def streams():
# Variables
distributions = [[0.1, 0.9], [0.2, 0.8], [0.3, 0.7], [0.4, 0.6]]
label_noises = [0.0, 0.1, 0.2, 0.3]
drift_types = ["incremental", "sudden"]
random_states = [1337, 666, 42]
# Prepare streams
streams = {}
for drift_type in drift_types:
for distribution in distributions:
for random_state in random_states:
for flip_y in label_noises:
stream = csm.StreamGenerator(
drift_type=drift_type,
distribution=distribution,
random_state=random_state,
flip_y=flip_y,
)
streams.update({str(stream): stream})
return streams
import csm
import numpy as np
from tqdm import tqdm
drift_types = ("sudden", "incremental")
ensemble_sizes = (2, 3, 4, 5, 10, 15, 20)
alphas = (0, 0.05, 0.1, 0.15, 0.2, 0.25, 0.3)
results = np.zeros((len(drift_types), len(ensemble_sizes), len(alphas), 199))
for i, drift_type in tqdm(enumerate(drift_types),
ascii=True,
total=len(drift_types)):
stream = csm.StreamGenerator(drift_type=drift_type,
distribution=[0.1, 0.9],
random_state=844,
flip_y=0.1)
for j, ensemble_size in tqdm(enumerate(ensemble_sizes),
ascii=True,
total=len(ensemble_sizes)):
for k, alpha in tqdm(enumerate(alphas), ascii=True, total=len(alphas)):
clf = csm.MDE(alpha=alpha, ensemble_size=ensemble_size)
learner = csm.TestAndTrain(stream, clf)
learner.run()
stream.reset()
results[i, j, k, :] = np.array(learner.scores)
np.save("results/experiment_1", results)
a = np.swapaxes(results_hypercube, 1, 0)
res = np.reshape(a, (5, -1))
return h.tabrow(what, res)
# Compare drift types
print("Drift types")
for e in [2, 3]:
for drift_type in drift_types:
for distribution in ldistributions:
for random_state in random_states:
for flip_y in label_noises:
stream = csm.StreamGenerator(
drift_type=drift_type,
distribution=distribution,
random_state=random_state,
flip_y=flip_y,
)
streams = [stream]
title = stream
filename = "figures/run/%s_%i" % (stream, e)
what = drift_type
tabrow = gather_and_present(title, filename, streams, what,
e)
print(stream)
print(tabrow)