diff options
author | Christoph Helma <helma@in-silico.ch> | 2016-03-01 13:52:38 +0100 |
---|---|---|
committer | Christoph Helma <helma@in-silico.ch> | 2016-03-01 13:52:38 +0100 |
commit | 799ac13e2424dafbabc817854ecd1f4a608c59d9 (patch) | |
tree | ef45fccf26151b54eb315d1607dcd4eca74a4f3b /loael.md | |
parent | afcaa6f7e8af8c9bc2cf419e8513051c6a595d88 (diff) |
svm regression with caret tune, no crossvalidation
Diffstat (limited to 'loael.md')
-rw-r--r-- | loael.md | 7 |
1 files changed, 5 insertions, 2 deletions
@@ -174,7 +174,8 @@ algorithm for the prediction of quantitative properties. First all fingerprint features with identical values across all neighbors are removed. The reamining set of features is used as descriptors for creating a local weighted PLS model with atom environments as descriptors and model similarities as weights. The -`plsr` function of the `pls` R package [@pls] is used for this purpose. +`pls` method from the `caret` R package [@Kuhn08] is used for this purpose. + Finally the local PLS model is applied to predict the activity of the query compound. @@ -186,6 +187,8 @@ weighted by its similarity to the query compound. Christoph: TODO +Prediction intervals were obtained from the `predict` function. + ### Validation For the comparison of experimental variability with predictive accuracies we @@ -311,7 +314,7 @@ These results are presented in [@fig:corr] and [@tbl:cv]. Please bear in mind th Training data | $r^2$ | RMSE --------------|---------------------------|------------------------- Experimental | 0.49 | 1.41 -Combined | 0.38 | 1.47 +Combined | 0.38 | 1.51 : Comparison of model predictions with experimental variability. {#tbl:common-pred} |