In-memory & self-contained sample template. #2979
Conversation
| public float Label { get; set; } | ||
| // Predicted score from the trainer. | ||
| public float Score { get; set; } | ||
| } |
There was a problem hiding this comment.
I like extending the DataPoint class, because that is effectively what happens, columns get added. not a biggie though. #Resolved
There was a problem hiding this comment.
There's cons and pros here. I think keeping input and output separate is easier to understand for users.
In reply to: 266150723 [](ancestors = 266150723)
| } | ||
|
|
||
| private static IEnumerable<DataPoint> GenerateRandomDataPoints(int count) | ||
| private static IEnumerable<DataPoint> GenerateRandomDataPoints(int count, int seed=0) |
There was a problem hiding this comment.
private static IEnumerable GenerateRandomDataPoints(int count, int seed=0) [](start = 7, length = 86)
i like this, but do you think it will be quick to create them artificially for each task? Ranking and time series come to mind. #Resolved
There was a problem hiding this comment.
sometimes it's not easy, ranking being an example. For those, I'll keep the text-loader style.
so this template is mostly suitable for regression and binary classification.
In reply to: 266151216 [](ancestors = 266151216)
There was a problem hiding this comment.
Text data is also not easy to randomly generate.
In reply to: 266157156 [](ancestors = 266157156,266151216)
There was a problem hiding this comment.
I believe Zeeshan A meant meaningful text data but I don't think we need meaningful data to demonstrate the functionality of a module. The amount of data might be a problem to trainers, but to my knowledge, there is no trainer directly consuming strings.
There was a problem hiding this comment.
This sample is just a template. If in some scenarios it doesn't make sense, we can try text-loader instead.
In reply to: 266160094 [](ancestors = 266160094,266157156,266151216)
Codecov Report
@@ Coverage Diff @@
## master #2979 +/- ##
==========================================
- Coverage 72.3% 72.29% -0.02%
==========================================
Files 796 796
Lines 142349 142349
Branches 16051 16051
==========================================
- Hits 102923 102908 -15
- Misses 35041 35060 +19
+ Partials 4385 4381 -4
|
| @@ -45,11 +75,21 @@ private static IEnumerable<DataPoint> GenerateRandomDataPoints(int count) | |||
| } | |||
| } | |||
|
|
|||
| // Example with label and 50 feature values. A data set is a collection of such examples. | |||
| private class DataPoint | |||
| { | |||
| public float Label { get; set; } | |||
| [VectorType(50)] | |||
| public float[] Features { get; set; } | |||
| } | |||
There was a problem hiding this comment.
I assume all the samples are going to use same DataPoints (to be consistent), right? #Resolved
There was a problem hiding this comment.
I guess so. We now have samples, examples, instances, which are kind of less precise than data point. #Resolved
There was a problem hiding this comment.
the label type and number of features might change depending on the scenario, but the outline is the same.
In reply to: 266160360 [](ancestors = 266160360)
ghost
locked as resolved and limited conversation to collaborators
Related to #2726 I created this in-memory and self-contained sample for FastTree. I'll use the final version from this PR as template for the following samples.