validmind 2.13.9__tar.gz → 2.13.12__tar.gz

{validmind-2.13.9 → validmind-2.13.12}/.gitignore

@@ -189,12 +189,8 @@ lending_club_loan_data_*.csv
 
 # Pickle files
 *.pkl
-# Sample application scorecard model for validation notebook — do not remove!
-!notebooks/use_cases/validation/xgb_model_champion.pkl
 # Sample logistic regression model for validation series — do not remove!
 !notebooks/tutorials/validation/lr_model_champion.pkl
-# Sample XGBoost model for validation quickstart — do not remove!
-!notebooks/quickstart/xgboost_model_champion.pkl
 
 notebooks/llm/datasets/*.jsonl
 

{validmind-2.13.9 → validmind-2.13.12}/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.4
+Metadata-Version: 2.5
 Name: validmind
-Version: 2.13.9
+Version: 2.13.12
 Summary: ValidMind Library
 Author-email: Andres Rodriguez <andres@validmind.ai>, Juan Martinez <juan@validmind.ai>, Anil Sorathiya <anil@validmind.ai>, Luis Pallares <luis@validmind.ai>, John Walz <john@validmind.ai>
 License:                          DUAL LICENSE NOTICE
@@ -735,6 +735,7 @@ Requires-Dist: pyarrow<23,>=22; (python_version >= '3.14') and extra == 'all'
 Requires-Dist: pycocoevalcap; extra == 'all'
 Requires-Dist: ragas<=0.2.7,>=0.2.3; extra == 'all'
 Requires-Dist: rouge>=1; extra == 'all'
+Requires-Dist: scikit-learn<1.8; extra == 'all'
 Requires-Dist: scipy; extra == 'all'
 Requires-Dist: scorecardpy==0.1.9.6; extra == 'all'
 Requires-Dist: sentencepiece<0.3.0,>=0.2.0; extra == 'all'
@@ -744,7 +745,7 @@ Requires-Dist: statsmodels<0.15.0,>=0.14.2; extra == 'all'
 Requires-Dist: textblob<0.19.0,>=0.18.0.post0; extra == 'all'
 Requires-Dist: torch>=2.0.0; extra == 'all'
 Requires-Dist: transformers<5.0.0,>=4.32.0; extra == 'all'
-Requires-Dist: xgboost<3,>=1.5.2; extra == 'all'
+Requires-Dist: xgboost<3.1,>=1.5.2; extra == 'all'
 Provides-Extra: credit-risk
 Requires-Dist: scorecardpy==0.1.9.6; extra == 'credit-risk'
 Requires-Dist: setuptools<82; extra == 'credit-risk'
@@ -802,7 +803,8 @@ Requires-Dist: arch>=7.0.0; extra == 'stats'
 Requires-Dist: scipy; extra == 'stats'
 Requires-Dist: statsmodels<0.15.0,>=0.14.2; extra == 'stats'
 Provides-Extra: xgboost
-Requires-Dist: xgboost<3,>=1.5.2; extra == 'xgboost'
+Requires-Dist: scikit-learn<1.8; extra == 'xgboost'
+Requires-Dist: xgboost<3.1,>=1.5.2; extra == 'xgboost'
 Description-Content-Type: text/markdown
 
 # ValidMind Library

{validmind-2.13.9 → validmind-2.13.12}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "validmind"
-version = "2.13.9"
+version = "2.13.12"
 description = "ValidMind Library"
 readme = "README.pypi.md"
 requires-python = ">=3.9,<3.15"
@@ -40,7 +40,8 @@ dependencies = [
 [project.optional-dependencies]
 all = [
   "torch (>=2.0.0)",
-  "xgboost (>=1.5.2,<3)",
+  "xgboost (>=1.5.2,<3.1)",
+  "scikit-learn (<1.8)",
   "transformers (>=4.32.0,<5.0.0)",
   "pycocoevalcap",
   "ragas (>=0.2.3,<=0.2.7)",
@@ -108,7 +109,10 @@ nlp = [
 ]
 pytorch = ["torch (>=2.0.0)"]
 stats = ["scipy", "statsmodels (>=0.14.2,<0.15.0)", "arch (>=7.0.0)"]
-xgboost = ["xgboost (>=1.5.2,<3)"]
+xgboost = [
+  "xgboost (>=1.5.2,<3.1)",
+  "scikit-learn (<1.8)",
+]
 explainability = [
   "shap (>=0.46.0)",
   "numba (>=0.65.1) ; python_version >= '3.14'",

validmind-2.13.12/validmind/__version__.py

@@ -0,0 +1 @@
+__version__ = "2.13.12"

