Predictive Systems Regression

The node returns values from CDF that correspond to provided percentile values, the probabilities of value being within quantiles, less the fixed values and less than test value depending the provided settings.

Options

Target column (calibration table)

A column with target column that is being predicted.

Prediction column

A column with predicted values.

Use normalization

In order to increase the informativeness, and to potentially minimize prediction regions, it is possible to obtain individual bounds for each sample, which is achieved using a normalized nonconformity function. In order to use normalization Prediction Variance or estimate of difficulty - Sigma (needs to be calculated separately) should be provided.

Difficulty column

The column that contains the values defining the Sigma. By default it is advisable to use Prediction Variance values.

Beta

A sensitivity parameter determining the relative importance of the normalization term.

Keep All Columns

If checked all columns from input table will be also included into output table along with calibration columns.

Keep ID column

If checked selected ID column from input table will be included into output table.

Probability distribution column

The column that contains the CDF values for the samples to be predicted.

Target value

A fixed value to compare prediction with. As the output there will be a calculated probability that the predicted values are lower than this fixed value.

Target column

Column that contains values to compare with (test set). As the output there will be a calculated probability that the predicted values are lower than corresponding values from the selected column.

Lower percentiles (%)

The desired percentile cutoff that controls the lower border of the interval from which the predictions could be sampled. The interval is used for estimating the range of possible errors. As the output there will be a calculated probability that the predicted values are higher than the border value corresponding to the percentile. Multiple lower percentiles are allowed, setting the lower interval is optional.

Upper percentiles (%)

The desired percentile cutoff that controls the upper border of the interval from which the predictions could be sampled. The interval is used for estimating the range of possible errors. As the output there will be a calculated probability that the predicted values are lower than the border value corresponding to the percentile. Multiple higher intervals are allowed, setting the upper interval is optional.

Views

This node has no views

Workflows

• Conformal predictive systems (simple)KNIME Hub
• COPA 2023 - Conformal predictive systems - simple use casesKNIME Hub
• COPA 2023 - Mondrian conformal predictive systems experimentKNIME Hub

Links

• No links available

Developers