MetricsApp

class rectools.visuals.metrics_app.MetricsApp(data: DataFrame, metric_names: List[str], meta_names: List[str], show_legend: bool = True, auto_display: bool = True, scatter_kwargs: Optional[Dict[str, Any]] = None)

Bases: object

Jupyter widgets app for metric visualization and comparison. Do not create instances of this class directly. Use MetricsApp.construct method instead.

Inherited-members

Parameters

• data (DataFrame) –

• metric_names (List[str]) –

• meta_names (List[str]) –

• show_legend (bool) –

• auto_display (bool) –

• scatter_kwargs (Optional[Dict[str, Any]]) –

Methods

construct(models_metrics[, models_metadata, ...])	Construct interactive widget for metric-to-metric trade-off analysis.
display()	Display MetricsApp widget

Attributes

fold_ids	Sorted list of fold identifiers from the models_metrics.
model_names	Sorted list of model names from models_metrics.

classmethod construct(models_metrics: DataFrame, models_metadata: Optional[DataFrame] = None, show_legend: bool = True, auto_display: bool = True, scatter_kwargs: Optional[Dict[str, Any]] = None) → MetricsApp

Construct interactive widget for metric-to-metric trade-off analysis.

Parameters

• models_metrics (pd.DataFrame) –

  A pandas DataFrame containing metrics for visualization. Required columns:

  • Columns.Models - model names

  • Columns.Split - fold number

  • Any other numeric columns which represent metric values

• models_metadata (tp.Optional[pd.DataFrame], optional, default None) –

  An optional pandas DataFrame containing any models metadata (hyperparameters, training info, etc.). Used for alternative ways of coloring scatterplot points. Required columns:

  • Columns.Model - model names

  • Any other columns with additional information

• show_legend (bool, default True) – Specifies whether to display the chart legend.

• auto_display (bool, default True) – Automatically displays the widgets immediately after initialization.

• scatter_kwargs (tp.Optional[tp.Dict[str, tp.Any]], optional, default None) – Additional arguments for plotly.express.scatter

Returns

An instance of MetricsApp, providing interactive Jupyter widget for metric visualization.

Return type

MetricsApp

Examples

Create interactive widget

>>> metrics_df = pd.DataFrame(
...    {
...        Columns.Model: ["Model1", "Model2", "Model1", "Model2", "Model1", "Model2"],
...        Columns.Split: [0, 0, 1, 1, 2, 2],
...        "prec@10": [0.031, 0.025, 0.027, 0.21, 0.031, 0.033],
...        "recall@10": [0.041, 0.045, 0.055, 0.08, 0.036, 0.021],
...        "novelty@10": [2.6, 11.3, 4.3, 9.8, 3.3, 11.2],
...    })
>>> # Optional metainfo about models
>>> metadata_df = pd.DataFrame(
...    {
...        Columns.Model: ["Model1", "Model2"],
...        "factors": [64, 32],
...        "regularization": [0.05, 0.05],
...        "alpha": [2.0, 0.5],
...    })
>>> app = MetricsApp.construct(
...    models_metrics=metrics_df,
...    models_metadata=metadata_df,
...    show_legend=True,
...    auto_display=False,
...    scatter_kwargs={"width": 800, "height": 600})

Get plotly chart from the current widget state

>>> fig = app.fig
>>> fig = fig.update_layout(title="Metrics comparison")

display() → None

Display MetricsApp widget

Return type

None

property fold_ids: Optional[List[int]]

Sorted list of fold identifiers from the models_metrics.

property model_names: List[str]

Sorted list of model names from models_metrics.