{validmind-2.13.9 → validmind-2.13.12}/validmind/ai/utils.py

@@ -8,7 +8,7 @@ import os
 from openai import AzureOpenAI, OpenAI
 
 from ..logging import get_logger
-from ..utils import md_to_html
+from ..utils import is_html, md_to_html
 
 logger = get_logger(__name__)
 
@@ -38,6 +38,7 @@ class DescriptionFuture:
 
     def __init__(self, future):
         self._future = future
+        self.markdown_source = None
 
     def get_description(self):
         if isinstance(self._future, tuple):
@@ -46,7 +47,9 @@ class DescriptionFuture:
             # This will block until the future is completed
             description = self._future.result()
 
-        return md_to_html(description[0], mathml=True), description[1]
+        source, was_generated = description
+        self.markdown_source = source if not is_html(source) else None
+        return md_to_html(source, mathml=True), was_generated
 
 
 def _get_google_api_key():

{validmind-2.13.9 → validmind-2.13.12}/validmind/api_client.py

@@ -588,6 +588,7 @@ async def alog_metadata(
     text: Optional[str] = None,
     _json: Optional[Dict[str, Any]] = None,
     section_id: Optional[str] = None,
+    text_format: Optional[str] = None,
 ) -> Dict[str, Any]:
     """Logs free-form metadata to ValidMind API.
 
@@ -597,6 +598,9 @@ async def alog_metadata(
         _json (dict, optional): Free-form key-value pairs to assign to the metadata. Defaults to None.
         section_id (str, optional): Section ID to append the text block to when the
             content ID does not already exist.
+        text_format (str, optional): Format of ``text``. Markdown is sent as-is when
+            the backend advertises support so conversion happens after the request
+            passes through the WAF. Older backends receive locally converted HTML.
 
     Raises:
         Exception: If the API call fails.
@@ -604,11 +608,17 @@ async def alog_metadata(
     Returns:
         dict: The response from the API.
     """
+    if text_format == "markdown" and not client_config.supports_log_metadata_markdown():
+        text = md_to_html(text, mathml=True)
+        text_format = None
+
     metadata_dict = {"content_id": content_id}
     if text is not None:
         metadata_dict["text"] = text
     if _json is not None:
         metadata_dict["json"] = _json
+    if text_format is not None:
+        metadata_dict["text_format"] = text_format
 
     request_params = {}
     if section_id:
@@ -808,11 +818,18 @@ def generate_qualitative_text(text_generation_data: Dict[str, Any]) -> Dict[str,
     return r.json()
 
 
-def _normalize_logged_text(text: str, field_name: str) -> str:
-    """Validate text content and convert Markdown to HTML when needed."""
+def _validate_logged_text(text: str, field_name: str) -> str:
+    """Validate text accepted by log_text."""
     if not isinstance(text, str) or not text:
         raise ValueError(f"`{field_name}` must be a non-empty string")
 
+    return text
+
+
+def _normalize_logged_text(text: str, field_name: str) -> str:
+    """Validate text content and convert Markdown to HTML for local rendering."""
+    text = _validate_logged_text(text, field_name)
+
     if not is_html(text):
         return md_to_html(text, mathml=True)
 
@@ -828,7 +845,7 @@ def _validate_manual_log_text_args(
     if context is not None:
         raise ValueError("`context` is only supported when `text` is omitted")
 
-    return _normalize_logged_text(text, "text")
+    return _validate_logged_text(text, "text")
 
 
 def _build_log_text_generation_request(
@@ -860,13 +877,13 @@ def _build_log_text_generation_request(
     return request_data
 
 
-def _generate_log_text(
+def _generate_log_text_source(
     content_id: str,
     prompt: Optional[str],
     context: Optional[Dict[str, Any]],
     section_id: Optional[str] = None,
 ) -> str:
-    """Generate text for log_text and normalize it to HTML."""
+    """Generate and validate source text without converting it to HTML."""
     request_data = _build_log_text_generation_request(
         content_id,
         prompt,
@@ -874,6 +891,22 @@ def _generate_log_text(
         section_id=section_id,
     )
     generated_text = generate_qualitative_text(request_data)["content"]
+    return _validate_logged_text(generated_text, "generated text")
+
+
+def _generate_log_text(
+    content_id: str,
+    prompt: Optional[str],
+    context: Optional[Dict[str, Any]],
+    section_id: Optional[str] = None,
+) -> str:
+    """Generate text and normalize it to HTML for local result rendering."""
+    generated_text = _generate_log_text_source(
+        content_id,
+        prompt,
+        context,
+        section_id=section_id,
+    )
     return _normalize_logged_text(generated_text, "generated text")
 
