summaryrefslogtreecommitdiff
path: root/analysis
diff options
context:
space:
mode:
authorLoïc Guégan <loic.guegan@mailbox.org>2025-09-19 13:37:36 +0200
committerLoïc Guégan <loic.guegan@mailbox.org>2025-09-19 13:37:36 +0200
commitd06f583dc59c6317e16370aa29b96bacaeaaa770 (patch)
treece7654a8e4ec784c8d53f7954d648dad1c04f83e /analysis
parent6d4a2b0d9baef41ce26e5dd4639dd06dc99ea664 (diff)
Minor changes
Diffstat (limited to 'analysis')
-rw-r--r--analysis/README.md14
1 files changed, 14 insertions, 0 deletions
diff --git a/analysis/README.md b/analysis/README.md
new file mode 100644
index 0000000..48e2e5e
--- /dev/null
+++ b/analysis/README.md
@@ -0,0 +1,14 @@
+# Analysis
+
+## Files
+- kmeans.R: Just here for doing some tests
+- offline.R: contains two major functions:
+ - build_models: To generate K-fold cross-validation results (note that hyper-parameters for decisions tree is fixed (no validation set))
+ - generate_inputs: generate the inputs for the simulations experiments + the decision tree plots
+- in-situ.R: Implement the in-situ learning approach (Figure 4a 4b and 4c)
+ - For figure 4a and 4b we train the model with increasing amount of data from previous results as if we were using one policy per day (see section IV.A)
+ - For figure 4c, delta is generated by comparing using each policies in round-robin (one per days to perform the training)
+ to each previous paper results with single policy only (see paper section IV.A)
+## Notes
+Todo: remove minbucket=1 (does not impact the results)
+