def demo(instances=2000):
""" _test_comparison_prequential
This demo will test a prequential evaluation when more than one learner is
passed, which makes it a comparison task.
Parameters
----------
instances: int
The evaluation's maximum number of instances.
"""
# Stream setup
stream = FileStream("../data/datasets/covtype.csv", -1, 1)
# stream = SEAGenerator(classification_function=2, sample_seed=53432, balance_classes=False)
stream.prepare_for_use()
# Setup the classifier
clf = SGDClassifier()
# classifier = KNNAdwin(n_neighbors=8, max_window_size=2000,leaf_size=40, categorical_list=None)
# classifier = OzaBaggingAdwin(base_estimator=KNN(n_neighbors=8, max_window_size=2000, leaf_size=30, categorical_list=None))
clf_one = KNNAdwin(n_neighbors=8, max_window_size=1000, leaf_size=30)
# clf_two = KNN(n_neighbors=8, max_window_size=1000, leaf_size=30)
# clf_two = LeverageBagging(base_estimator=KNN(), n_estimators=2)
t_one = OneHotToCategorical([[10, 11, 12, 13],
[
14, 15, 16, 17, 18, 19, 20, 21, 22, 23,
24, 25, 26, 27, 28, 29, 30, 31, 32, 33,
34, 35, 36, 37, 38, 39, 40, 41, 42, 43,
44, 45, 46, 47, 48, 49, 50, 51, 52, 53
]])
# t_two = OneHotToCategorical([[10, 11, 12, 13],
# [14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35,
# 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53]])
pipe_one = Pipeline([('one_hot_to_categorical', t_one), ('KNN', clf_one)])
# pipe_two = Pipeline([('one_hot_to_categorical', t_two), ('KNN', clf_two)])
classifier = [clf, pipe_one]
# classifier = SGDRegressor()
# classifier = PerceptronMask()
# Setup the pipeline
# pipe = Pipeline([('Classifier', classifier)])
# Setup the evaluator
evaluator = EvaluatePrequential(
pretrain_size=2000,
output_file='test_comparison_prequential.csv',
max_samples=instances,
batch_size=1,
n_wait=200,
max_time=1000,
show_plot=True,
metrics=['performance', 'kappa_t'])
# Evaluate
evaluator.evaluate(stream=stream, model=classifier)
def demo(output_file=None, instances=50000):
""" _test_sam_knn_prequential
This demo shows how to produce a prequential evaluation.
The first thing needed is a stream. For this case we use a file stream
which gets its samples from the movingSquares.csv file, inside the datasets
folder.
Then we need to setup a classifier, which in this case is an instance
of scikit-multiflow's SAMKNN. Then, optionally we create a
pipeline structure, initialized on that classifier.
The evaluation is then run.
Parameters
----------
output_file: string
The name of the csv output file
instances: int
The evaluation's max number of instances
"""
# Setup the File Stream
stream = FileStream("../data/datasets/movingSquares.csv", -1, 1)
# stream = WaveformGenerator()
stream.prepare_for_use()
# Setup the classifier
# classifier = SGDClassifier()
# classifier = KNNAdwin(n_neighbors=8, max_window_size=2000,leaf_size=40, categorical_list=None)
# classifier = OzaBaggingAdwin(base_estimator=KNN(n_neighbors=8, max_window_size=2000, leaf_size=30, categorical_list=None))
classifier = SAMKNN(n_neighbors=5,
weighting='distance',
max_window_size=1000,
stm_size_option='maxACCApprox',
use_ltm=False)
# classifier = SGDRegressor()
# classifier = PerceptronMask()
# Setup the pipeline
# pipe = Pipeline([('Classifier', classifier)])
# Setup the evaluator
evaluator = EvaluatePrequential(pretrain_size=0,
max_samples=instances,
batch_size=1,
n_wait=100,
max_time=1000,
output_file=output_file,
show_plot=True,
metrics=['performance'])
# Evaluate
evaluator.evaluate(stream=stream, model=classifier)
def demo():
""" _test_knn
This demo tests the KNNClassifier on a file stream, which gives
instances coming from a SEA generator.
The test computes the performance of the KNNClassifier as well as
the time to create the structure and classify max_samples (5000 by
default) instances.