 
@@ -902,14 +935,21 @@ async def alog_text(
     if text is not None:
         text = _validate_manual_log_text_args(text, prompt, context)
     else:
-        text = _generate_log_text(
+        text = _generate_log_text_source(
             content_id,
             prompt,
             context,
             section_id=section_id,
         )
 
-    return await alog_metadata(content_id, text, _json, section_id=section_id)
+    text_format = None if is_html(text) else "markdown"
+    return await alog_metadata(
+        content_id,
+        text,
+        _json,
+        section_id=section_id,
+        text_format=text_format,
+    )
 
 
 def log_text(
@@ -924,9 +964,9 @@ def log_text(
 
     Args:
         content_id (str): Unique content identifier for the text.
-        text (str, optional): The text to log. Will be converted to HTML with
-            MathML support when Markdown is provided. If omitted, text is
-            generated using the qualitative text generation backend.
+        text (str, optional): The text to log. Markdown is sent to the backend for
+            HTML and math conversion. If omitted, text is generated using the
+            qualitative text generation backend.
         prompt (str, optional): Custom prompt used for AI-assisted text
             generation. Only supported when `text` is omitted.
         context (dict, optional): Context object for AI-assisted text

{validmind-2.13.9 → validmind-2.13.12}/validmind/client.py

@@ -38,7 +38,7 @@ from .template import preview_template as _preview_template
 from .test_suites import get_by_id as get_test_suite_by_id
 from .tests.run import _get_run_metadata
 from .utils import display as vm_display
-from .utils import get_dataset_info, get_model_info
+from .utils import get_dataset_info, get_model_info, is_html
 from .vm_models import TestSuite, TestSuiteRunner
 from .vm_models.dataset import DataFrameDataset, PolarsDataset, TorchDataset, VMDataset
 from .vm_models.html_progress import HTMLLabel, HTMLProgressBar
@@ -479,9 +479,12 @@ def run_text_generation(
         A TextGenerationResult containing the generated text.
     """
     start_time = time.perf_counter()
-    description = api_client._generate_log_text(
+    description_source = api_client._generate_log_text_source(
         content_id, prompt, context, section_id=section_id
     )
+    description = api_client._normalize_logged_text(
+        description_source, "generated text"
+    )
     metadata = _get_run_metadata(duration_seconds=time.perf_counter() - start_time)
     result = TextGenerationResult(
         result_id=content_id,
@@ -494,6 +497,9 @@ def run_text_generation(
         section_id=section_id,
         context=context,
         _was_description_generated=True,
+        _description_source=(
+            description_source if not is_html(description_source) else None
+        ),
     )
 
     if show:

{validmind-2.13.9 → validmind-2.13.12}/validmind/client_config.py

@@ -39,6 +39,10 @@ class ClientConfig:
         """Returns True if the client can generate LLM-based test descriptions."""
         return self.feature_flags.get("llm_test_descriptions", True)
 
+    def supports_log_metadata_markdown(self):
+        """Return whether the backend converts Markdown logged as metadata."""
+        return self.feature_flags.get("log_metadata_markdown", False)
+
 
 client_config = ClientConfig(
     model=None,

{validmind-2.13.9 → validmind-2.13.12}/validmind/experimental/agents.py

@@ -54,7 +54,8 @@ def run_task(
     else:
         raise ValueError(f"Unsupported task: {task}")
 
-    if not is_html(generated_text):
+    description_source = generated_text if not is_html(generated_text) else None
+    if description_source is not None:
         generated_text = md_to_html(generated_text, mathml=True)
 
     # Create a test result with the generated text
@@ -68,6 +69,7 @@ def run_task(
         section_id=input.get("section_id"),
         context=input.get("context"),
         _was_description_generated=True,
+        _description_source=description_source,
     )
     if show:
         result.show()

{validmind-2.13.9 → validmind-2.13.12}/validmind/tests/data_validation/LaggedCorrelationHeatmap.py

