         child=Div(text=SHAP_TEXT.format(output_html(title))),
                         title="About SHAP",
                     ),
@@ -343,8 +343,8 @@ def _get_plots(self):
 
             cf_tabbed = Tabs(
                 tabs=[
-                    Panel(child=cf_figures[k], title="Available Counterfactuals"),
-                    Panel(
+                    TabPanel(child=cf_figures[k], title="Available Counterfactuals"),
+                    TabPanel(
                         child=Div(
                             text=CF_TEXT.format(output_html(title)),
                         ),
@@ -369,7 +369,7 @@ def _get_plots(self):
                 trustyai_content,
                 sizing_mode="scale_width",
             )
-            tabs.append(Panel(child=joint, title=title))
+            tabs.append(TabPanel(child=joint, title=title))
 
         full_dash = Tabs(tabs=tabs, sizing_mode="scale_width")
         return full_dash

src/trustyai/utils/tyrus.py

@@ -6,8 +6,9 @@
 
 from pytest import approx
 import pandas as pd
-from xgboost import XGBClassifier
+from sklearn.ensemble import RandomForestClassifier
 import os
+import joblib
 import pathlib
 
 from trustyai.metrics.fairness.group import statistical_parity_difference, disparate_impact_ratio, \
@@ -23,8 +24,8 @@
 INCOME_DF_BIASED = pd.read_csv(os.path.join(TEST_DIR, "data/income-biased.zip"), index_col=False)
 INCOME_DF_UNBIASED = pd.read_csv(
     os.path.join(TEST_DIR, "data/income-unbiased.zip"), index_col=False)
-XGB_MODEL = XGBClassifier()
-XGB_MODEL.load_model(os.path.join(TEST_DIR, "models/income-xgb-biased.ubj"))
+
+XGB_MODEL = joblib.load(os.path.join(TEST_DIR, "models/income-xgd-biased.joblib"))
 
 
 def test_statistical_parity_difference_random():