Plot factory

Supercharge your flow with ready made interactive plots.

PoniardPlotFactory is a series of plots meant to enhance the PoniardBaseEstimator.get_results experience. Like the rest of Poniard, wherever possible it will try to use cross validation (for example, PoniardPlotFactory.confusion_matrix uses predictions from scikit-learn’s cross_val_predict).

---

source

PoniardPlotFactory

 PoniardPlotFactory (template:str='plotly_white',
                     discrete_colors:List[str]=['rgb(127, 60, 141)',
                     'rgb(17, 165, 121)', 'rgb(57, 105, 172)', 'rgb(242,
                     183, 1)', 'rgb(231, 63, 116)', 'rgb(128, 186, 90)',
                     'rgb(230, 131, 16)', 'rgb(0, 134, 149)', 'rgb(207,
                     28, 144)', 'rgb(249, 123, 114)', 'rgb(165, 170,
                     153)'], font_family:str='Helvetica',
                     font_color:str='#8C8C8C')

Helper class that handles plotting for Poniard Estimators.

It has access to the Poniard estimator instance through the _poniard attribute.

	Type	Default	Details
template	str	plotly_white	Plotly template. Default “plotly_white”.
discrete_colors	typing.List[str]	[‘rgb(127, 60, 141)’, ‘rgb(17, 165, 121)’, ‘rgb(57, 105, 172)’, ‘rgb(242, 183, 1)’, ‘rgb(231, 63, 116)’, ‘rgb(128, 186, 90)’, ‘rgb(230, 131, 16)’, ‘rgb(0, 134, 149)’, ‘rgb(207, 28, 144)’, ‘rgb(249, 123, 114)’, ‘rgb(165, 170, 153)’]	A list of colors, default Bold. See the Plotly reference.
font_family	str	Helvetica	See the Plotly reference
font_color	str	#8C8C8C	See the Plotly reference

General plots

---

source

PoniardPlotFactory.metrics

 PoniardPlotFactory.metrics (kind:str='strip', facet:str='col',
                             metrics:Union[str,Sequence[str]]=None,
                             only_test:bool=True, exclude_dummy:bool=True,
                             show_means:bool=True, **kwargs)

Plot metrics obtained by running PoniardBaseEstimator.fit.

	Type	Default	Details
kind	str	strip	Either “strip” or “bar”. Default “strip”.
facet	str	col	Either “col” or “row”. Default “col”.
metrics	typing.Union[str, typing.Sequence[str]]	None	String or list of strings. This must follow the names passed to the Poniard constructor. For example, if during init a dict of metrics was passed, its keys can be passed here. Default None, which plots every estimator metric available.
only_test	bool	True	Whether to plot only test scores. Default True.
exclude_dummy	bool	True	Whether to exclude dummy estimators. Default True.
show_means	bool	True	Whether to plot means along with fold scores. Default True.
kwargs
Returns	Figure		Plotly strip or bar plot.

from sklearn.datasets import load_iris
from poniard import PoniardClassifier

X, y = load_iris(return_X_y=True, as_frame=True)
pnd = PoniardClassifier().setup(X, y, show_info=False).fit()

pnd.plot.metrics(metrics="f1_macro")

pnd.plot.metrics(metrics=["roc_auc", "accuracy"], facet="col", kind="bar")

---

source

PoniardPlotFactory.overfitness

 PoniardPlotFactory.overfitness (metric:Optional[str]=None,
                                 exclude_dummy:bool=True)

Plot the ratio of test scores to train scores for every estimator.

	Type	Default	Details
metric	typing.Optional[str]	None	String representing a metric. This must follow the names passed to the Poniard constructor. For example, if during init a dict of metrics was passed, one of its keys can be passed here. Default None, which plots the first metric.
exclude_dummy	bool	True	Whether to exclude dummy estimators. Default True.
Returns	Figure		Plotly strip plot.

Classification plots

Poniard offers additional plots based on cross validated predictions:

---

source

PoniardPlotFactory.roc_curve

 PoniardPlotFactory.roc_curve
                               (estimator_names:Optional[Sequence[str]]=No
                               ne, response_method:str='auto', **kwargs)

Plot ROC curve with cross validated predictions for multiple estimators.

	Type	Default	Details
estimator_names	typing.Optional[typing.Sequence[str]]	None	Estimators to include. If None, all estimators are used.
response_method	str	auto	Either “auto”, “predict_proba” or “decision_function”. “auto” will try to use predict_proba if all estimators have it, otherwise it will try decision_function If there is no common response_method, it will raise an error.
kwargs			Passed to sklearn.metrics.roc_curve().
Returns	Figure		Plotly line plot.

