lettersim-ot-rsa/fig_code/99_estimate_spaces.py

# %% import modules ======================
import os
n_threads = str(1)
os.environ["OMP_NUM_THREADS"] = n_threads
os.environ["OPENBLAS_NUM_THREADS"] = n_threads
os.environ["MKL_NUM_THREADS"] = n_threads
os.environ["VECLIB_MAXIMUM_THREADS"] = n_threads
os.environ["NUMEXPR_NUM_THREADS"] = n_threads

import sys
import os.path as op
sys.path.append(op.join(op.dirname(__file__), '..'))
import scold
from scold import draw, text_arr_sim, arr_sim, text_arr_sim_wasserstein

import numpy as np
from tqdm import tqdm
import itertools

from joblib import Parallel, delayed
n_jobs = -1

# %% setup

font = 'Arial-Lgt.ttf'
font_size = 75
round_method = None

char1 = 'a'
char2 = 'c'

char2_arr = draw.text_array(char2, font=font, size=font_size, method=round_method)

n_scale = 250
n_rotation = 250

logscale = np.linspace(np.log(0.5), np.log(2), num=n_scale, endpoint=True)
rotation = np.linspace(-180, 180, num=n_rotation, endpoint=True)

# %% the Jaccard distances

space_j_list = Parallel(n_jobs=n_jobs)(
    delayed(text_arr_sim.text_arr_sim)(
        a=char1, b_arr=char2_arr,
        font_a=font,
        size=font_size, method=round_method,
        measure='jaccard',
        translate=True,
        scale_val=np.exp(ls), fliplr=False, flipud=False, rotate_val=r)
        for ls, r in tqdm(
            itertools.product(logscale, rotation),
            desc='Mapping Jaccard space', total=n_scale*n_rotation))

space_j = np.array([1 - x['jaccard'] for x in space_j_list]).reshape((n_scale, n_rotation))

np.save(op.join('fig_code', '99_estimate_spaces_jaccard'), space_j)

# %% the wasserstein distances

space_w_list = Parallel(n_jobs=n_jobs)(
    delayed(text_arr_sim.text_arr_sim)(
        a=char1, b_arr=char2_arr,
        font_a=font,
        size=font_size, method=round_method,
        measure='partial_wasserstein',
        translate=True,
        scale_val=np.exp(ls), fliplr=False, flipud=False, rotate_val=r,
        partial_wasserstein_kwargs={'scale_mass':True, 'scale_mass_method':'proportion', 'mass_normalise':False, 'distance_normalise':False, 'ins_weight':0.0, 'del_weight':0.0, 'translation': 'opt'})
        for ls, r in tqdm(
            itertools.product(logscale, rotation),
            desc='Mapping Wasserstein space', total=n_scale*n_rotation))

space_w = np.array([x['partial_wasserstein'] for x in space_w_list]).reshape((n_scale, n_rotation))

np.save(op.join('fig_code', '99_estimate_spaces_wasserstein'), space_w)