I enjoy using the package and wonder if you can add the weight feature into the NN2 function where distances are normalised/adjusted according the weights ? (meanwhile, if there is any hack I can do, I appreciate any comments)

I'm creating a package which requires "RANN.L1" as a dependency. The package "RANN.L1" is therefore listed under Imports: in the DESCRIPTION file of the package, and the remote dependency location is specified as jefferis/RANN@master-L1 under Remotes: in the same file. However, as RANN.L1 is a branch of the RANN repo, the package build under Travis CI fails, presumably as it can't associate the "RANN.L1" import with the "RANN" remote. Any suggestions? Having the jefferis/RANN@master-L1 branch as a separate repo under jefferis/RANN.L1 could be a quick fix? Thanks.

I am using the version in CRAN, 2.4.1. I have a dataset that can contain objects with distances of zero. When I run a nearest neighbor search for all objects within a specified radius, the indexes can become switched. For example:

nn.idx[70,] = 71 70
instead of the expected 70 71.

I'm not sure if objects with zero distance are not allowed, or if its a bug. I can supply example data if necessary.

input with >0 rows but 0 cols can cause R to crash

On 11 Dec 2018, at 09:00, Prof Brian Ripley [email protected] wrote:

This concerns packages

FUNLDA RANN RJafroc SGL distances mixAK projmanr yaImpute

R-devel has switched to C++11 as the default standard, and your packages failed to compile on Solaris: see
https://www.stats.ox.ac.uk/pub/bdr/Solaris-gcc/ for the installation logs. (g++ 5.2 was used: Solaris is not often
updated ....)

You were warned about ERR in 'Writing R Extensions' §1.6.4 : it looks like 'EBP' gives another header clash.

For 'distances', you are trying to compile in a sub-directory with C++98 yet the src directory with C++11: they cannot
portably be mixed. Use one or the other for both.

Please correct ASAP and before Jan 11 to safely retain the package on CRAN.

Hi, thanks for the nn2 function, it has transformed the efficiency of distance calculations for large datasets. I have a related problem, where I would like to calculate the distance from each point to the closest point on a line. I can convert the line to a set of coordinates and then run nn2, however, this is sort of cheating as it doesn't formally calculate distance to the line that connects each of those coordinates. I just wondered whether you were working on (or could be persuaded to! :)) a function to do this? At the moment dist2line from geosphere is the best option I have, but it is very slow, even if parallelized, on large datasets. I'm working on a function to first find the nearest line, then sample points along that line and run nn2, but its pretty inelegant...

Thanks again!

In page 10 of the manual of ANN (https://www.cs.umd.edu/~mount/ANN/Files/1.1.2/ANNmanual_1.1.pdf) . It mentioned that we can return the number of points within a radius when setting the ANNdist to sqRad and k = 0. This is very useful when we are trying to count the number of points for each sample whose distance is within the radius. Can you help to also return this value?

I think ANNkdFRPtsInRange is actually the value we want. But I don't know how can we return that value in the get_NN_2Set and nn2 function

int ANNkd_tree::annkFRSearch(
	ANNpoint			q,				// the query point
	ANNdist				sqRad,			// squared radius search bound
	int					k,				// number of near neighbors to return
	ANNidxArray			nn_idx,			// nearest neighbor indices (returned)
	ANNdistArray		dd,				// the approximate nearest neighbor
	double				eps)			// the error bound
{
	ANNkdFRDim = dim;					// copy arguments to static equivs
	ANNkdFRQ = q;
	ANNkdFRSqRad = sqRad;
	ANNkdFRPts = pts;
	ANNkdFRPtsVisited = 0;				// initialize count of points visited
	ANNkdFRPtsInRange = 0;				// ...and points in the range

	ANNkdFRMaxErr = ANN_POW(1.0 + eps);
	ANN_FLOP(2)							// increment floating op count

	ANNkdFRPointMK = new ANNmin_k(k);	// create set for closest k points
										// search starting at the root
	root->ann_FR_search(annBoxDistance(q, bnd_box_lo, bnd_box_hi, dim));

	for (int i = 0; i < k; i++) {		// extract the k-th closest points
		if (dd != NULL)
			dd[i] = ANNkdFRPointMK->ith_smallest_key(i);
		if (nn_idx != NULL)
			nn_idx[i] = ANNkdFRPointMK->ith_smallest_info(i);
	}

	delete ANNkdFRPointMK;				// deallocate closest point set
	return ANNkdFRPtsInRange;			// return final point count
}

Hi @jefferis, it would be great if your c++ functions were exported by adding // [[Rcpp::interfaces(r, cpp)]] to the relevant c++ functions for using your indexing methods.

I've already used those in my projects because it uses one of the fastest implementations of indexing, but currently I'm just copying your code and using it my personal projects.

Thanks a lot!

RANN1 (by @krlmlr on top of the original RANN) supports the Manhattan (L1) norm rather than the standard Euclidean (L2) metric. It should be pretty much for release to CRAN – and I propose that he is the maintainer! Presumably the version numbers should be kept in sync as much as possible.

This was difficult to do, because the metric is defined via preprocessor macros. I still think we could make it work, by including the library twice in two namespaces. This would be a cleaner approach, one package less to maintain.

The ANN docs suggest that the Euclidean metric is used by default. Some digging in the code and in above docs suggest that this is also true for RANN.

According to Section 2.2.1 (p. 14 ff.), only the macros in the code linked above need to be redefined to use other Minkowski norms. To define a code base where Manhattan, Euclidean and "max" metric are simultaneously accessible, this probably needs to be replaced by a template parameter that defines the implementation of the four operators.

The purpose is to accelerate statistical matching using the Gower distance: I think I can transform a "Gower distance problem" to a "Manhattan distance problem" but then RANN should be able to solve it (it is called by the StatMatch package).

Does this make sense to you?

See jefferis/nabor#9

the following line leaves my R process unresponsive, no segfault, no 100% cpu:

RANN::nn2(iris[,1:4], treetype = "bd")

treetype="kd" works fine.

Could you please install

remotes::install_github(c("ropenscilabs/tic", "ropenscilabs/travis"))

and run

tic::use_tic()

If asked to overwrite .travis.yml, select "yes". You may need to set up a Travis access token with travis::browse_travis_token() .

Benefits:

1. Automatic building of pkgdown documentation
2. We can build pkgdown for {RANN} and {RANN.L1} and host in parallel
3. Testing on Windows

I can also do it if you increase my permissions to this repository.

I just noticed {RANN.L1} has been orphaned on CRAN, I'd like to update.