def conduct(self, file=None):
for r_i, r in enumerate(self._streams_random_seeds):
if self._gradual_drift:
stream = StreamGenerator(n_chunks=self.n_chunks,
chunk_size=500,
n_drifts=5,
weights=self._proportions,
n_features=2,
n_informative=2,
n_redundant=0,
n_repeated=0,
random_state=r,
concept_sigmoid_spacing=5)
else:
stream = StreamGenerator(n_chunks=self.n_chunks,
chunk_size=500,
n_drifts=5,
weights=self._proportions,
n_features=2,
n_informative=2,
n_redundant=0,
n_repeated=0,
random_state=r)
self._evaluator.process(stream, self._ensembles)
self._scores[r_i, :, :, :] = self._evaluator.scores[:, :, :]
if file is not None:
print("Saving file " + file)
np.save(file, self._scores)
def streams(random_state):
# Variables
# distributions = [[0.95, 0.05], [0.90, 0.10], [0.85, 0.15]]
distributions = [[0.97, 0.03]]
label_noises = [
0.01,
0.03,
0.05,
]
incremental = [False, True]
ccs = [5, None]
n_drifts = 1
# Prepare streams
streams = {}
for drift_type in incremental:
for distribution in distributions:
for flip_y in label_noises:
for spacing in ccs:
stream = StreamGenerator(
incremental=drift_type,
weights=distribution,
random_state=random_state,
y_flip=flip_y,
concept_sigmoid_spacing=spacing,
n_drifts=n_drifts,
chunk_size=250,
n_chunks=200,
n_clusters_per_class=1,
n_features=8,
n_informative=8,
n_redundant=0,
n_repeated=0,
)
if spacing == None and drift_type == True:
pass
else:
streams.update({str(stream): stream})
return streams
def timestream(chunk_size):
# Variables
distributions = [[0.80, 0.20]]
label_noises = [
0.01,
]
incremental = [False]
ccs = [None]
n_drifts = 1
# Prepare streams
streams = {}
for drift_type in incremental:
for distribution in distributions:
for flip_y in label_noises:
for spacing in ccs:
stream = StreamGenerator(
incremental=drift_type,
weights=distribution,
random_state=1994,
y_flip=flip_y,
concept_sigmoid_spacing=spacing,
n_drifts=n_drifts,
chunk_size=chunk_size,
n_chunks=5,
n_clusters_per_class=1,
n_features=8,
n_informative=8,
n_redundant=0,
n_repeated=0,
)
if spacing == None and drift_type == True:
pass
else:
streams.update({str(stream): stream})
return streams
import numpy as np
from tqdm import tqdm
from strlearn.evaluators import TestThenTrain
mcargs = {
"n_classes": 2,
"n_chunks": 250,
"chunk_size": 400,
"n_features": 10,
}
streams = {
"incremental1":
StreamGenerator(**mcargs,
n_drifts=1,
concept_sigmoid_spacing=5,
incremental=True,
weights=[0.8, 0.2],
random_state=14),
"incremental2":
StreamGenerator(**mcargs,
n_drifts=1,
concept_sigmoid_spacing=5,
incremental=True,
weights=[0.8, 0.2],
random_state=67),
"incremental3":
StreamGenerator(**mcargs,
n_drifts=1,
concept_sigmoid_spacing=5,
incremental=True,
weights=[0.8, 0.2],
import numpy as np
from sklearn.naive_bayes import GaussianNB
import matplotlib.pyplot as plt
from strlearn.streams import StreamGenerator
from strlearn.evaluators import TestThenTrain
from strlearn.ensembles import SEA, OnlineBagging ,OOB, UOB
from strlearn.metrics import recall, specificity, balanced_accuracy_score
stream = StreamGenerator(n_classes=2,
n_chunks=200,
chunk_size=400,
n_features=10,
n_drifts=1,
weights=[0.8, 0.2],
random_state=95),
clfs = [
SEA(GaussianNB(), n_estimators=5),
OnlineBagging(GaussianNB(), n_estimators=5),
OOB(GaussianNB(), n_estimators=5),
UOB(GaussianNB(), n_estimators=5),
]
clf_names = [
"SEA",
"OB",
"OOB",
"UOB",
]
metrics = [
streams = {}
for i, clf in enumerate(clfs):
for j, random_state in enumerate(random_states):
for k, kurwa in enumerate(drifttype):
for l, distribution in enumerate(distributions):
for m, flip_y in enumerate(label_noises):
# for n, spacing in enumerate(css):
spacing, drift_type = kurwa
stream = StreamGenerator(
incremental=drift_type,
weights=distribution,
random_state=random_state,
y_flip=flip_y,
concept_sigmoid_spacing=spacing,
n_drifts=1,
n_chunks=200,
chunk_size=250,
n_clusters_per_class = 1,
n_features = 8,
n_informative= 8,
n_redundant= 0,
n_repeated = 0,
)
if spacing == None and drift_type == True:
pass
else:
results = np.load(
"results/experiment1_%s/%s.npy" % (clf, stream)
)
scores[i, j, k, l, m] = results
scores_metrics = scores
#ewaluator z odpowiednia metryka
evaluator = TestThenTrain(metrics=accuracy_score)
#przygotowanie pliku z wynikami
f_sudden = open("wynikidryfnagly.csv", "a")
f_gradual = open("wynikidryfgradualny.csv", "a")
f_incremental = open("wynikidryfinkrementalny", "a")
#usrednianie po n_chunks i wypisywanie wynikow, osobne pliki dla kazdego dryfu
#wyniki sa zapisywane w jednej kolumnie ->
#ciag: srednia_str1_clf1, srednia_str1_clf2, srednia_str1_clf3, srednia_str2_clf1, ...
