predictions.py

import sys, os
import numpy as np

sys.path.append(os.path.dirname(os.path.dirname(os.path.realpath(__file__))))
import regressionAnalysis as reg

#TODO find way to grab last item in yFit to get predictions offset
def predict(predictMatrix, model, modelBuilder, dataset, knownYCumulative):
	print(f"\nPredicting {modelBuilder.__name__} for {len(predictMatrix[0])} iterations")
	dataset = np.array(dataset)
	#print(f"\n\n{dataset}\n\n")
	#print(f"\n\n{predictMatrix}\n\n")
	# Put last item in covariates at front of predictMatrix
	predictMatrixMod = []
	for cov in range(len(predictMatrix)):
		predictMatrixMod.append([dataset[cov + 1][-1]])
		for x in predictMatrix[cov]:
			predictMatrixMod[cov].append(x)

	print(f"Prediction Matrix: {predictMatrixMod}")
	predictMatrixMod = np.array(predictMatrixMod)
	length = len(predictMatrixMod[0])
	# handling for when an interaction is zero
	culled = []
	if modelBuilder == reg.interactionCoefficients:
		knownCoeff_eqs = [modelBuilder(dataset[1:, i]) for i in range(len(dataset[1:]))]
		culled = reg.cullZeros(knownCoeff_eqs)

	coeff_eqs = [modelBuilder(predictMatrixMod[:, i]) for i in range(length)]
	if culled:
		for interaction in culled[::-1]:
			for i, eq in enumerate(coeff_eqs):
				coeff_eqs[i] = np.delete(eq, interaction)

	# Construct Y Cumulative from betas
	# Do not use model[:-1] here. reg.main returns it like that already
	YFitList = [np.dot(model, coeff_eqs[i]) for i in range(len(coeff_eqs))]
	YPredictions = reg.yFitCumulative(YFitList, length)

	# Fix offset by comparing to last known yCumulative in original dataset
	# I have no idea why, but yPredictions is off by a constant offset. This fixes it
	offset = YPredictions[0] - knownYCumulative[-1]
	YPredictionsCorrected = [x - offset for x in YPredictions]
	return YPredictionsCorrected[1:]

# ex usage: predict(userMatrix, MLRRules, mlrCoefficients, dataset)