Please enable JavaScript.
Coggle requires JavaScript to display documents.
Chapter 19- Interpret Model (reason codes (The focus of this chapter has…
Chapter 19- Interpret Model
feature impacts on target
There are four kinds of relationships that are commonly useful for exploring why a model predicts certain outcomes. One of the advantages of supervised machine learning is that all relationships are measured in terms of their relationship with the target.
overall impact of a feature without consideration of the impact of other features
treats each feat. as a standalone effect on the target
importance score = really useful, it allows data scientists to focus attention on features most likely to yield additional predictive value if misinterpreted by AutoML
these scores are not fully reliable indicators of value of feat.
overall impact of feature adjusted for the impact of other features
usain bolt example
directional impact of feature
whether the presence of a value helps the model by assisting it in predicting readmissions or no readmissions
partial impact of feature
the power of language
at this point access to SME is important during eval of the text model(S)
info displayed in word cloud
reps words that have highest coefficients
hotspots
insights then hotspots
uses rule fit classifier (immediate red flag)
hotspot screen shows most relevant...up to four... combos of feat. and their effect on the target
largest and most overlapping hotspots are in the middle, deeper the tone of blue or red, more of an impact that particular combo of feat. has on target
not recommended to show during presentations because its very detailed and complex
reason codes
The focus of this chapter has thus far been on the features and feature values that
drive positive or negative changes in the target; however, understanding of the
data at the more granular individual patient level remains limited.
strong (+++/---) medium (++/--) weak (+/-)
powerful feature, can supplement business decisions
computing these is slower than computing predictions...these engage in additional evals of why a prediction was set as the given probability for that case