for rnd_st in range(10, 110,
10): #rnd_st przechowuje aktualna wartosc random_state
#wypisywanie wynikow dla dryfu naglego dla wszystkich klasyfikatorow
str_sudden = StreamGenerator(n_drifts=1, random_state=rnd_st)
evaluator.process(str_sudden, clfs)
array2d = evaluator.scores.reshape(249, 3)
resultsmean = np.mean(array2d, axis=0)
np.savetxt(f_sudden, resultsmean, delimiter=",", fmt='%0.3f')
#wypisywanie wynikow dla dryfu gradualnego dla wszystkich klasyfikatorow
str_gradual = StreamGenerator(n_drifts=1,
concept_sigmoid_spacing=5,
random_state=rnd_st)
evaluator.process(str_gradual, clfs)
array2d = evaluator.scores.reshape(249, 3)
resultsmean = np.mean(array2d, axis=0)
np.savetxt(f_gradual, resultsmean, delimiter=",", fmt='%0.3f')
#wypisywanie wynikow dla dryfu inkremetalnego dla wszystkich klasyfikatorow
"n_chunks": 100,
"chunk_size": 500,
"random_state": 5,
"n_features": 2,
"n_informative": 2,
"n_redundant": 0,
"n_repeated": 0,
"n_features": 2,
"n_clusters_per_class": 1,
}
streams = {
"9_first":
StreamGenerator(n_drifts=3,
concept_sigmoid_spacing=5,
reocurring=True,
incremental=True,
weights=(2, 5, 0.9),
**mcargs),
"9_second":
StreamGenerator(n_drifts=1, concept_sigmoid_spacing=5, **mcargs),
}
for stream_name in streams:
print(stream_name)
stream = streams[stream_name]
checkpoints = np.linspace(0, stream.n_chunks - 1, 8).astype(int)
fig = plt.figure(constrained_layout=True, figsize=(8, 6))
gs = GridSpec(5, len(checkpoints), figure=fig)
mcargs = {
"n_classes": 3,
"n_chunks": 100,
"chunk_size": 500,
"random_state": 105,
"n_features": 2,
"n_informative": 2,
"n_redundant": 0,
"n_repeated": 0,
"n_features": 2,
"n_clusters_per_class": 1,
}
streams = {
"0_stationary": StreamGenerator(**mcargs),
"1_sudden": StreamGenerator(n_drifts=1, **mcargs),
"2_gradual": StreamGenerator(n_drifts=1, concept_sigmoid_spacing=5, **mcargs),
"3_incremental": StreamGenerator(
n_drifts=1, concept_sigmoid_spacing=5, incremental=True, **mcargs
),
"4_reocurring": StreamGenerator(
n_drifts=2, concept_sigmoid_spacing=5, reocurring=True, **mcargs
),
"5_nonreocurring": StreamGenerator(
n_drifts=2, concept_sigmoid_spacing=5, reocurring=False, **mcargs
),
}
mcargs.update({"n_classes": 2, "random_state": 5})
streams.update(
from sklearn.naive_bayes import GaussianNB
from strlearn.metrics import recall, specificity, balanced_accuracy_score
import numpy as np
from tqdm import tqdm
from strlearn.evaluators import TestThenTrain
mcargs = {
"n_classes": 2,
"n_chunks": 250,
"chunk_size": 400,
"n_features": 10,
}
streams = {
"sudden1":
StreamGenerator(**mcargs, n_drifts=1, weights=[0.9, 0.1], random_state=14),
"sudden2":
StreamGenerator(**mcargs, n_drifts=1, weights=[0.9, 0.1], random_state=67),
"sudden3":
StreamGenerator(**mcargs, n_drifts=1, weights=[0.9, 0.1], random_state=95),
"sudden4":
StreamGenerator(**mcargs, n_drifts=1, weights=[0.9, 0.1],
random_state=234),
"sudden5":
StreamGenerator(**mcargs, n_drifts=1, weights=[0.9, 0.1],
random_state=876),
"sudden6":
StreamGenerator(**mcargs,
n_drifts=1,
weights=[0.9, 0.1],
random_state=1410),