For now, PoniardPlotFactory.roc_curve only works with binary classification tasks (you can track this issue here, so we’ll need a different dataset, like the well known German credit dataset.

from sklearn.datasets import fetch_openml
from sklearn.preprocessing import LabelEncoder
from poniard import PoniardClassifier

X, y = fetch_openml(data_id=31, return_X_y=True, as_frame=True)
categorical_cols = X.select_dtypes(include="category").columns
X[categorical_cols] = X[categorical_cols].astype(str)
y = LabelEncoder().fit_transform(y)
pnd = PoniardClassifier().setup(X, y, show_info=False).fit()

estimators = ["LogisticRegression", "XGBClassifier", "SVC"]
pnd.plot.roc_curve(estimator_names=estimators)

---

source

PoniardPlotFactory.confusion_matrix

 PoniardPlotFactory.confusion_matrix (estimator_name:str, **kwargs)

Plot confusion matrix with cross validated predictions for a single estimator.

	Type	Details
estimator_name	str	Estimator to include.
kwargs		Passed to sklearn.metrics.confusion_matrix().
Returns	Figure	Plotly image plot.

Like roc_curve, confusion_matrix accepts kwargs that are passed on to the appropiate sklearn functions.

pnd.plot.confusion_matrix(estimator_name="LogisticRegression", normalize="all")

Regression plots

---

source

PoniardPlotFactory.residuals

 PoniardPlotFactory.residuals (estimator_names:List[str])

Plot regression residuals vs predictions for a list of estimators.

	Type	Details
estimator_names	typing.List[str]	Estimators to include.
Returns	Figure	Residuals plot.

---

source

PoniardPlotFactory.residuals_histogram

 PoniardPlotFactory.residuals_histogram (estimator_names:List[str])

Plot a histogram of regression residuals for a list of estimators.

	Type	Details
estimator_names	typing.List[str]	Estimators to include.
Returns	Figure	Residuals histogram plot.

from sklearn.datasets import fetch_california_housing
from sklearn.linear_model import LinearRegression
from xgboost import XGBRegressor

from poniard import PoniardRegressor

X, y = fetch_california_housing(return_X_y=True, as_frame=True)
pnd = PoniardRegressor(estimators=[LinearRegression(), XGBRegressor()]).setup(
    X, y, show_info=False
)
pnd.fit()

PoniardRegressor(estimators=[LinearRegression(), XGBRegressor(base_score=None, booster=None, colsample_bylevel=None,
             colsample_bynode=None, colsample_bytree=None,
             enable_categorical=False, gamma=None, gpu_id=None,
             importance_type=None, interaction_constraints=None,
             learning_rate=None, max_delta_step=None, max_depth=None,
             min_child_weight=None, missing=nan, monotone_constraints=None,
             n_estimators=100, n_jobs=None, num_parallel_tree=None,
             predictor=None, random_state=0, reg_alpha=None, reg_lambda=None,
             scale_pos_weight=None, subsample=None, tree_method=None,
             validate_parameters=None, verbosity=0)])

pnd.plot.residuals(estimator_names=["LinearRegression", "XGBRegressor"])

pnd.plot.residuals_histogram(estimator_names=["LinearRegression", "XGBRegressor"])

Feature plots

These plots help in understanding how features interact with models.

---

source

PoniardPlotFactory.permutation_importance

 PoniardPlotFactory.permutation_importance (estimator_name:str,
                                            n_repeats:int=10,
                                            kind:str='bar', **kwargs)

Plot permutation importances for an estimator.

This shuffles features randomly one at a time and measures the change in the estimator’s performance. If the feature is important for the model, the estimator’s performance should decrease (represented by positive values in the plot). See the scikit-learn guide.

	Type	Default	Details
estimator_name	str		Estimator to include.
n_repeats	int	10	How many times to repeat random permutations of a single feature. Default 10.
kind	str	bar	Either “bar” or “strip”. Default “bar”. “strip” plots each permutation repetition as well as the mean. Bar plots only the mean.
kwargs			Passed to sklearn.inspection.permutation_importance().
Returns	Figure		Plotly bar or strip plot.

pnd.plot.permutation_importance(estimator_name="XGBRegressor")

---

source

PoniardPlotFactory.partial_dependence

 PoniardPlotFactory.partial_dependence (estimator_name:str,
                                        feature:Union[str,int], **kwargs)

Plot partial dependence for a single feature of a single estimator.

In essence, visualize how the target changes within the feature’s range.

Only plots average partial dependence for all samples and not individual samples (ICE).

	Type	Details
estimator_name	str	Estimator to include.
feature	typing.Union[str, int]	Feature for which to plot partial dependence. Can be a pandas column name or index.
kwargs		Passed to sklearn.inspection.partial_dependence().
Returns	Figure	Plotly line plot.

pnd.plot.partial_dependence("XGBRegressor", "AveOccup")