Please enable JavaScript.
Coggle requires JavaScript to display documents.
Build Candidate Models (Starting the process (select the target feature),…
Build Candidate Models
Starting the process
select the target feature
find targets in feature list by clicking "use as target"
there will be true or false options that come up
data robot will chose which metric to use or you can choose
for regression problems other measures are used
logloss evaluates accuracy on the model
Advanced Options
click show advanced options
try different options to see what they are used for
"Random" pulls out a percentage of data that you specify
Stratified works similarly but tries to get random samples that have the same target
cross - validation illustrates two things
Partition Feature
method for determining cases that have different folds
different from other features because the user has to do the randomization
1 more item...
reminder of the distribution between samples spilt into folds and holdouts
specifies the cross-validation scores between averages and models used
Starting the analytical process
Quick run / auto pilot
start samples from 16% of the data
Informative feature represents data that hasn't been automatically tagged and puts out two options, duplicate and too few values
Keep autopilot up and click start
follow steps outlined in book on DataRobot
Model Selection Process
introduction sidebar shows all running algorithms
More CPU's to an algorithm means more processing power needed
Look over the tournament round!