Please enable JavaScript.
Coggle requires JavaScript to display documents.
Chapter 19. Interpret Model (19.1 Feature Impacts on Targets (The overall…
Chapter 19. Interpret Model
19.1 Feature Impacts on Targets
The overall impact of a feature without consideration of the impact of other
features.
The overall impact of a feature adjusted for the impact of other features.
The directional impact of a feature.
The partial impact of a feature
19.2 The Overall Impact of Features on the Target without considerations of other Features
A full green bar indicates that the feature explains at least 30%
of the target value
the optimization
measure score for the individual feature model becomes visible
importance score is exceedingly useful because it allows a data scientist to
focus attention on the features most likely to yield additional predictive value if
misinterpreted by the AutoML
importance scores should not be
relied on for feature selection and model interpretation
19.3 The Overall Impact of a Feature Adjusted for the Impact of other Features
select Feature Impact, followed by Compute Feature Impact
will initiate a calculation of the value of each feature in
the context of this model
DataRobot then examines the model’s
performance relative to the model that retained all the features
Usain Bolt Ex.
This figure shows that the number of inpatient visits during the last year
(number_inpatient) is the most predictive feature overall. Discharge
(disposition_id) is the second most predictive feature, and num_medications the
fifth most predictive feature
19.4 The Directional Impact of Features on Target
The third type of relationship is what has been termed in this book the directional
impact of the feature, or whether the presence of a value helps the model by
assisting it in predicting readmissions or non-readmissions
Go to insights
shown is the result of a logistic
regression analysis
Autopilot stage 3, Run with new Sample Size
examine its LogLoss score
does provide what are commonly known as coefficients
19.5 The Partial Impact of Features on Target
Now navigate to the Model X-Ray screen for the ENET Blender model
The Model X-Ray constructs a list of features ranked by
their influence on the target
left Y-axis contains the
frequency of cases in the validation set. The X-axis contains the values of the most
predictive feature, discharge_disposition_id
DataRobot averages its prediction probabilities of these 73 patients and
places these probabilities as a blue cross, labelled “Predicted.”
Try out the remaining features to gain
experience with the Model X-Ray view
19.6 The Power of Language
head to the Insights screen again and select Text Mining
The two most important
terms related to patients avoiding readmission were diag_3_desc and sarcoidosis,
an inflammatory disease
Diseases of tricuspid valve
● Mitral valve disorders
● Congenital pulmonary valve anomaly, unspecified
● Mitral valve stenosis and aortic valve stenosis
The key then, to interpreting the word cloud,
is to simply analyze large terms with vivid colors
Filter Stop
Words
19.7 Hotspots
The final portion of chapter 19 will outline the Hotspots view in DataRobot
run the
same model again with access to the full data
Click Run with New Sample Size
The Hotspot screen shows the most relevant (up to four) combinations of features
and their effect on the target
The Hotspot screen shows the most relevant (up to four) combinations of features
and their effect on the target or cold
it is not recommended to show
this particular screen during presentations due to its exceedingly high level of detail
and complexity
19.8 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
feature values need to be examined along with
an analysis of how these values determined that patient’s probability of
readmission
The left and right thresholds allow for the specification of the probability cutoffs to
be used in this view
The importance of each reason is valued at
strong (+++/---), medium (++/--), or weak (+/-)
The reason codes are a powerful feature that can supplement business decisions