- Fixed an issue where the session would crash if the Annoy nearest neighbor search was unable to find k neighbors for an item.

Even with a fix for the bug mentioned above, if the nearest neighbor index file is larger than 2GB in size, Annoy may not be able to read the data back in. This should only occur with very large or high-dimensional datasets. The nearest neighbor search will fail under these conditions. A work-around is to set `n_threads = 0`

, because the index will not be written to disk and re-loaded under these circumstances, at the cost of a longer search time. Alternatively, set the `pca`

parameter to reduce the dimensionality or lower `n_trees`

, both of which will reduce the size of the index on disk. However, either may lower the accuracy of the nearest neighbor results.

Initial CRAN release.

- New parameter,
`tmpdir`

, which allows the user to specify the temporary directory where nearest neighbor indexes will be written during Annoy nearest neighbor search. The default is`base::tempdir()`

. Only used if`n_threads > 1`

and`nn_method = "annoy"`

.

Fixed an issue with

`lvish`

where there was an off-by-one error when calculating input probabilities.Added a safe-guard to

`lvish`

to prevent the gaussian precision, beta, becoming overly large when the binary search fails during perplexity calibration.The

`lvish`

perplexity calibration uses the log-sum-exp trick to avoid numeric underflow if beta becomes large.

- New parameter:
`pcg_rand`

. If`TRUE`

(the default), then a random number generator from the PCG family is used during the stochastic optimization phase. The old PRNG, a direct translation of an implementation of the Tausworthe “taus88” PRNG used in the Python version of UMAP, can be obtained by setting`pcg_rand = FALSE`

. The new PRNG is slower, but is likely superior in its statistical randomness. This change in behavior will be break backwards compatibility: you will now get slightly different results even with the same seed. - New parameter:
`fast_sgd`

. If`TRUE`

, then the following combination of parameters are set:`n_sgd_threads = "auto"`

,`pcg_rand = FALSE`

and`approx_pow = TRUE`

. These will result in a substantially faster optimization phase, at the cost of being slightly less accurate and results not being exactly repeatable.`fast_sgd = FALSE`

by default but if you are only interested in visualization, then`fast_sgd`

gives perfectly good results. For more generic dimensionality reduction and reproducibility, keep`fast_sgd = FALSE`

. - New parameter:
`init_sdev`

which specifies how large the standard deviation of each column of the initial coordinates should be. This will scale any input coordinates (including user-provided matrix coordinates).`init = "spca"`

can now be thought of as an alias of`init = "pca", init_sdev = 1e-4`

. This may be too aggressive scaling for some datasets. The typical UMAP spectral initializations tend to result in standard deviations of around`2`

to`5`

, so this might be more appropriate in some cases. If spectral initialization detects multiple components in the affinity graph and falls back to scaled PCA, it uses`init_sdev = 1`

. - As a result of adding
`init_sdev`

, the`init`

options`sspectral`

,`slaplacian`

and`snormlaplacian`

have been removed (they weren’t around for very long anyway). You can get the same behavior by e.g.`init = "spectral", init_sdev = 1e-4`

.`init = "spca"`

is sticking around because I use it a lot.

- Spectral initialization (the default) was sometimes generating coordinates that had too large a range, due to an erroneous scale factor that failed to account for negative coordinate values. This could give rise to embeddings with very noticeable outliers distant from the main clusters.
- Also during spectral initialization, the amount of noise being added had a standard deviation an order of magnitude too large compared to the Python implementation (this probably didn’t make any difference though).
- If requesting a spectral initialization, but multiple disconnected components are present, fall back to
`init = "spca"`

. - Removed dependency on C++
`<random>`

header. This breaks backwards compatibility even if you set`pcg_rand = FALSE`

. `metric = "cosine"`

results were incorrectly using the unmodified Annoy angular distance.- Numeric matrix columns can be specified as the target for the
`categorical`

metric (fixes https://github.com/jlmelville/uwot/issues/20).

- Data is now stored column-wise during optimization, which should result in an increase in performance for larger values of
`n_components`

(e.g. approximately 50% faster optimization time with MNIST and`n_components = 50`

). - New parameter:
`pca_center`

, which controls whether to center the data before applying PCA. It would be typical to set this to`FALSE`

if you are applying PCA to binary data (although note you can’t use this with setting with`metric = "hamming"`

) - PCA will now be used when the
`metric`

is`"manhattan"`

and`"cosine"`

. It’s still*not*applied when using`"hamming"`

(data still needs to be in binary format, not real-valued). - If using mixed datatypes, you may override the
`pca`

and`pca_center`

parameter values for a given data block by using a list for the value of the metric, with the column ids/names as an unnamed item and the overriding values as named items, e.g. instead of`manhattan = 1:100`

, use`manhattan = list(1:100, pca_center = FALSE)`

to turn off PCA centering for just that block. This functionality exists mainly for the case where you have mixed binary and real-valued data and want to apply PCA to both data types. It’s normal to apply centering to real-valued data but not to binary data.