"""
stream = FileStream('../data/datasets/sea_big.csv', -1, 1)
stream.prepare_for_use()
train = 200
X, y = stream.next_sample(train)
# t = OneHotToCategorical([[10, 11, 12, 13],
# [14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35,
# 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53]])
# t2 = OneHotToCategorical([[10, 11, 12, 13],
# [14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35,
# 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53]])
start = timer()
knn = KNNClassifier(n_neighbors=8, max_window_size=2000, leaf_size=40)
# pipe = Pipeline([('one_hot_to_categorical', t), ('KNNClassifier', knn)])
# compare = KNeighborsClassifier(n_neighbors=8, algorithm='kd_tree', leaf_size=40, metric='euclidean')
# pipe2 = Pipeline([('one_hot_to_categorical', t2), ('KNNClassifier', compare)])
# pipe.fit(X, y)
# pipe2.fit(X, y)
knn.partial_fit(X, y)
# compare.fit(X, y)
n_samples = 0
max_samples = 5000
my_corrects = 0
# compare_corrects = 0
while n_samples < max_samples:
X, y = stream.next_sample()
# my_pred = pipe.predict(X)
my_pred = knn.predict(X)
# compare_pred = pipe2.predict(X)
# compare_pred = compare.predict(X)
if y[0] == my_pred[0]:
my_corrects += 1
# if y[0] == compare_pred[0]:
# compare_corrects += 1
n_samples += 1
end = timer()
print('Evaluation time: ' + str(end - start))
print(str(n_samples) + ' samples analyzed.')
print('My performance: ' + str(my_corrects / n_samples))
def demo_parameterized(h, filename="covtype.csv", show_plot=True, model_names=None):
# Setup Stream
stream = FileStream("../data/datasets/" + filename)
stream.prepare_for_use()
# For each classifier, e...
pretrain = 100
evaluator = EvaluatePrequential(pretrain_size=pretrain, output_file='test_parametrized.csv', max_samples=10000,
batch_size=1, n_wait=500, show_plot=show_plot)
evaluator.evaluate(stream=stream, model=h, model_names=model_names)
def demo():
""" _test_mol
This demo tests the MOL learner on a file stream, which reads from
the music.csv file.
The test computes the performance of the MOL learner as well as
the time to create the structure and classify all the samples in
the file.
"""
# Setup logging
logging.basicConfig(format='%(message)s', level=logging.INFO)
# Setup the file stream
stream = FileStream("../data/datasets/music.csv", 0, 6)
stream.prepare_for_use()
# Setup the classifier, by default it uses Logistic Regression
# classifier = MultiOutputLearner()
# classifier = MultiOutputLearner(base_estimator=SGDClassifier(n_iter=100))
classifier = MultiOutputLearner(base_estimator=Perceptron())
# Setup the pipeline
pipe = Pipeline([('classifier', classifier)])
pretrain_size = 150
logging.info('Pre training on %s samples', str(pretrain_size))
logging.info('Total %s samples', str(stream.n_samples))
X, y = stream.next_sample(pretrain_size)
# classifier.fit(X, y)
classes = stream.target_values
classes_flat = list(set([item for sublist in classes for item in sublist]))
pipe.partial_fit(X, y, classes=classes_flat)
count = 0
true_labels = []
predicts = []
init_time = timer()
logging.info('Evaluating...')
while stream.has_more_samples():
X, y = stream.next_sample()
# p = classifier.predict(X)
p = pipe.predict(X)
predicts.extend(p)
true_labels.extend(y)
count += 1
perf = hamming_score(true_labels, predicts)
logging.info('Evaluation time: %s s', str(timer() - init_time))
logging.info('Total samples analyzed: %s', str(count))
logging.info('The classifier\'s static Hamming score : %0.3f' % perf)
def demo(output_file=None, instances=50000):
""" _test_sam_knn_prequential
This demo shows how to produce a prequential evaluation.
The first thing needed is a stream. For this case we use a file stream
which gets its samples from the moving_squares.csv file, inside the datasets
folder.
Then we need to setup a classifier, which in this case is an instance
of scikit-multiflow's SAMKNNClassifier. Then, optionally we create a
pipeline structure, initialized on that classifier.
The evaluation is then run.