@@ -6,7 +6,6 @@ from typing import Tuple
 
 import numpy as np
 import pandas as pd
-import plotly.figure_factory as ff
 import plotly.graph_objects as go
 
 from validmind import RawData, tags, tasks
@@ -86,13 +85,15 @@ def LaggedCorrelationHeatmap(
         index=dataset.feature_columns,
     )
 
-    fig = ff.create_annotated_heatmap(
-        z=correlation_df.values,
-        x=list(correlation_df.columns),
-        y=list(correlation_df.index),
-        colorscale=COOLWARM,
-        annotation_text=correlation_df.round(2).values,
-        showscale=True,
+    fig = go.Figure(
+        go.Heatmap(
+            z=correlation_df.values,
+            x=list(correlation_df.columns),
+            y=list(correlation_df.index),
+            colorscale=COOLWARM,
+            showscale=True,
+            texttemplate="%{z:.2f}",
+        )
     )
     fig.update_layout(
         title={
@@ -103,7 +104,8 @@ def LaggedCorrelationHeatmap(
             "yanchor": "top",
         },
         font=dict(size=14),
-        xaxis_title="Lags",
+        xaxis=dict(title="Lags", side="top", dtick=1),
+        yaxis=dict(dtick=1),
     )
 
     return fig, RawData(correlation_matrix=correlation_df, dataset=dataset.input_id)

{validmind-2.13.9 → validmind-2.13.12}/validmind/tests/data_validation/TimeSeriesMissingValues.py

@@ -6,7 +6,6 @@ from typing import Any, Dict, List, Tuple
 
 import pandas as pd
 import plotly.express as px
-import plotly.figure_factory as ff
 import plotly.graph_objects as go
 
 from validmind import RawData, tags, tasks
@@ -94,16 +93,23 @@ def TimeSeriesMissingValues(
     z = missing_mask.T.astype(int).values
     x = missing_mask.index.tolist()
     y = missing_mask.columns.tolist()
-    heatmap = ff.create_annotated_heatmap(
-        z=z,
-        x=x,
-        y=y,
-        colorscale="Reds",
-        showscale=False,
+    heatmap = go.Figure(
+        go.Heatmap(
+            z=z,
+            x=x,
+            y=y,
+            colorscale="Reds",
+            showscale=False,
+            texttemplate="%{z}",
+        )
     )
 
     # Update the layout after creation
-    heatmap.update_layout(title="Missing Values Heatmap")
+    heatmap.update_layout(
+        title="Missing Values Heatmap",
+        xaxis=dict(side="top", dtick=1),
+        yaxis=dict(dtick=1),
+    )
 
     return (
         [

{validmind-2.13.9 → validmind-2.13.12}/validmind/tests/model_validation/RegressionResidualsPlot.py

@@ -5,12 +5,29 @@
 from typing import Tuple
 
 import numpy as np
-import plotly.figure_factory as ff
 import plotly.graph_objects as go
 
 from validmind import RawData, tags, tasks
+from validmind.errors import MissingDependencyError
 from validmind.vm_models import VMDataset, VMModel
 
+# The first colour in Plotly's default distplot palette, kept so the ported figure
+# matches what `ff.create_distplot` produced.
+KDE_COLOR = "rgb(31, 119, 180)"
+
+try:
+    from scipy.stats import gaussian_kde
+except ImportError as e:
+    if "scipy" in str(e):
+        raise MissingDependencyError(
+            "Missing required package `scipy` for RegressionResidualsPlot. "
+            "Please run `pip install validmind[stats]` to use statistical tests",
+            required_dependencies=["scipy"],
+            extra="stats",
+        ) from e
+
+    raise e
+
 
 @tags("model_performance", "visualization")
 @tasks("regression")
@@ -65,12 +82,30 @@ def RegressionResidualsPlot(
 
     # Residuals plot
     residuals = y_true.flatten() - y_pred.flatten()
-    fig = ff.create_distplot(
-        hist_data=[residuals],
-        group_labels=["Residuals"],
-        bin_size=[bin_size],
-        show_hist=True,
-        show_rug=False,
+    # A 500-point KDE across the data range, which is what `ff.create_distplot`
+    # computed internally with scipy before Plotly 7 removed it.
+    curve_x = np.linspace(residuals.min(), residuals.max(), 500, endpoint=False)
+    curve_y = gaussian_kde(residuals)(curve_x)
+    fig = go.Figure(
+        data=[
+            go.Histogram(
+                x=residuals,
+                histnorm="probability density",
+                autobinx=False,
+                xbins=dict(start=residuals.min(), end=residuals.max(), size=bin_size),
+                name="Residuals",
+                marker=dict(color=KDE_COLOR),
+                opacity=0.7,
+            ),
+            go.Scatter(
+                x=curve_x,
+                y=curve_y,
+                mode="lines",
+                name="Residuals",
+                showlegend=False,
+                marker=dict(color=KDE_COLOR),
+            ),
+        ]
     )
     fig.update_layout(
         title="Distribution of Residuals",

