find_best_at_all_thresh

post_processors.find_best_at_all_thresh(thresh, batch_size, err_name='errors', time_name='timings')

This utility function finds the method was the fastest to reach a given threshold, at each threshold in the list thres.

Parameters:

• df (Pandas DataFrame) –

  The dataframe containing the errors and timings for each algorithm at each iterations, for several runs. Because I am a lazy coder:

  • Batch size must be constant

  • The algorithms must always be stored in df in the same order

• thresh (list) – A list of thresholds to be used for computing which method was faster.

• batch_size (int) – Number of algorithm runs to compare for the max pooling. Should be a multiple (typically 1x) of the number of algorithms.

Returns:

• scores_time (nd array) – A table (method x thresh) with how many times each method was the fastest to reach a given threshold. Here faster is understood in runtime.

• scores_it (nd array) – A table (method x thresh) with how many times each method was the fastest to reach a given threshold. Here faster is understood in number of iterations.