Parameters
----------
output_file: string
The name of the csv output file
instances: int
The evaluation's max number of instances
"""
# Setup the File Stream
stream = FileStream("../data/datasets/moving_squares.csv")
# stream = WaveformGenerator()
stream.prepare_for_use()
# Setup the classifier
classifier = SAMKNNClassifier(n_neighbors=5,
weighting='distance',
max_window_size=1000,
stm_size_option='maxACCApprox',
use_ltm=False)
# Setup the evaluator
evaluator = EvaluatePrequential(pretrain_size=0,
max_samples=instances,
batch_size=1,
n_wait=100,
max_time=1000,
output_file=output_file,
show_plot=True)
# Evaluate
evaluator.evaluate(stream=stream, model=classifier)
def demo():
# The classifier we will use (other options: SAMKNNClassifier, LeverageBaggingClassifier, SGD)
h = HoeffdingTreeClassifier()
# Setup Stream
stream = FileStream("../data/datasets/sea_stream.csv")
stream.prepare_for_use()
pretrain = 100
evaluator = EvaluatePrequential(pretrain_size=pretrain,
output_file='test_filestream.csv',
max_samples=10000,
batch_size=1,
n_wait=1000,
show_plot=True)
evaluator.evaluate(stream=stream, model=h)
def demo():
""" _test_streams
This demo tests if the streams are correctly generating samples.
:return:
"""
stream = FileStream('../data/datasets/covtype.csv', -1, 1)
stream.prepare_for_use()
rbf_drift = RandomRBFGeneratorDrift(change_speed=41.00,
n_centroids=50,
model_seed=32523423,
instance_seed=5435,
n_classes=2,
n_features=10,
num_drift_centroids=50)
rbf_drift.prepare_for_use()
sea = SEAGenerator()
print('1 instance:\n')
X, y = stream.next_sample()
print(X)
print(y)
X, y = sea.next_sample()
print(X)
print(y)
print('\n\n10 instances:\n')
X, y = stream.next_sample(10)
print(X)
print(y)
X, y = sea.next_sample(10)
print(X)
print(y)
def demo():
""" _test_knn_adwin
This demo tests the KNNAdwin classifier on a file stream, which gives
instances coming from a SEA generator.
The test computes the performance of the KNNAdwin classifier as well as
the time to create the structure and classify max_samples (10000 by
default) instances.
"""
start = timer()
logging.basicConfig(format='%(message)s', level=logging.INFO)
# warnings.filterwarnings("ignore", ".*Passing 1d.*")
stream = FileStream('../data/datasets/sea_big.csv', -1, 1)
# stream = RandomRBFGeneratorDrift(change_speed=41.00, n_centroids=50, model_random_state=32523423,
# sample_seed=5435, n_classes=2, num_att=10, num_drift_centroids=50)
stream.prepare_for_use()
t = OneHotToCategorical([[10, 11, 12, 13],
[
14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24,
25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35,
36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46,
47, 48, 49, 50, 51, 52, 53
]])
t2 = OneHotToCategorical([[10, 11, 12, 13],
[
14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24,
25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35,
36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46,
47, 48, 49, 50, 51, 52, 53
]])
# knn = KNN(n_neighbors=8, max_window_size=2000, leaf_size=40)
knn = KNNAdwin(n_neighbors=8, leaf_size=40, max_window_size=2000)
# pipe = Pipeline([('one_hot_to_categorical', t), ('KNN', knn)])
compare = KNeighborsClassifier(n_neighbors=8,
algorithm='kd_tree',
leaf_size=40,
metric='euclidean')
# pipe2 = Pipeline([('one_hot_to_categorical', t2), ('KNN', compare)])
first = True
train = 200
if train > 0:
X, y = stream.next_sample(train)
# pipe.partial_fit(X, y, classes=stream.target_values)
# pipe.partial_fit(X, y, classes=stream.target_values)
# pipe2.fit(X, y)
knn.partial_fit(X, y, classes=stream.target_values)
compare.fit(X, y)
first = False
n_samples = 0
max_samples = 10000
my_corrects = 0
compare_corrects = 0
while n_samples < max_samples:
if n_samples % (max_samples / 20) == 0:
logging.info('%s%%', str((n_samples // (max_samples / 20) * 5)))
X, y = stream.next_sample()
# my_pred = pipe.predict(X)
my_pred = knn.predict(X)
# my_pred = [1]
if first:
# pipe.partial_fit(X, y, classes=stream.target_values)
# pipe.partial_fit(X, y, classes=stream.target_values)
knn.partial_fit(X, y, classes=stream.target_values)
first = False
else:
# pipe.partial_fit(X, y)
knn.partial_fit(X, y)
# compare_pred = pipe2.predict(X)
compare_pred = compare.predict(X)
if y[0] == my_pred[0]:
my_corrects += 1
if y[0] == compare_pred[0]:
compare_corrects += 1
n_samples += 1
end = timer()
print('Evaluation time: ' + str(end - start))
print(str(n_samples) + ' samples analyzed.')
