Utilities¶

Pairwise Cross Mapping¶

Overview¶

edm-xmap performs pairwise cross mapping between many time series using Simplex projection. The input is given as a 2D array where a column corresponds to a time series. The output is a 2D array where each element represents the cross map rho between two time series. Optionally, pairwise cross-correlation can also be computed.

Both the input and output are stored as HDF5 files. The output HDF5 file contains up to four datasets, which are:

ccm: Cross map rho
e: Optimal embedding dimension
rho: Pairwise cross-correlation (if --rho is enabled)
rhodiff: ccm - abs(rho) (if --rho-diff is enabled)

Command-line options¶

edm-xmap: All-to-all Convergent Cross Mapping Tool

Usage:
  edm-xmap [OPTION...] input output
  -d, --dataset arg           HDF5 dataset name (default: "values")
  -e, --max-embedding-dim arg Embedding dimension (default: 20)
  -t, --tau arg               Time delay (default: 1)
  --rho                       Compute cross correlation (default: false)
  --rho-diff                  Compute rho diff(default: false)
  -v, --verbose               Enable verbose output
  -h, --help                  Show this help

Example¶

Here, an HDF5 file Fish1_150a_Normo.h5 containing 154 time series with 1,600 time steps, is processed using edm-xmap.

$ h5ls Fish1_150a_Normo.h5
names                    Dataset {154}
values                   Dataset {1600, 154}

$ ./edm-xmap --rho --rho-diff Fish1_150a_Normo.h5 Fish1_150a_Normo_xmap.h5

$ h5ls Fish1_150a_Normo_xmap.h5
ccm                      Dataset {154, 154}
e                        Dataset {154}
rho                      Dataset {154, 154}
rhodiff                  Dataset {154, 154}

Micro benchmarks¶

Overview¶

kEDM includes several micro benchmarks to quickly measure the performance of bottleneck kernels using dummy datasets.

Command-line options¶

knn-bench: k-Nearest Neighbors Search Benchmark

Usage:
  knn-bench [OPTION...]
  -l, --length arg        Length of time series (default: 10,000)
  -e, --embedding-dim arg Embedding dimension (default: 20)
  -t, --tau arg           Time delay (default: 1)
  -i, --iteration arg     Number of iterations (default: 10)
  -v, --verbose           Enable verbose logging (default: false)
  -h, --help              Show this help

lookup-bench: kNN Lookup Benchmark

Usage:
  build-cuda/lookup-bench [OPTION...]
  -l, --length arg        Length of time series (default: 10,000)
  -n, --num-ts arg        Number of time series (default: 10,000)
  -e, --embedding-dim arg Embedding dimension (default: 20)
  -t, --tau arg           Time delay (default: 1)
  -i, --iteration arg     Number of iterations (default: 10)
  -v, --verbose           Enable verbose logging (default: false)
  -h, --help              Show this help