{validmind-2.13.9 → validmind-2.13.12}/validmind/tests/model_validation/sklearn/ConfusionMatrix.py

@@ -6,7 +6,6 @@
 from typing import Tuple
 
 import numpy as np
-import plotly.figure_factory as ff
 import plotly.graph_objects as go
 from sklearn.metrics import confusion_matrix
 
@@ -42,9 +41,9 @@ def ConfusionMatrix(
     The mechanism used involves taking the predicted results (`y_test_predict`) from the classification model and
     comparing them against the actual values (`y_test_true`). A confusion matrix is built using the unique labels
     extracted from `y_test_true`, employing scikit-learn's metrics. The matrix is then visually rendered with the help
-    of Plotly's `create_annotated_heatmap` function. A heatmap is created which provides a two-dimensional graphical
-    representation of the model's performance, showcasing distributions of True Positives (TP), True Negatives (TN),
-    False Positives (FP), and False Negatives (FN).
+    of a Plotly heatmap. A heatmap is created which provides a two-dimensional graphical representation of the
+    model's performance, showcasing distributions of True Positives (TP), True Negatives (TN), False Positives (FP),
+    and False Negatives (FN).
 
     ### Signs of High Risk
 
@@ -110,21 +109,32 @@ def ConfusionMatrix(
             ],
         ]
 
-    fig = ff.create_annotated_heatmap(
-        z=cm,
-        colorscale="Blues",
-        x=labels,
-        y=labels,
-        annotation_text=text,
+    if text is None:
+        text = cm.astype(str)
+
+    fig = go.Figure(
+        go.Heatmap(
+            z=cm,
+            x=labels,
+            y=labels,
+            colorscale="Blues",
+            showscale=False,
+            text=text,
+            texttemplate="%{text}",
+            # Heatmap text auto-shrinks to fit the cell; pin it to the layout font
+            # size so the cells read the same as the old annotation-based figure.
+            textfont=dict(size=12),
+            hovertemplate=(
+                "True Label:%{y}<br>Predicted Label:%{x}<br>Count:%{z}<extra></extra>"
+            ),
+        )
     )
 
-    fig["data"][0][
-        "hovertemplate"
-    ] = "True Label:%{y}<br>Predicted Label:%{x}<br>Count:%{z}<extra></extra>"
-
     fig.update_layout(
-        xaxis=dict(title="Predicted label"),
-        yaxis=dict(title="True label"),
+        # `side="top"` and `dtick=1` keep the axis placement and per-class ticks that
+        # `ff.create_annotated_heatmap` used to apply for us.
+        xaxis=dict(title="Predicted label", side="top", dtick=1),
+        yaxis=dict(title="True label", dtick=1),
         autosize=False,
         width=600,
         height=600,

{validmind-2.13.9 → validmind-2.13.12}/validmind/tests/ongoing_monitoring/TargetPredictionDistributionPlot.py

@@ -4,8 +4,8 @@
 
 from typing import Dict, List, Tuple
 
+import numpy as np
 import pandas as pd
-import plotly.figure_factory as ff
 import plotly.graph_objects as go
 
 from validmind import RawData, tags, tasks
@@ -13,7 +13,7 @@ from validmind.errors import MissingDependencyError
 from validmind.vm_models import VMDataset, VMModel
 
 try:
-    from scipy.stats import kurtosis, skew
+    from scipy.stats import gaussian_kde, kurtosis, skew
 except ImportError as e:
     if "scipy" in str(e):
         raise MissingDependencyError(
@@ -26,6 +26,17 @@ except ImportError as e:
     raise e
 
