LukasRuppert

Hi, sorry for the late reply, do you mean updating the values of existing points in the dataset? In that case, running one or two refinement iterations (depending on how...

Sorry, there is no simple API. Have a look at the examples for different datasets in the `src` folder. The metadata of each dataset, i.e. data type, dimension, integer types...

Due to several reasons, I sadly can't answer this as precisely as you might like. On the one hand, there are actually several kernels involved in the graph construction each...

Hi, I've looked up the corresponding configuration. Use K = 24, tau_build = 0.7, refinements = 0, tau_query = 1.08. At the end of `demo.cpp` add `query_function(1.08f);` and run the...

Yes, I still have them: https://gist.github.com/LukasRuppert/6f80e283d40eb09b734738772b30bff0 The config shown there is for NYTimes. For GloVe, you need to change the dimension from 256 to 200. And there might be nicer...

The recall rate (as in "did I find the 1 closest neighbor within the k results") is `rKQuery_including_duplicates`. The consensus (as in "did I find the k nearest neighbors within...

We share the loop over all N elements, but for every result `j`, you only look at the `j`th entry in the ground truth. Now, suppose `gt[i][2]` was not found...

Memory usage is described in Appendix C.1.1 If you run the program in verbose mode (`--v=4`), it will also tell you the sizes. Alternatively, look at the code: https://github.com/cgtuebingen/ggnn/blob/release_0.5/include/ggnn/cuda_knn_ggnn_gpu_instance.cuh#L208-L260

Yes, you will get the following output: ``` I20231031 10:04:19.683550 779255 cuda_knn_ggnn_graph_device.cuh:76] GGNNGraphDevice(): allocating GPU memory... (0.0926831 GB graph + 0.476837 GB base) ``` Which is exactly the amount of...