def generate_bayesian_diagram(result_matrices,
algo_names=["algo1", "algo2"],
rope=0.01,
rho=1 / 5,
show_diagram=True,
save_diagram=None):
#rope=0.01 #we consider two classifers equivalent when the difference of accuracy is less that 1%
print(rope, rho)
#rho=1/5 #we are performing 10 folds, 10 runs cross-validation
pl, pe, pr = bt.hierarchical(result_matrices,
rope,
rho,
verbose=True,
names=algo_names)
samples = bt.hierarchical_MC(result_matrices, rope, rho, names=algo_names)
#plt.rcParams['figure.facecolor'] = 'black'
fig = bt.plot_posterior(samples, algo_names)
if show_diagram:
plt.show()
if save_diagram is not None:
plt.savefig(save_diagram)
return (pl, pe, pr)
# Algorithm names
alg_names = (('ReliefSeq', 'TuRF'), )
# Saved files names
save_names = (('reliefseq', 'turf'), )
for save_names_nxt, file_names, names in zip(save_names, comp_pairs,
alg_names):
# Load scores matrices.
scores_l = sio.loadmat('./raw_scores/' + file_names[0])['data']
scores_r = sio.loadmat('./raw_scores/' + file_names[1])['data']
# Compute difference.
scores = scores_l - scores_r
msk = np.logical_not(
np.apply_along_axis(lambda x: np.all(x == 0), 1, scores))
scores = scores[msk, :]
# Compute probabilities.
pleft, prope, pright = bt.hierarchical(scores, rope, rho)
with open('results_bhctt.res', 'a') as f:
f.write('{0}, {1}, {2}, {3}, {4}\n'.format(names[0], names[1], pleft,
prope, pright))
# Sample posterior and make simplex plot.
samples = bt.hierarchical_MC(scores, rope, rho, names=alg_names)
fig = bt.plot_posterior(samples, names)
plt.savefig(save_names_nxt[0] + '_' + save_names_nxt[1] + '_final.png')