print('My performance: ' + str(my_corrects / n_samples))
print('Compare performance: ' + str(compare_corrects / n_samples))
from skmultiflow.data import FileStream
from skmultiflow.lazy.knn import KNN
from skmultiflow.evaluation import EvaluatePrequential
n_neighbors = 8
max_window_size = 2000
leaf_size = 30
n_estimators = 30
show_plot = True
pretrain_size = 100
max_samples = 7000
metrics = ['accuracy']
stream = FileStream('data/stream1.csv')
stream.prepare_for_use()
mdl = KNN(n_neighbors=n_neighbors,
max_window_size=max_window_size,
leaf_size=leaf_size)
evaluator = EvaluatePrequential(show_plot=show_plot,
pretrain_size=pretrain_size,
max_samples=max_samples,
metrics=metrics)
evaluator.evaluate(stream=stream, model=mdl)
def test_evaluate_and_adapt_trees():
expected_accuracies = [
0.86, 0.876, 0.914, 0.858, 0.77, 0.894, 0.876, 0.91, 0.898, 0.884,
0.804, 0.808
]
expected_trees = [30, 60, 30, 60, 30]
# Load the meta-model
dictMeta = {
0.0: 60,
0.1: 30,
0.2: 30,
0.3: 30,
0.4: 60,
0.5: 70,
0.6: 60,
0.7: 30,
0.8: 30,
0.9: 30
} # dict = {'pourc redund feat':best nb tree}
n_trees = 10
n_samples_max = 6000
n_samples_meas = 500
stream = FileStream('./recurrent-data/real-world/elec.csv')
stream.prepare_for_use()
# Evaluate model (with adaptation or not)
arf = AdaptiveRandomForest(n_estimators=n_trees,
lambda_value=6,
grace_period=10,
split_confidence=0.1,
tie_threshold=0.005,
warning_detection_method=ADWIN(delta=0.01),
drift_detection_method=ADWIN(delta=0.001),
random_state=0)
modelsList = [arf]
modelsNames = ['ARF']
evaluator = EvaluatePrequentialAndAdaptTreesARF(
metrics=['accuracy', 'kappa', 'running_time', 'ram_hours'],
show_plot=False,
n_wait=n_samples_meas,
pretrain_size=200,
max_samples=n_samples_max,
output_file=None,
metaKB=dictMeta)
# Run evaluation
model, acc, n_trees = evaluator.evaluate(stream=stream,
model=modelsList,
model_names=modelsNames)
assert np.alltrue(acc[0] == expected_accuracies)
assert np.alltrue(n_trees[0] == expected_trees)
def demo():
""" _test_kdtree_compare
This demo compares creation and query speed for different kd tree
implementations. They are fed with instances from the covtype dataset.
Three kd tree implementations are compared: SciPy's KDTree, NumPy's
KDTree and scikit-multiflow's KDTree. For each of them the demo will
time the construction of the tree on 1000 instances, and then measure
the time to query 100 instances. The results are displayed in the
terminal.
"""
warnings.filterwarnings("ignore", ".*Passing 1d.*")
stream = FileStream('../data/datasets/covtype.csv', -1, 1)
stream.prepare_for_use()
filter = OneHotToCategorical([[10, 11, 12, 13],
[14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33,
34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53]])
X, y = stream.next_sample(1000)
X = filter.transform(X)
# print(X)
X_find, y = stream.next_sample(100)
X_find = filter.transform(X_find)
print(X_find[4])
# Normal kdtree
start = timer()
scipy = spatial.KDTree(X, leafsize=40)
end = timer()
print("\nScipy KDTree construction time: " + str(end-start))
start = timer()
for i in range(10):
ind = scipy.query(X_find[i], 8)
# print(ind)
end = timer()
print("Scipy KDTree query time: " + str(end - start))
del scipy
# Fast kdtree
start = timer()
opt = KDTree(X, metric='euclidean', return_distance=True)
end = timer()
print("\nOptimal KDTree construction time: " + str(end-start))
start = timer()
for i in range(100):
ind, dist = opt.query(X_find[i], 8)
# print(ind)
# print(dist)
end = timer()
print("Optimal KDTree query time: " + str(end - start))
del opt
# Sklearn kdtree
start = timer()
sk = ng.KDTree(X, metric='euclidean')
end = timer()
print("\nSklearn KDTree construction time: " + str(end-start))
start = timer()
for i in range(100):
ind, dist = sk.query(np.asarray(X_find[i]).reshape(1, -1), 8, return_distance=True)
# print(ind)
# print(dist)
end = timer()
print("Sklearn KDTree query time: " + str(end - start) + "\n")
del sk