NEWS.md

RcppHNSW (development version)

New features

• New class: HnswEuclidean. This uses Euclidean distances internally and will be returned from hnsw_build when distance = "euclidean" is specified. This fixes an issue where if you created an index with hnsw_build and distance = "euclidean" (the default), then after saving, you would be unable to reload the index and have it find Euclidean distances. You would have to create it as an HsnwL2 object and take the square root of the distances yourself (#21).
• The Hnsw constructors now expose a random_seed parameter that you can use to set the random seed used in constructing the HNSW index. Internally, the hnsw_build and hnsw_knn functions will use a random seed based on R's RNG state. This means that if you want to reproduce results, you need to set the random seed in R via set.seed before calling those functions. Based on a request by Maciej Beręsewicz (#23).

RcppHNSW 0.6.0

New features

• Updated hnswlib to version 0.8.0.

RcppHNSW 0.5.0

New features

• Updated hnswlib to version 0.7.0. Note that I made some very minor changes to the code to silence some compiler warnings. These changes have been submitted up-stream to the hnswlib project.
• For high-dimensional data, there can be a noticeable CPU overhead in copying data out of the non-contiguous memory regions when row-wise data is used. If you wish to provide data where each column of the input matrix contains an item to be indexed/search then see the following additions to the API:
  • For the class-based API: addItemsCol, getAllNNsCol and getAllNNsListCol are the column-based equivalents of addItems, getAllNNs and getAllNNsList, respectively. Note that the returned nearest neighbor data from getAllNNsCol and getAllNNsListCol are also stored by column, i.e. the matrices have dimensions k x n where k is the number of neighbors, and n the number of items in the data being searched.
  • For the function-based API, a new parameter byrow has been added to hnsw_knn, hnsw_build and hnsw_search. By default this is set to TRUE and indicates that the items in the input matrix are found in each row. To pass column-stored items, set byrow = FALSE. Any matrices returned by hnsw_search and hnsw_knn will now follow the convention provided by the value of byrow: i.e. if byrow = FALSE, the matrices contain nearest neighbor information in each column.
• new method: getItems, which returns a matrix of the data vectors in the index with the specified integer identifiers. From a feature request made by d4tum (#18).

Bug fixes and minor improvements

• The progress parameter in the functional interface no longer does anything. When verbose = TRUE, a progress bar is no longer shown.
• Due to a breaking change in roxygen2 7.0.0, there was a missing package alias in the documentation.

RcppHNSW 0.4.1

Bug fixes and minor improvements

• Rolled back to hnswlib v0.4.0 due to valgrind problems in v0.6.2

RcppHNSW 0.4.0

New features

• Updated hnswlib to version 0.6.2.

Bug fixes and minor improvements

• Minor future-proofing of licensing: RcppHNSW is now GPLv3 or later, rather than GPLv3 only.

RcppHNSW 0.3.0

New features

• Multi-threading support is now available. Use the setNumThreads method if using the object-based API, and the n_threads parameter in the hnsw_* function API. For finer control, a setGrainSize and grain_size option is also available in the object and function interface respectively. Thank you to Dmitriy Selivanov for a lot of the work on this.
• Updated hnswlib to version 0.4.0.

Bug fixes and minor improvements

• Setting verbose = TRUE now has incurs substantially less computational overhead associated with calculating the progress bar. Thank you to Samuel Granjeaud for spotting the problem and coming up with various solutions.
• New parameter: progress. By default this is set to "bar" and will show the progress bar when verbose = TRUE. If you want a more terse output, set progress = NULL. progress = NULL will eventually be the default setting: for now, verbose = TRUE will get you the progress bar by default for backwards compatibility.
• No progress bar will be shown if you have less than 50 items to process.

RcppHNSW 0.2.0

New features

• Updated hnswlib to https://github.com/nmslib/hnswlib/commit/c5c38f0 (20 September 2019).
• A new method, markDeleted, that will remove an object from being retrieved from the index.
• A new method, resizeIndex, that allows the index to be increased without having to save and reload the index.
• A new method, size is available for the index objects and reports the number of items added to the index.

Bug fixes and minor improvements

• hnsw_search would stop if the number of rows in the input matrix was smaller than k. This check has been removed. Note that the correct behavior is to ensure that k is smaller than or equal to index$size() where index is the index you are searching. Because the size() method is new to this version, to preserve compatibility with old indexes, this check hasn't been added to hnsw_search. If this matters to you, manually compare index$size() with k before running hnsw_search. An error will be thrown if k neighbors can't be found in the index. Thank you to Yuxing Liao for spotting this and the pull request to remove the check.

RcppHNSW 0.1.0

Initial release.