Add a meta information extractor #1031
Conversation
Add support for specifying the minimum number of additional arguments.
Add a tool for extracting meta information about machine learning algorithms.
| typename PredictionsType, | ||
| typename WeightsType, | ||
| bool DatasetInfo, | ||
| bool NumClasses> |
There was a problem hiding this comment.
If I am understanding correctly, you will use this to determine regression or classification types, right? NumClasses should be true for all classification types (some changes still necessary there, would you like me to do them?) and false for all regression types.
There was a problem hiding this comment.
Yes, you're right again. It will be nice if you normalize API across classification algorithms.
| bool NumClasses> | ||
| struct TrainForm; | ||
|
|
||
| template<typename PT, typename WT, typename...SignatureParams> |
There was a problem hiding this comment.
Minor style comment---should there be a space between typename... and SignatureParams?
There was a problem hiding this comment.
I don't mind adding a space. Maybe it's worth to explicitly state about this convention in the style guidelines?
|
|
||
| /* Forms of Train with weights for classification algorithms */ | ||
| using WTF4 = TrainForm<MT, PT, WT, false, true>; | ||
| using WTF5 = TrainForm<MT, PT, WT, true, true>; |
There was a problem hiding this comment.
I like the names of these variables. :)
| CheckPredictionsType<DecisionTree<>, arma::Row<size_t>, arma::mat, | ||
| arma::Row<size_t>>(); | ||
| CheckPredictionsType<DecisionTree<>, arma::Row<char>, arma::mat, | ||
| arma::Row<char>>(); |
There was a problem hiding this comment.
Just out of curiosity, did you try alternate types for other classes than DecisionTree<> and have a failure? I think since #290 is not done yet, there may be problems with that for some classes.
There was a problem hiding this comment.
I was trying it also for HoeffdingTree<> in the line 121:
CheckPredictionsType<HoeffdingTree<>, arma::Row<size_t>, arma::imat>();
Here arma::imat is used is as the data type in the Train method.
| * A wrapper struct for holding a Train form. | ||
| * | ||
| * @tparam MatType The type of data. | ||
| * @tparam PredictionsType The type of predictions. |
There was a problem hiding this comment.
The assumption here is that LabelsType == PredictionsType, right? I think that's completely reasonable, I just want to make sure I am comprehending fully.
src/mlpack/tests/cv_test.cpp
Outdated
| @@ -113,7 +113,7 @@ void CheckPredictionsType() | |||
| BOOST_AUTO_TEST_CASE(PredictionsTypeTest) | |||
| { | |||
| CheckPredictionsType<LinearRegression, arma::rowvec>(); | |||
| CheckPredictionsType<FFN<>, arma::mat>(); | |||
There was a problem hiding this comment.
I have disabled the check since API for FFN have changed (now the Train method takes data and predictions as values, rather than as const references).
|
By the way, if you don't see any problems, feel free to merge. |
Add a tool for extracting meta information about machine learning algorithms.