[FeaturesRefactor] Features::view()/preprocess() #3970
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- implement DotIterator, SGMatrix and SGVector iterators - Perceptron training with iterators + unit test SGVector: make SGVector(index_t len) 'ctor zero-initialize memory with SG_CALLOC like SGMatrix(index_t rows, index_t cols) 'ctor.
vigsterkr
reviewed
Aug 24, 2017
| SGMatrix<ST> target(first_vec.vlen, get_num_vectors()); | ||
| target.set_column(0, first_vec); | ||
|
|
||
| for (index_t i = 1; i < get_num_vectors(); ++i) |
There was a problem hiding this comment.
this whole part above should be rather done by a smart memcpy instead of setting the columns separately. needs a stride etc.
| @@ -1067,6 +1048,24 @@ CFeatures* CDenseFeatures<ST>::create_merged_copy(CFeatures* other) | |||
| return create_merged_copy(list); | |||
| } | |||
|
|
|||
| template <class ST> | |||
| Some<CDenseFeatures<ST>> CDenseFeatures<ST>::view(const SGVector<index_t>& subset) | |||
There was a problem hiding this comment.
the problem with this is that this right away makes it inaccessible/usable from SWIG interfaces... we are having difficulties with exposing Some in SWIG
| auto feats_view = wrap(this->duplicate()); | ||
|
|
||
| auto sg_subset = SGVector<index_t>(subset.size()); | ||
| std::copy(subset.cbegin(), subset.cend(), sg_subset.data()); |
There was a problem hiding this comment.
sg_memcpy ? as both std::vector and SGVector backed by a continuous memory chunk
|
This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions. |
|
This issue is now being closed due to a lack of activity. Feel free to reopen it. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
[Work in progress]
Continues #3968, only the last commit is relevant.
The design is outlined in this gist.
I might split this into a few PRs that builds on the fundamental changes [1-2-3],
[4] cross validation, [5]
add/remove_subset()replacement, [6] preprocessors.Anyway, what's in here:
Base methods to construct new features
Features::preprocess()/view().Base method to construct new labels
Labels::view(), this requires a shallow copy 'ctorlike in
Featuresclasses (duplicate()+ copy 'ctors),implemented for (Dense/Binary/Multiclass/Regression)Labels.
DenseFeatures:get_feature_vector()method;eval()method;get_feature_matrix()constness.CrossValidation: useview()instead ofadd/remove_subset().Replace
add/remove_subset()withview()+ unit tests failing due toDenseFeatureschanges [see below]:StochasticGBMachine: refactorget_subset()and related functionsto use a new labels object instead of adding the subset to
m_labels,refactor unit tests and add case that covers
subset_frac < 1branch;LMNNImpl: refactor to usepreprocess()instead ofapply_to_feature_matrix();CARTree: fix feature matrix constness, minor changes to unit test (some);NormOne::apply_to_feature_vector()uses linalg;PruneVarSubMean::init()to useDotFeatures::get_mean()/get_std()(the latter was added to
DotFeatures), add unit test;RandomFourierGaussPreproc: fix feature matrix constness;MultipleProcessorsunit test with eval case;LeastAngleRegression.ols_equivalenceunit test to usepreprocess()[Misc] Initializer list constructor for
SGMatrix/Vector(not a big breakthrough but anyway useful for unit tests with fixed data...)
Unit tests: