mlpce

Machine Learning Prediction Confidence Estimation

Let's say you have a cool XGBOOST model that you've built and now you're wanting to make predictions with it on new data points - how well does your training data cover that model space? In classic statistical analysis, especially DOEs, there are many characteristics about the data used to cover a space that can be considered (e.g. A-, D-, G-, I-optimality). I-optimality is the average prediction variance in the design space, that is, a measure of how precisely a model built on that data should be able to make new predictions.

mlpce is a Python package which provides an expression of confidence in any given prediction by using an approximating linear function to calculate the standard error of prediction for the new point and comparing it to the same value for the training data. The approximating linear function can either be specified as a string or the module will simply pick a high-order polynomial model based on the available degrees of freedom in the training data.

Usage

Consider a dataset picked to be I-Optimal for evaluating a full third-order response surface model. There are 54 rows and 6 columns. This pandas data frame can then be passed into the Confidence class where an approximating linear model will be created and the necessary matrices will be calculated. Now we can pass in a few new rows to be evaluated.

import pandas as pd
from mlpce import Confidence

pd_x = pd.DataFrame(data=[[-1, -0.5, 0.5, -1, 1, 1], [1, -1, 1, -1, -1, -1], [-0.5, 0.5, 1, -0.5, 0, 1],
                          [0.5, 1, 1, 0.5, -1, -1], [-0.5, 0.5, -0.5, 1, -1, 0.5], [-0.5, 0.5, -1, -0.5, 0.5, 1],
                          [1, 1, -1, -1, -1, 0.5], [1, -1, -1, -0.5, 1, 0.5], [1, 0.5, -1, 1, 0.5, 0],
                          [0, -0.5, 0.5, -0.5, -0.5, 0.5], [1, 1, 1, 1, 1, -0.5], [0.5, 1, -0.5, 0.5, -0.5, 1],
                          [0.5, -0.5, -0.5, -0.5, 0.5, -0.5], [1, -1, 1, -1, 0.5, 1], [-1, 1, 0, 1, 1, 1],
                          [1, 1, 0.5, -1, 1, 1], [-0.5, -0.5, -1, -1, 0.5, -1], [1, -1, -1, 0.5, 1, -1],
                          [0.5, -1, -1, -1, -0.5, -0.5], [-1, -1, 0, -0.5, -1, -1], [1, -0.5, 1, 0.5, 1, 0],
                          [0.5, -1, 0.5, 1, 0, -0.5], [1, 0.5, 0.5, -0.5, -0.5, -0.5], [1, -1, 1, 0.5, -1, 1],
                          [0.5, 0.5, -0.5, -1, 1, -1], [0.5, 0.5, 0.5, 0.5, 0.5, 0.5], [0.5, -0.5, 0, 1, 1, 1],
                          [-0.5, -0.5, 1, 0.5, -1, -0.5], [-1, 1, 0, -0.5, 1, 0], [1, 1, -0.5, -1, -0.5, -1],
                          [0.5, 0.5, -1, 1, -1, -0.5], [0.5, 1, 1, -1, -1, 0.5], [1, -1, -1, 1, -1, 0.5],
                          [-0.5, -1, -0.5, 0.5, 1, 0], [1, -0.5, -0.5, -1, -1, 1], [-1, -0.5, -1, 1, -0.5, -1],
                          [-1, 1, -1, 1, 0.5, -1], [-0.5, -1, -1, -0.5, -1, 1], [-1, 0, -0.5, -1, -0.5, 0.5],
                          [1, -1, 0.5, -1, 1, -1], [-1, 0.5, -1, -0.5, -1, -1], [1, 1, 1, 1, -1, 1],
                          [1, -1, -0.5, 0.5, -1, -1], [-1, 0.5, 1, 1, -1, -1], [-1, -1, 1, -0.5, 1, -0.5],
                          [-1, -0.5, -1, 0.5, 0, 1], [-1, -1, 1, -1, -1, 1], [-1, 0, 0.5, 1, 1, -1], 
                          [0.5, 1, 1, -1, 0.5, -1], [-0.5, 0.5, 1, -1, -1, -1], [-1, 0, 1, 1, -1, 1], 
                          [-1, 1, 0.5, -0.5, -1, 1], [-0.5, 1, 0.5, 0.5, 0, -0.5], [-1, -1, 1, 1, 0.5, 0.5]],
                    columns=['a', 'b', 'c', 'd', 'e', 'f'])
pd_x_k = pd.DataFrame(data=[[0, 0, 0, 0, 0, 0], [2, 2, 2, 2, 2, 2]],
                      columns=['a', 'b', 'c', 'd', 'e', 'f'])

emm = Confidence(known=pd_x)
pred_variance, confidence = emm.assess_x(pd_x_k)

The results are dictionaries with keys matching any responses provided as well as a 'Full' key which evaluates the row in the setting of all x values (without regard for missing values in responses). The first element is the calculated, unscaled prediction variance. The second element is a string of 'High', 'Mid' or 'Low' indicating how confident you can feel in the model's ability to make predictions in this space.

• High - the prediction variance is less than the 90th percentile of training data's prediction variances
• Mid - the prediction variance is no greater than the maximum prediction variance of the training data
• Low - the prediction variance is greater than the maximum prediction variance of the training data