- Fixed bug that affected
`umap_transform`

, where negative sampling was over the size of the test data (should be the training data). - Some other performance improvements (around 10% faster for the optimization stage with MNIST).
- When
`verbose = TRUE`

, log the Annoy recall accuracy, which may help tune values of`n_trees`

and`search_k`

.

- New parameter:
`n_sgd_threads`

, which controls the number of threads used in the stochastic gradient descent. By default this is now single-threaded and should result in reproducible results when using`set.seed`

. To get back the old, less consistent, but faster settings, set`n_sgd_threads = "auto"`

. - API change for consistency with Python UMAP:
`alpha`

is now`learning_rate`

.`gamma`

is now`repulsion_strength`

.

- Default spectral initialization now looks for disconnected components and initializes them separately (also applies to
`laplacian`

and`normlaplacian`

). - New
`init`

options:`sspectral`

,`snormlaplacian`

and`slaplacian`

. These are like`spectral`

,`normlaplacian`

,`laplacian`

respectively, but scaled so that each dimension has a standard deviation of 1e-4. This is like the difference between the`pca`

and`spca`

options.

- Hamming distance support (was actually using Euclidean distance).
- Smooth knn/perplexity calibration results had a small dependency on the number of threads used.
- Anomalously long spectral initialization times should now be reduced.
- Internal changes and fixes thanks to a code review by Aaron Lun (https://github.com/ltla).

- New parameter
`pca`

: set this to a positive integer to reduce matrix of data frames to that number of columns using PCA. Only works if`metric = "euclidean"`

. If you have > 100 columns, this can substantially improve the speed of the nearest neighbor search. t-SNE implementations often set this value to 50.

- Laplacian Eigenmap initialization convergence failure is now correctly detected.
- C++ code was over-writing data passed from R as a function argument.

- Highly experimental mixed data type support for
`metric`

: instead of specifying a single metric name (e.g.`metric = "euclidean"`

), you can pass a list, where the name of each item is the metric to use and the value is a vector of the names of the columns to use with that metric, e.g.`metric = list("euclidean" = c("A1", "A2"), "cosine" = c("B1", "B2", "B3"))`

treats columns`A1`

and`A2`

as one block, using the Euclidean distance to find nearest neighbors, whereas`B1`

,`B2`

and`B3`

are treated as a second block, using the cosine distance. - Factor columns can also be used in the metric, using the metric name
`categorical`

. `y`

may now be a data frame or matrix if multiple target data is available.- New parameter
`target_metric`

, to specify the distance metric to use with numerical`y`

. This has the same capabilities as`metric`

. - Multiple external nearest neighbor data sources are now supported. Instead of passing a list of two matrices, pass a list of lists, one for each external metric.
- More details on mixed data types can be found at https://github.com/jlmelville/uwot#mixed-data-types.
- Compatibility with older versions of RcppParallel (contributed by sirusb).
`scale = "Z"`

To Z-scale each column of input (synonym for`scale = TRUE`

or`scale = "scale"`

).- New scaling option,
`scale = "colrange"`

to scale columns in the range (0, 1).

- Hamming distance is now supported, due to upgrade to RcppAnnoy 0.0.11.

- For supervised UMAP with numeric
`y`

, you may pass nearest neighbor data directly, in the same format as that supported by`X`

-related nearest neighbor data. This may be useful if you don’t want to use Euclidean distances for the`y`

data, or if you have missing data (and have a way to assign nearest neighbors for those cases, obviously). See the Nearest Neighbor Data Format section for details.

- New parameter
`ret_nn`

: when`TRUE`

returns nearest neighbor matrices as a`nn`

list: indices in item`idx`

and distances in item`dist`

. Embedded coordinates are in`embedding`

. Both`ret_nn`

and`ret_model`

can be`TRUE`

, and should not cause any compatibility issues with supervised embeddings. `nn_method`

can now take precomputed nearest neighbor data. Must be a list of two matrices:`idx`

, containing integer indexes, and`dist`

containing distances. By no coincidence, this is the format return by`ret_nn`

.

- Embedding to
`n_components = 1`

was broken (https://github.com/jlmelville/uwot/issues/6) - User-supplied matrices to
`init`

parameter were being modified, in defiance of basic R pass-by-copy semantics.

`metric = "cosine"`

is working again for`n_threads`

greater than`0`

(https://github.com/jlmelville/uwot/issues/5)

*August 5 2018*. You can now use an existing embedding to add new points via`umap_transform`

. See the example section below.*August 1 2018*. Numerical vectors are now supported for supervised dimension reduction.*July 31 2018*. (Very) initial support for supervised dimension reduction: categorical data only at the moment. Pass in a factor vector (use`NA`

for unknown labels) as the`y`

parameter and edges with bad (or unknown) labels are down-weighted, hopefully leading to better separation of classes. This works remarkably well for the Fashion MNIST dataset.*July 22 2018*. You can now use the cosine and Manhattan distances with the Annoy nearest neighbor search, via`metric = "cosine"`

and`metric = "manhattan"`

, respectively. Hamming distance is not supported because RcppAnnoy doesn’t yet support it.