""" =================================================================== The simplest experiment example with one classifier and two metrics =================================================================== Lorem impsum. """ from sklearn.naive_bayes import GaussianNB clf = GaussianNB() from strlearn.streams import StreamGenerator stream = StreamGenerator(n_chunks=30, n_drifts=1) from sklearn.metrics import accuracy_score from strlearn.metrics import recall metrics = [accuracy_score, recall] from strlearn.evaluators import TestThenTrain evaluator = TestThenTrain(metrics) evaluator.process(stream, clf) print(evaluator.scores.shape) evaluator.scores
number_of_classifiers=5)
sea = SEA(base_estimator=StratifiedBagging(base_estimator=GaussianNB(),
random_state=42),
n_estimators=5,
metric=roc_auc_score)
ros_knorau2 = SEA(base_estimator=StratifiedBagging(base_estimator=GaussianNB(),
random_state=42,
oversampler="ROS"),
oversampled="ROS",
des="KNORAU2")
stream = StreamGenerator(n_chunks=250,
chunk_size=200,
random_state=1410,
n_drifts=1,
weights=[0.9, 0.1])
eval = TestThenTrain(metrics=(geometric_mean_score_1))
eval.process(stream, [kmc])
value = np.squeeze(eval.scores[0])
val = gaussian_filter1d(value, sigma=3, mode="nearest")
plt.plot(val)
plt.savefig("zzz")
# print(kmc.new_auc)
# print(kmc.auc_array)
# print(kmc.worst)
# print(eval.scores)
"n_features": 2,
"n_informative": 2,
"n_redundant": 0,
"n_repeated": 0,
}
#stream = sl.streams.StreamGenerator(**concept_kwargs)
#stream1 = sl.streams.StreamGenerator(**concept_kwargs)
cm = LinearSegmentedColormap.from_list("lokomotiv",
colors=[(0.3, 0.7, 0.3),
(0.7, 0.3, 0.3)])
chunks_plotted = np.linspace(0, n_chunks - 1, 8).astype(int)
stream = StreamGenerator(n_chunks=100, n_drifts=1)
stream1 = StreamGenerator(n_chunks=100, n_drifts=1)
plot_stream(stream1, "stationary", "Stationary stream")
from sklearn.metrics import accuracy_score
from strlearn.metrics import precision
metrics = [accuracy_score, precision]
from strlearn.evaluators import TestThenTrain
evaluator = TestThenTrain(metrics)
evaluator.process(stream, clf)
plt.figure(figsize=(6, 3))
mcargs = {
"n_classes": 3,
"n_chunks": 100,
"chunk_size": 500,
"random_state": 105,
"n_features": 2,
"n_informative": 2,
"n_redundant": 0,
"n_repeated": 0,
"n_features": 2,
"n_clusters_per_class": 1,
}
streams = {
"0_stationary":
StreamGenerator(**mcargs),
"1_sudden":
StreamGenerator(n_drifts=1, **mcargs),
"2_gradual":
StreamGenerator(n_drifts=1, concept_sigmoid_spacing=5, **mcargs),
"3_incremental":
StreamGenerator(n_drifts=1,
concept_sigmoid_spacing=5,
incremental=True,
**mcargs),
"4_reocurring":
StreamGenerator(n_drifts=2,
concept_sigmoid_spacing=5,
reocurring=True,
**mcargs),
"5_nonreocurring":
import numpy as np
from sklearn.naive_bayes import GaussianNB
import matplotlib.pyplot as plt
from strlearn.streams import StreamGenerator
from strlearn.evaluators import TestThenTrain
from strlearn.ensembles import SEA, OnlineBagging ,OOB, UOB
from strlearn.metrics import recall, specificity, balanced_accuracy_score
stream = StreamGenerator(n_classes=2,
n_chunks=200,
chunk_size=400,
n_features=10,
concept_sigmoid_spacing=5,
incremental=True,
n_drifts=1,
weights=[0.8, 0.2],
random_state=95)
clfs = [
SEA(GaussianNB(), n_estimators=5),
OnlineBagging(GaussianNB(), n_estimators=5),
OOB(GaussianNB(), n_estimators=5),
UOB(GaussianNB(), n_estimators=5),
]
clf_names = [
"SEA",
"OB",
"OOB",
"UOB",
]