I found the following solution from a early solved issue.
"""Prediction_local is intercept + weights * features.
Sorry, this is poorly documented, but the weights are actually relative to the scaled data, so your feature_value column should be scaled. You can get the scaled values by running:
(data_toexplain - explainer.scaler.mean) / explainer.scaler.scale_
Or you can believe that I did it right when computing prediction_local : )"""
But I tried this formula to compare the prediction_local with the manual calculated prediction using the feature weights. But unfortunately they are different. (In my case "discretize_continuous" to false.)
Has the Prediction_local = sum (feature_weight*scaled_feature_value) + intercept formula been changed now?
charlie9526
commented
4 years ago
I found the following solution from a early solved issue. """Prediction_local is intercept + weights * features. Sorry, this is poorly documented, but the weights are actually relative to the scaled data, so your feature_value column should be scaled. You can get the scaled values by running: (data_toexplain - explainer.scaler.mean) / explainer.scaler.scale_ Or you can believe that I did it right when computing prediction_local : )"""
But I tried this formula to compare the prediction_local with the manual calculated prediction using the feature weights. But unfortunately they are different. (In my case "discretize_continuous" to false.)
Has the Prediction_local = sum (feature_weight*scaled_feature_value) + intercept formula been changed now?