Binary FastTree/Forest samples using T4 templates. #3035
Conversation
| yield return new DataPoint | ||
| { | ||
| Label = label, | ||
| // Create random features that are correlated with label. |
There was a problem hiding this comment.
Please add Note that data points with Label=false tend to have larger feature values. #Resolved
There was a problem hiding this comment.
| // This example requires installation of additional NuGet package | ||
| // <a href="https://www.nuget.org/packages/Microsoft.ML.FastTree/">Microsoft.ML.FastTree</a>. | ||
| public static void Example() | ||
| { |
There was a problem hiding this comment.
Is it possible to execute those Example()s in a test? Just want to make sure they are always executable. #Resolved
There was a problem hiding this comment.
| var mlContext = new MLContext(seed: 0); | ||
|
|
||
| // Create a list of training examples. | ||
| var examples = GenerateRandomDataPoints(1000); |
There was a problem hiding this comment.
examples [](start = 16, length = 8)
dataPoints? #Resolved
There was a problem hiding this comment.
| } | ||
|
|
||
| // Class used to capture predictions. | ||
| private class Prediction |
There was a problem hiding this comment.
Prediction [](start = 22, length = 10)
Not sure if we want to produce probability for calibrated models or socre for non-calibrated models. #Resolved
There was a problem hiding this comment.
I had probabilities originally. But was surprised that FastForest wasn't calibrated and FastTree was. So for the sake of sharing the same T4 template and not having too much conditional formatting I'm using PredictedLabel which all binary trainers are guaranteed to produce. Scores are a bit confusing because it's not clear what -50 or +40.1 mean without calibration.
In reply to: 267449313 [](ancestors = 267449313)
| public static class FastTreeWithOptions | ||
| { | ||
| // This example requires installation of additional NuGet package | ||
| // <a href="https://www.nuget.org/packages/Microsoft.ML.FastTree/">Microsoft.ML.FastTree</a>. |
There was a problem hiding this comment.
Very useful comment! #ByDesign
There was a problem hiding this comment.
It looks great now. I think .ttinclude can be made more generic as we create more samples. Thanks for doing it @shmoradims.
Codecov Report
@@ Coverage Diff @@
## master #3035 +/- ##
==========================================
+ Coverage 72.41% 72.41% +<.01%
==========================================
Files 803 803
Lines 143851 143851
Branches 16173 16173
==========================================
+ Hits 104171 104172 +1
+ Misses 35258 35257 -1
Partials 4422 4422
|
ghost
locked as resolved and limited conversation to collaborators
Related to #2522. The *.cs files are auto-generated. Please review the .tt and .ttinclude files.