 
+def _kde_curve(values):
+    """A 500-point KDE across the data range.
+
+    This is what `ff.create_distplot(..., show_hist=False)` computed internally
+    before Plotly 7 removed the Figure Factory functions.
+    """
+    x = np.linspace(values.min(), values.max(), 500, endpoint=False)
+
+    return x, gaussian_kde(values)(x)
+
+
 @tags("visualization")
 @tasks("monitoring")
 def TargetPredictionDistributionPlot(
@@ -110,18 +121,8 @@ def TargetPredictionDistributionPlot(
     moments = moments.set_index("Statistic", drop=False)
 
     # Create KDE for both distributions
-    ref_kde = ff.create_distplot(
-        [pred_ref["Reference Prediction"].values],
-        ["Reference"],
-        show_hist=False,
-        show_rug=False,
-    )
-    monitor_kde = ff.create_distplot(
-        [pred_monitor["Monitoring Prediction"].values],
-        ["Monitoring"],
-        show_hist=False,
-        show_rug=False,
-    )
+    ref_x, ref_y = _kde_curve(pred_ref["Reference Prediction"].values)
+    monitor_x, monitor_y = _kde_curve(pred_monitor["Monitoring Prediction"].values)
 
     # Create new figure
     fig = go.Figure()
@@ -129,8 +130,8 @@ def TargetPredictionDistributionPlot(
     # Add reference distribution
     fig.add_trace(
         go.Scatter(
-            x=ref_kde.data[0].x,
-            y=ref_kde.data[0].y,
+            x=ref_x,
+            y=ref_y,
             fill="tozeroy",
             name="Reference Prediction",
             line=dict(color="blue", width=2),
@@ -141,8 +142,8 @@ def TargetPredictionDistributionPlot(
     # Add monitoring distribution
     fig.add_trace(
         go.Scatter(
-            x=monitor_kde.data[0].x,
-            y=monitor_kde.data[0].y,
+            x=monitor_x,
+            y=monitor_y,
             fill="tozeroy",
             name="Monitor Prediction",
             line=dict(color="red", width=2),

{validmind-2.13.9 → validmind-2.13.12}/validmind/tests/output.py

@@ -184,7 +184,10 @@ class StringOutputHandler(OutputHandler):
 
     def process(self, item: Any, result: TestResult) -> None:
         if not is_html(item):
+            result._description_source = item
             item = md_to_html(item, mathml=True)
+        else:
+            result._description_source = None
 
         result.description = item
 

{validmind-2.13.9 → validmind-2.13.12}/validmind/tests/run.py

@@ -16,7 +16,7 @@ from validmind.ai.test_descriptions import get_result_description
 from validmind.errors import MissingRequiredTestInputError
 from validmind.input_registry import input_registry
 from validmind.logging import get_logger
-from validmind.utils import test_id_to_name
+from validmind.utils import is_html, test_id_to_name
 from validmind.vm_models.input import VMInput
 from validmind.vm_models.result import TestResult
 
@@ -451,7 +451,7 @@ def run_test(  # noqa: C901
         result = post_process_fn(result)
 
     if not result.description:
-        result.description = get_result_description(
+        description = get_result_description(
             test_id=test_id,
             test_description=result.doc,
             tables=result.tables,
@@ -463,6 +463,9 @@ def run_test(  # noqa: C901
             additional_context=additional_context,
             params=result.params,
         )
+        result.description = description
+        if isinstance(description, str) and result.doc and not is_html(result.doc):
+            result._description_source = result.doc
 
     if show:
         result.show()

{validmind-2.13.9 → validmind-2.13.12}/validmind/unit_metrics/regression/GiniCoefficient.py

@@ -7,6 +7,10 @@ import numpy as np
 from validmind import tags, tasks
 from validmind.vm_models import VMDataset, VMModel
 
+# NumPy 2 removed `np.trapz` in favour of `np.trapezoid`; the library supports both
+# (numpy<2 below Python 3.14, numpy>=2.3 from 3.14 on).
+_trapezoid = getattr(np, "trapezoid", None) or np.trapz
+
 
 @tags("regression")
 @tasks("regression")
@@ -29,7 +33,7 @@ def GiniCoefficient(dataset: VMDataset, model: VMModel) -> float:
     cumsum_pred_norm = cumsum_pred / np.max(cumsum_pred)
 
     # Compute area under the Lorenz curve
-    area_lorenz = np.trapz(cumsum_pred_norm, x=cumsum_true_norm)
+    area_lorenz = _trapezoid(cumsum_pred_norm, x=cumsum_true_norm)
 
     # Compute Gini coefficient
     return 1 - 2 * area_lorenz