blackboxopt.visualizations.visualizer

create_hover_information(sections)

Create a hovertemplate which is used to render hover hints in plotly charts.

The data for the chart hovertext has to be provided as custom_data attribute to the chart and can be e.g. a list of column names.

One oddness is, that in the template the columns can't be referenced by name, but only by index. That's why it is important to have the same ordering in the template as in the custom_data and the reason why this is done together in one function.

Parameters:

Name	Type	Description	Default
sections	dict	Sections to render. The kyeys will show up as the section titles, values are expected to be a list of column names to be rendered under the section. E.g.: { "info": ["Objective #1", "Objective #2", "fidelity"] }	required

Returns:

Type	Description
Tuple[str, List]	(plotly hover template, data column names)

Source code in blackboxopt/visualizations/visualizer.py

def create_hover_information(sections: dict) -> Tuple[str, List]:
    """
    Create a [hovertemplate](https://plotly.com/python/reference/pie/#pie-hovertemplate)
    which is used to render hover hints in plotly charts.

    The data for the chart hovertext has to be provided as `custom_data` attribute to
    the chart and can be e.g. a list of column names.

    One oddness is, that in the template the columns can't be referenced by name, but
    only by index. That's why it is important to have the same ordering in the template
    as in the `custom_data` and the reason why this is done together in one function.

    Args:
        sections: Sections to render. The kyeys will show up as the section titles,
            values are expected to be a list of column names to be rendered under
            the section. E.g.: { "info": ["Objective #1", "Objective #2", "fidelity"] }

    Returns:
        (plotly hover template, data column names)
    """
    template = ""
    idx = 0
    for section, columns in sections.items():
        template += f"<br><b>{section.replace('_', ' ').title()}</b><br>"
        for column in columns:
            template += f"{column}: %{{customdata[{idx}]}}<br>"
            idx += 1
    template += "<extra></extra>"

    data_columns: list = sum(sections.values(), [])

    return template, data_columns

evaluations_to_df(evaluations)

Convert evaluations into multi index dataframe.

The evaluations will be casted to dictionaries which will be normalized. The keys of the dicts will be used as secondary column index. Evaluations with one or more missing objective-value will be dropped.

Examples:

Evaluation(objectives={'loss_1': 1.0, 'loss_2': -0.0}, stacktrace=None, ...)

Will be transformed into:

| objectives | stacktrace | ... | <- "group" index | loss_1 | loss_2 | stacktrace | ... | <- "field" index | ------ | ------ | ---------- | --- | | 1.0 | -0.0 | None | ... |

Source code in blackboxopt/visualizations/visualizer.py

def evaluations_to_df(evaluations: List[Evaluation]) -> pd.DataFrame:
    """Convert evaluations into multi index dataframe.

    The evaluations will be casted to dictionaries which will be normalized.
    The keys of the dicts will be used as secondary column index. Evaluations
    with one or more missing objective-value will be dropped.

    Example:

    ```
    Evaluation(objectives={'loss_1': 1.0, 'loss_2': -0.0}, stacktrace=None, ...)
    ```

    Will be transformed into:

    |    objectives   | stacktrace | ... |  <- "group" index
    | loss_1 | loss_2 | stacktrace | ... |  <- "field" index
    | ------ | ------ | ---------- | --- |
    |    1.0 |   -0.0 | None       | ... |
    """
    if not evaluations or len(evaluations) == 0:
        raise NoSuccessfulEvaluationsError

    # Filter out e.g. EvaluationSpecifications which might be passed into
    evaluations = [e for e in evaluations if isinstance(e, Evaluation)]

    # Transform to dicts, filter out evaluations with missing objectives
    evaluation_dicts = [e.__dict__ for e in evaluations if not e.any_objective_none]

    if len(evaluation_dicts) == 0:
        raise NoSuccessfulEvaluationsError

    df = pd.DataFrame(evaluation_dicts)

    # Flatten json/dict columns into single multi-index dataframe
    dfs_expanded = []
    for column in df.columns:
        # Normalize json columns keep original column for non-json columns
        try:
            df_temp = pd.json_normalize(df[column], errors="ignore", max_level=0)
        except AttributeError:
            df_temp = df[[column]]

        # Use keys of dicts as second level of column index
        df_temp.columns = pd.MultiIndex.from_product(
            [[column], df_temp.columns], names=["group", "field"]
        )
        # Drop empty columns
        df_temp = df_temp.dropna(axis=1, how="all")

        dfs_expanded.append(df_temp)

    df = pd.concat(dfs_expanded, join="outer", axis=1)

    # Parse datetime columns
    date_columns = [c for c in df.columns if "unixtime" in str(c)]
    df[date_columns] = df[date_columns].apply(pd.to_datetime, unit="s")

    # Calculate duration in seconds
    df["duration", "duration"] = (
        df["finished_unixtime", "finished_unixtime"]
        - df["created_unixtime", "created_unixtime"]
    )

    return df