Please enable JavaScript.
Coggle requires JavaScript to display documents.
Ch. 15 Build Candidate Models (Model Selection Process (Models ordered by…
Ch. 15 Build Candidate Models
Starting the Process
select target feature during this stage
"Use as target" button found directly in feature list
or name of feature can be typed into target area
after selecting target feature
distribution of target feature displayed
DataRobot offers option of which metrics to optimize models for
LogLoss (Accuracy)
model evaluated on probabilities
Starting Analytical Process
Prep data with three options
Autopilot
starts analysis at 16% of the sample
eventually determines which models to run w/ 32% of sample
Quick
abbreviated version of Autopilot
produces almost as good models by shortcutting DataRobot ML process
starts analysis at 32% of sample
only 4 models automatically cross validated
only 1 blender algorithm applied
Manual
After hitting "Start"
sidebar shows 7 steps
1) Setting target feature
2) Creating CV and holdout partitions
3) Characterizing target variable
4) Loading data set and preparing data
Saving target and partitioning information
6) Analyzing features
7) Calculating list of models
New column added to feature list: Importance
green bar shows relative importance of feature when examined against target independently of all other features
Model Selection Process
sidebar showing several algorithms running on Amazon Cloud Environment
"Worker" can be thought of as a dedicated computer
Models ordered by validation score
Tournament rounds
Larger selection of sample being test between models each time
cross-validation process run on top 8 models
blending