convergence_analysis module

convergence_analysis.py. Study of the convergence through the evolution of the Spearman and Pearson correlation + measure of the mAP value every 5 iteration.

Convergence analysis

Usage:

python convergence_analysis.py -N [1] -t [2] -d [3] -c [4] -ts [5] -s [6] -nmin [7] -nmax [8] -di [9] -o [10] -te [11] -in [12] -g [13] -cfg [14] -v [15] --debug --help
where:
  • Default options are found in the configuration.ini file.
  • [1] -i, --iter: number of classification iterations.
  • [2] -t, --threads: number of cores to use.
  • [3] -d, --dataset: dataset to use.
  • [4] -c, --classifier: classifier to use.
  • [5] -ts, --trainsize: proportion of dataset to use for training.
  • [6] -s, --sim: similarity type to use (EM not supported).
  • [7] -nmin: minimum number of synthetic labels.
  • [8] -nmax: maximum number of synthetic labels.
  • [9] -di, --distrib: synthetic annotation mode (RND, UNI, OVA).
  • [10] -o, --output: output folder.
  • [11] -te, --temp: temporary folder.
  • [12] -in: input data file.
  • [13] -g, --ground: ground-truth file.
  • [14] -cfg, --config_file: provide a custom configuration file.
  • [15] -v, --verbose: controls verbosity level (0 to 4).
  • -db, --debug: debug mode (save temporary files).
  • -h, --help

Computes N iterations of SIC and compares the final similarity matrix to partial matrices in past iterations (see steps in convergence_analysis.py).