plot_predictions
Sequentially vary the value of the specified features to put them into all buckets and calculate predictions for the input objects accordingly.
Alert
- Only models trained on datasets that do not contain categorical features are supported.
- Multiclassification modes are not supported.
Parameters
data
Description
The data to plot predictions for.
For example, use a two-document slice of the original dataset (refer to the example below).
Possible types
- numpy.ndarray
- pandas.DataFrame
- pandas.SparseDataFrame
- scipy.sparse.spmatrix (all subclasses except dia_matrix)
- catboost.Pool
Default value
Required parameter
features_to_change
Description
The list of numerical features to vary the prediction value for.
For example, chose the required features by selecting top N most important features that impact the prediction results for a pair of objects according to PredictionDiff (refer to the example below).
Possible types
- list of int
- string
- combination of list of int & string
Default value
Required parameter
plot
Description
Plot a Jupyter Notebook chart based on the calculated predictions.
Possible types
bool
Default value
True
plot_file
Description
The name of the output HTML-file to save the chart to.
Possible types
string
Default value
None (the files is not saved)
Return value
Dict with two fields:
A list of dictionaries with predictions for all objects in the data float feature index -> [prediction for the object with corresponding feature values in the bucket : for all buckets used in the model]
Examples
import numpy as np
from catboost import Pool, CatBoost
train_data = np.random.randint(0, 100, size=(100, 10))
train_label = np.random.randint(0, 1000, size=(100))
train_pool = Pool(train_data, train_label)
train_pool_slice = train_pool.slice([2, 3])
model = CatBoost()
model.fit(train_pool)
prediction_diff = model.get_feature_importance(train_pool_slice,
type='PredictionDiff',
prettified=True)
model.plot_predictions(data=train_pool_slice,
features_to_change=prediction_diff["Feature Id"][:2],
plot=True,
plot_file="plot_predictions_file.html")
An example of the first plotted chart:
