def demo():
""" _test_filters
This demo test the MissingValuesCleaner filter. The transform is set
to clean any value equal to -47, replacing it with the median value
of the last 10 samples, or less if there aren't 10 samples available.
The output will be the 10 instances used in the transform. The first
9 are kept untouched, as they don't have any feature value of -47. The
last samples has its first feature value equal to -47, so it's replaced
by the median of the 9 first samples.
"""
stream = FileStream("https://raw.githubusercontent.com/scikit-multiflow/streaming-datasets/"
"master/covtype.csv")
filter = MissingValuesCleaner(-47, 'median', 10)
X, y = stream.next_sample(10)
X[9, 0] = -47
for i in range(10):
temp = filter.partial_fit_transform([X[i].tolist()])
print(temp)
def get_file_stream2(self, path):
from skmultiflow.data.file_stream import FileStream
stream = FileStream(path, n_targets=1, target_idx=-1)
stream2 = FileStream(path, n_targets=1, target_idx=-1)
stream3 = FileStream(path, n_targets=1, target_idx=-1)
return stream, stream2, stream3
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
# opt = FileOption("FILE", "OPT_NAME", "../datasets/covtype.csv", "CSV", False)
opt = FileOption("FILE", "OPT_NAME", "../datasets/movingSquares.csv",
"CSV", False)
stream = FileStream(opt, -1, 1)
# stream = WaveformGenerator()
stream.prepare_for_use()
# Setup the classifier
# classifier = SGDClassifier()
# classifier = KNNAdwin(k=8, max_window_size=2000,leaf_size=40, categorical_list=None)
# classifier = OzaBaggingAdwin(h=KNN(k=8, max_window_size=2000, leaf_size=30, categorical_list=None))
classifier = SAMKNN(n_neighbors=5,
knnWeights='distance',
maxSize=1000,
STMSizeAdaption='maxACCApprox',
useLTM=False)
# classifier = SGDRegressor()
# classifier = PerceptronMask()
# Setup the pipeline
#pipe = Pipeline([('Classifier', classifier)])
# Setup the evaluator
eval = EvaluatePrequential(pretrain_size=0,
max_instances=instances,
batch_size=1,
n_wait=100,
max_time=1000,
output_file=output_file,
task_type='classification',
show_plot=True,
plot_options=['performance'])
# Evaluate
eval.eval(stream=stream, classifier=classifier)
def demo():
""" _test_filters
This demo test the MissingValuesCleaner filter. The transform is set
to clean any value equal to -47, replacing it with the median value
of the last 10 samples, or less if there aren't 10 samples available.
The output will be the 10 instances used in the transform. The first
9 are kept untouched, as they don't have any feature value of -47. The
last samples has its first feature value equal to -47, so it's replaced
by the median of the 9 first samples.
"""
opt = FileOption('FILE', 'OPT_NAME', '../datasets/covtype.csv', 'csv', False)
stream = FileStream(opt, -1, 1)
stream.prepare_for_use()
filter = MissingValuesCleaner(-47, 'median', 10)
X, y = stream.next_instance(10)
X[9, 0] = -47
for i in range(10):
temp = filter.partial_fit_transform([X[i].tolist()])
print(temp)
def test_sam_knn(package_path):
test_file = os.path.join(package_path, 'src/skmultiflow/data/datasets/sea_big.csv')
stream = FileStream(test_file)
stream.prepare_for_use()
hyperParams = {'maxSize': 1000, 'nNeighbours': 5, 'knnWeights': 'distance', 'STMSizeAdaption': 'maxACCApprox',
'useLTM': False}
learner = SAMKNN(n_neighbors=hyperParams['nNeighbours'], max_window_size=hyperParams['maxSize'],
weighting=hyperParams['knnWeights'],
stm_size_option=hyperParams['STMSizeAdaption'], use_ltm=hyperParams['useLTM'])
cnt = 0
max_samples = 5000
predictions = array('d')
wait_samples = 100
while cnt < max_samples:
X, y = stream.next_sample()
# Test every n samples
if (cnt % wait_samples == 0) and (cnt != 0):
predictions.append(learner.predict(X)[0])
learner.partial_fit(X, y)
cnt += 1
expected_predictions = array('d', [1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0,
0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0,
1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0,
0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0,
0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0])
assert np.alltrue(predictions == expected_predictions)
def test_KNN_adwin(test_path, package_path):
test_file = os.path.join(package_path,
'src/skmultiflow/data/datasets/sea_big.csv')
stream = FileStream(test_file, -1, 1)
stream.prepare_for_use()
learner = KNNAdwin(n_neighbors=8, leaf_size=40, max_window_size=2000)
cnt = 0
max_samples = 5000
predictions = []
correct_predictions = 0
wait_samples = 100
while cnt < max_samples:
X, y = stream.next_sample()
# Test every n samples
if (cnt % wait_samples == 0) and (cnt != 0):
predictions.append(learner.predict(X)[0])
if y[0] == predictions[-1]:
correct_predictions += 1
learner.partial_fit(X, y)
cnt += 1
performance = correct_predictions / len(predictions)
expected_predictions = [
1, 0, 0, 1, 0, 0, 1, 1, 1, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1, 1, 0, 0, 1,
1, 1, 1, 1, 0, 1, 0, 1, 0, 1, 1, 0, 1, 0, 1, 0, 1, 1, 0, 1, 0, 1, 1, 1,
1
]
expected_correct_predictions = 40
expected_performance = 0.8163265306122449
assert np.alltrue(predictions == expected_predictions)
assert np.isclose(expected_performance, performance)
assert correct_predictions == expected_correct_predictions
def train_tree(csv_path, tree):
print("Training the tree")
stream = FileStream(csv_path)
accuracy = 0
n_samples = 0
correct_cnt = 0
t0 = time.time()
while stream.has_more_samples():
X, y = stream.next_sample()
y_pred = tree.predict(X)
if y[0] == y_pred[0]:
correct_cnt += 1
tree = tree.partial_fit(X, y)
n_samples += 1
t1 = time.time()
total = t1 - t0
accuracy = 100.0 * correct_cnt / n_samples
print("Training data instances: ", n_samples)
print("Tree trained on ", n_samples, " instances & has ", accuracy,
"% accuracy.")
print("Training tree completed in ", total, " (s)")
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("../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(k=8, max_window_size=2000,leaf_size=40, categorical_list=None)
# classifier = OzaBaggingAdwin(h=KNN(k=8, max_window_size=2000, leaf_size=30, categorical_list=None))
clf_one = KNNAdwin(k=8, max_window_size=1000, leaf_size=30)
# clf_two = KNN(k=8, max_window_size=1000, leaf_size=30)
# clf_two = LeverageBagging(h=KNN(), ensemble_length=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='teste.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_parameterized(h, dset="sea_stream.csv", show_plot=True):
# Setup Stream
opt = FileOption("FILE", "OPT_NAME", "../datasets/"+dset, "CSV", False)
stream = FileStream(opt, -1, 1)
stream.prepare_for_use()
# For each classifier, e...
T_init = 100
eval = EvaluatePrequential(pretrain_size=T_init, output_file='output.csv', max_instances=10000, batch_size=1, n_wait=1000, task_type='classification', show_plot=show_plot, plot_options=['performance'])
eval.eval(stream=stream, classifier=h)
def demo(output_file=None, instances=40000):
""" _test_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 sea_big.csv file, inside the datasets
folder.
Then we need to setup a classifier, which in this case is an instance
of sklearn's PassiveAggressiveClassifier. 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/sea_big.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, nominal_attributes=None)
# classifier = OzaBaggingAdwin(base_estimator=KNN(n_neighbors=8, max_window_size=2000, leaf_size=30, categorical_list=None))
classifier = PassiveAggressiveClassifier()
# classifier = SGDRegressor()
# classifier = PerceptronMask()
# Setup the pipeline
pipe = Pipeline([('Classifier', classifier)])
# Setup the evaluator
evaluator = EvaluatePrequential(
pretrain_size=200,
max_samples=instances,
batch_size=1,
n_wait=100,
max_time=1000,
output_file=output_file,
show_plot=True,
metrics=['kappa', 'kappa_t', 'performance'])
# Evaluate
evaluator.evaluate(stream=stream, model=pipe)
def demo_parameterized(h, filename="covtype.csv", show_plot=True):
# Setup Stream
stream = FileStream("../datasets/" + filename, -1, 1)
stream.prepare_for_use()
# For each classifier, e...
pretrain = 100
evaluator = EvaluatePrequential(pretrain_size=pretrain,
output_file='output.csv',
max_samples=10000,
batch_size=1,
n_wait=1000,
show_plot=show_plot,
metrics=['performance'])
evaluator.evaluate(stream=stream, model=h)
def demo():
# The classifier we will use (other options: SAMKNN, LeverageBagging, SGD)
h = HoeffdingTree()
# Setup Stream
stream = FileStream("../datasets/sea_stream.csv", -1, 1)
stream.prepare_for_use()
pretrain = 100
evaluator = EvaluatePrequential(pretrain_size=pretrain,
output_file='output.csv',
max_samples=10000,
batch_size=1,
n_wait=1000,
show_plot=True,
metrics=['performance'])
evaluator.evaluate(stream=stream, model=h)
def test_sam_knn_coverage(package_path):
test_file = os.path.join(package_path,
'src/skmultiflow/data/datasets/sea_big.csv')
stream = FileStream(test_file)
stream.prepare_for_use()
hyperParams = {
'maxSize': 50,
'n_neighbors': 3,
'weighting': 'uniform',
'stm_size_option': 'maxACC',
'min_stm_size': 10,
'useLTM': True
}
learner = SAMKNN(n_neighbors=hyperParams['n_neighbors'],
max_window_size=hyperParams['maxSize'],
weighting=hyperParams['weighting'],
stm_size_option=hyperParams['stm_size_option'],
min_stm_size=hyperParams['min_stm_size'],
use_ltm=hyperParams['useLTM'])
cnt = 0
max_samples = 1000
predictions = array('i')
wait_samples = 20
while cnt < max_samples:
X, y = stream.next_sample()
# Test every n samples
if (cnt % wait_samples == 0) and (cnt != 0):
predictions.append(learner.predict(X)[0])
learner.partial_fit(X, y)
cnt += 1
expected_predictions = array('i', [
1, 1, 1, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0,
0, 1, 0, 1, 0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 1, 1,
1
])
assert np.alltrue(predictions == expected_predictions)
def demo():
# The classifier we will use (other options: SAMKNN, LeverageBagging, SGD)
h = HoeffdingTree()
# Setup Stream
opt = FileOption("FILE", "OPT_NAME", "../datasets/sea_stream.csv", "CSV",
False)
stream = FileStream(opt, -1, 1)
stream.prepare_for_use()
T_init = 100
eval = EvaluatePrequential(pretrain_size=T_init,
output_file='output.csv',
max_instances=10000,
batch_size=1,
n_wait=1000,
task_type='classification',
show_plot=True,
plot_options=['performance'])
eval.eval(stream=stream, classifier=h)
def test_tree(csv_path, tree):
print("Testing the tree")
stream = FileStream(csv_path)
n_samples = 0
correct_cnt = 0
t2 = time.time()
y_true_all = list()
y_pred_all = list()
while stream.has_more_samples():
X, y = stream.next_sample()
y_pred = tree.predict(X)
if y[0] == y_pred[0]:
correct_cnt += 1
tree = tree.partial_fit(X, y)
n_samples += 1
y_true_all.append(y[0])
y_pred_all.append(y_pred[0])
t3 = time.time()
total = t3-t2
accuracy = 100.0 * correct_cnt / n_samples
fscore = f1_score(y_true_all, y_pred_all, average='binary')
gm = geometric_mean_score(y_true_all, y_pred_all, average='binary')
print("Test data instances: ", n_samples)
print("Tree tested on ", n_samples, " instances & has ", accuracy, "% accuracy.")
print("Tree has F-score: %.3f" % fscore)
print("Tree has GM: %.3f" % gm)
print("Testing tree completed in ", total, " (s)")
return round(fscore,3), round(gm,3)
def validate(self):
# self.load_merged()
y_true = self.validation['R1'].tolist()
models_train = [
self.model_1_train, self.model_2_train, self.model_3_train
]
models_with_annotations = [
self.model_1_with_annotations, self.model_2_with_annotations,
self.model_3_with_annotations
]
self.logger.log(
'Predicting with models trained only on the training set...')
self.stream = FileStream('data/validation.csv')
num_samples = self.stream.n_remaining_samples()
for model in models_train:
self.stream.restart()
y_pred_train = self.predict(model, self.stream)
cr = classification_report(y_true, y_pred_train, labels=labels)
cm = confusion_matrix(y_true, y_pred_train, labels=labels)
print(cr)
pprint.pprint(cm)
self.logger.log(
'Predicting with models trained with the annotations added...')
for model in models_with_annotations:
self.stream.restart()
y_pred_with_annotations = self.predict(model, self.stream)
cr = classification_report(y_true,
y_pred_with_annotations,
labels=labels)
cm = confusion_matrix(y_true,
y_pred_with_annotations,
labels=labels)
print(cr)
pprint.pprint(cm)
def demo():
""" _test_stream_speed
This demo tests the sample generation speed of the file stream.
"""
# Setup the stream
opt = FileOption("FILE", "OPT_NAME", "../datasets/covtype.csv", "CSV", False)
stream = FileStream(opt, -1, 1)
stream = RandomRBFGeneratorDrift()
stream.prepare_for_use()
# Test with RandomTreeGenerator
#opt_list = [['-c', '2'], ['-o', '0'], ['-u', '5'], ['-v', '4']]
#stream = RandomTreeGenerator(opt_list)
#stream.prepare_for_use()
# Setup the evaluator
eval = EvaluateStreamGenerationSpeed(100000, float("inf"), None, 5)
# Evaluate
eval.eval(stream)
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
opt = FileOption("FILE", "OPT_NAME", "../datasets/music.csv", "CSV", False)
stream = FileStream(opt, 0, 6)
stream.prepare_for_use()
# Setup the classifier, by default it uses Logistic Regression
#classifier = MultiOutputLearner()
#classifier = MultiOutputLearner(h=SGDClassifier(n_iter=100))
classifier = MultiOutputLearner(h=Perceptron())
# Setup the pipeline
pipe = Pipeline([('classifier', classifier)])
pretrain_size = 150
logging.info('Pre training on %s samples', str(pretrain_size))
X, y = stream.next_instance(pretrain_size)
#classifier.fit(X, y)
pipe.partial_fit(X, y, classes=stream.get_classes())
count = 0
true_labels = []
predicts = []
init_time = timer()
logging.info('Evaluating...')
while stream.has_more_instances():
X, y = stream.next_instance()
#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():
""" _test_knn
This demo tests the KNN classifier on a file stream, which gives
instances coming from a SEA generator.
The test computes the performance of the KNN classifier as well as
the time to create the structure and classify max_samples (5000 by
default) instances.
"""
opt = FileOption('FILE', 'OPT_NAME', '../datasets/sea_big.csv', 'csv',
False)
stream = FileStream(opt, -1, 1)
stream.prepare_for_use()
train = 200
X, y = stream.next_instance(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 = KNN(k=8, max_window_size=2000, leaf_size=40)
#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)])
#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_instance()
#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 test_KNN(test_path, package_path):
test_file = os.path.join(package_path, 'src/skmultiflow/data/datasets/sea_big.csv')
stream = FileStream(test_file, -1, 1)
stream.prepare_for_use()
learner = KNN(n_neighbors=8, max_window_size=2000, leaf_size=40)
cnt = 0
max_samples = 5000
predictions = []
wait_samples = 100
while cnt < max_samples:
X, y = stream.next_sample()
# Test every n samples
if (cnt % wait_samples == 0) and (cnt != 0):
predictions.append(learner.predict(X)[0])
learner.partial_fit(X, y)
cnt += 1
expected_predictions = [1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0,
0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0,
1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0]
assert np.alltrue(predictions == expected_predictions)
def demo():
""" _test_streams
This demo tests if the streams are correctly generating samples.
:return:
"""
opt = FileOption('FILE', 'OPT_NAME', '../datasets/covtype.csv', 'csv',
False)
stream = FileStream(opt, -1, 1)
stream.prepare_for_use()
rbf_drift = RandomRBFGeneratorDrift(change_speed=41.00,
num_centroids=50,
model_seed=32523423,
instance_seed=5435,
num_classes=2,
num_att=10,
num_drift_centroids=50)
rbf_drift.prepare_for_use()
sea = SEAGenerator()
print('1 instance:\n')
X, y = stream.next_instance()
print(X)
print(y)
X, y = sea.next_instance()
print(X)
print(y)
print('\n\n10 instances:\n')
X, y = stream.next_instance(10)
print(X)
print(y)
X, y = sea.next_instance(10)
print(X)
print(y)
from skmultiflow.classification.lazy.knn_adwin import KNN
from skmultiflow.classification.trees.hoeffding_tree import HoeffdingTree
from skmultiflow.data.file_stream import FileStream
from skmultiflow.evaluation.evaluate_prequential import EvaluatePrequential
from skmultiflow.options.file_option import FileOption
from my_classifier import BatchClassifier
dataset = "elec"
# 1. Create a stream
opt = FileOption("FILE", "OPT_NAME", "./data/" + dataset + ".csv", "CSV",
False)
stream = FileStream(opt, -1, 1)
# 2. Prepare for use
stream.prepare_for_use()
# 2. Instantiate the HoeffdingTree classifier
h = [
KNN(k=10, max_window_size=100, leaf_size=30),
HoeffdingTree(),
BatchClassifier(window_size=100, max_models=10),
]
# 3. Setup the evaluator
eval = EvaluatePrequential(pretrain_size=1000,
output_file='result_' + dataset + '.csv',
max_instances=10000,
batch_size=1,
n_wait=500,
max_time=1000000000,
task_type='classification',
show_plot=True,
from skmultiflow.data.file_stream import FileStream
from skmultiflow.trees.hoeffding_tree import HoeffdingTreeClassifier
from skmultiflow.evaluation.evaluate_prequential import EvaluatePrequential
# Create a stream
stream = FileStream("elec.csv")
stream.prepare_for_use() # Not required for v0.5.0+
# Instantiate the HoeffdingTreeClassifier
ht = HoeffdingTreeClassifier()
# Setup the evaluator
evaluator = EvaluatePrequential(pretrain_size=1000,
max_samples=10000,
output_file='results.csv')
# Run evaluation
evaluator.evaluate(stream=stream, model=ht)
def test_file_stream(test_path, package_path):
test_file = os.path.join(package_path,
'src/skmultiflow/data/datasets/sea_stream.csv')
stream = FileStream(test_file)
stream.prepare_for_use()
assert stream.n_remaining_samples() == 40000
expected_names = ['attrib1', 'attrib2', 'attrib3']
assert stream.feature_names == expected_names
expected_targets = [0, 1]
assert stream.target_values == expected_targets
assert stream.target_names == ['class']
assert stream.n_features == 3
assert stream.n_cat_features == 0
assert stream.n_num_features == 3
assert stream.n_targets == 1
assert stream.get_data_info() == 'sea_stream.csv - 1 target(s), 2 classes'
assert stream.has_more_samples() is True
assert stream.is_restartable() is True
# Load test data corresponding to first 10 instances
test_file = os.path.join(test_path, 'sea_stream_file.npz')
data = np.load(test_file)
X_expected = data['X']
y_expected = data['y']
X, y = stream.next_sample()
assert np.alltrue(X[0] == X_expected[0])
assert np.alltrue(y[0] == y_expected[0])
X, y = stream.last_sample()
assert np.alltrue(X[0] == X_expected[0])
assert np.alltrue(y[0] == y_expected[0])
stream.restart()
X, y = stream.next_sample(10)
assert np.alltrue(X == X_expected)
assert np.alltrue(y == y_expected)
assert stream.n_targets == np.array(y).ndim
assert stream.n_features == X.shape[1]
def test_random_rbf_generator(test_path, package_path):
test_file = os.path.join(package_path,
'src/skmultiflow/datasets/sea_stream.csv')
file_option = FileOption('FILE', 'sea', test_file, 'csv', False)
stream = FileStream(file_option)
stream.prepare_for_use()
assert stream.estimated_remaining_instances() == 40000
expected_header = ['attrib1', 'attrib2', 'attrib3']
assert stream.get_attributes_header() == expected_header
expected_classes = [0, 1]
assert stream.get_classes() == expected_classes
assert stream.get_classes_header() == ['class']
assert stream.get_num_attributes() == 3
assert stream.get_num_nominal_attributes() == 0
assert stream.get_num_numerical_attributes() == 3
assert stream.get_num_targets() == 1
assert stream.get_num_values_per_nominal_attribute() == 0
assert stream.get_plot_name() == 'sea_stream.csv - 2 class labels'
assert stream.has_more_instances() is True
assert stream.is_restartable() is True
# Load test data corresponding to first 10 instances
test_file = os.path.join(test_path, 'sea_stream.npz')
data = np.load(test_file)
X_expected = data['X']
y_expected = data['y']
X, y = stream.next_instance()
assert np.alltrue(X[0] == X_expected[0])
assert np.alltrue(y[0] == y_expected[0])
X, y = stream.get_last_instance()
assert np.alltrue(X[0] == X_expected[0])
assert np.alltrue(y[0] == y_expected[0])
stream.restart()
X, y = stream.next_instance(10)
assert np.alltrue(X == X_expected)
assert np.alltrue(y == y_expected)
from skmultiflow.classification.lazy.knn import KNN
from skmultiflow.data.file_stream import FileStream
import time
import pudb;pu.db
f="/opt/techgig/scikit-multiflow/src/skmultiflow/datasets/sea_big.csv"
stream = FileStream(f, -1, 1)
stream.prepare_for_use()
X, y = stream.next_sample(200)
knn = KNN(k=8, max_window_size=2000, leaf_size=40)
knn.partial_fit(X, y)
n_samples = 0
corrects = 0
while n_samples < 5000:
X, y = stream.next_sample()
my_pred = knn.predict(X)
if y[0] == my_pred[0]:
corrects += 1
knn = knn.partial_fit(X, y)
n_samples += 1
print("KNN's performance: " + str(corrects/n_samples))
time.sleep(1)
print('KNN usage example')
print(str(n_samples) + ' samples analyzed.')
print("KNN's performance: " + str(corrects/n_samples))
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.*")
opt = FileOption('FILE', 'OPT_NAME', '../datasets/covtype.csv', 'csv', False)
stream = FileStream(opt, -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_instance(1000)
X = filter.transform(X)
#print(X)
X_find, y = stream.next_instance(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
# output- (10000, 11)
# Store it in csv
data.to_csv('data_stream.csv', index=False)
# Applying Hoeffding Tree on the synthetic data stream
# Import the relevant libraries
from skmultiflow.trees import HoeffdingTreeClassifier
from skmultiflow.evaluation import EvaluatePrequential
from skmultiflow.data.file_stream import FileStream
import pandas as pd
import numpy as np
# Load the synthetic data stream
dstream = FileStream('data_stream.csv')
dstream.prepare_for_use()
# Create the model instance
ht_class = HoeffdingTreeClassifier()
# perform prequential evaluation
evaluate1 = EvaluatePrequential(show_plot=False,
pretrain_size=400,
max_samples=10000,
metrics=['accuracy'])
evaluate1.evaluate(stream=dstream, model=ht_class)
###################################################
# Hoeffding Adaptive tree
from skmultiflow.trees import HoeffdingTree
from hoeffdingOptionTree import HOT
from skmultiflow.evaluation.evaluate_prequential import EvaluatePrequential
from skmultiflow.data.file_stream import FileStream
import matplotlib as plt
plt.interactive(True)
dataset = "elec"
# 1. Create a stream
stream = FileStream(dataset+".csv", n_targets=1, target_idx=-1)
# 2. Prepare for use
stream.prepare_for_use()
# 2. Instantiate the HoeffdingTree classifier
h = [
HoeffdingTree(),
HOT()
]
# 3. Setup the evaluator
evaluator = EvaluatePrequential(pretrain_size=1000, max_samples=20000, show_plot=True,
metrics=['accuracy', 'kappa'], output_file='result_'+dataset+'.csv',
batch_size=1)
# 4. Run
evaluator.evaluate(stream=stream, model=h)
class ARASActiveLearningTools(object):
def __init__(self):
self.id = 'ARAS_active_learning_tools'
self.logger = Log(self.id)
self.load_data()
self.load_models()
def load_data(self):
self.logger.log(
'Loading data files: train, validation, and annotations...')
self.train = pd.read_csv('data/train.csv', dtype=int)
print(self.train.shape)
print(self.train.head())
self.header = list(self.train.columns.values)
self.validation = pd.read_csv('data/validation.csv', dtype=int)
print(self.validation.shape)
print(self.validation.head())
self.annotations = pd.read_csv('data/annotations.csv', dtype=int)
self.annotations.columns = self.header
print(self.annotations.shape)
print(self.annotations.head())
def load_merged(self):
self.merged = pd.read_csv('data/merged.csv', dtype=int)
print(self.merged.shape)
print(self.merged.head())
def load_models(self):
self.model_1_train = pickle.load(open('models/train/Model1.p', 'rb'))
self.model_2_train = pickle.load(open('models/train/Model2.p', 'rb'))
self.model_3_train = pickle.load(open('models/train/Model3.p', 'rb'))
self.model_1_with_annotations = pickle.load(
open('models/with_annotations/Model1.p', 'rb'))
self.model_2_with_annotations = pickle.load(
open('models/with_annotations/Model2.p', 'rb'))
self.model_3_with_annotations = pickle.load(
open('models/with_annotations/Model3.p', 'rb'))
def merge(self):
self.logger.log(
'Merging the annotated data with the original training set...')
self.merged = pd.concat([self.train, self.annotations])
print(self.merged.shape)
print(self.merged.head())
self.logger.log('Saving dataframe...')
self.merged.to_csv('data/merged.csv', index=False)
self.logger.log_great('Done.')
# see https://scikit-learn.org/stable/modules/classes.html#module-sklearn.metrics for available metrics
def validate(self):
# self.load_merged()
y_true = self.validation['R1'].tolist()
models_train = [
self.model_1_train, self.model_2_train, self.model_3_train
]
models_with_annotations = [
self.model_1_with_annotations, self.model_2_with_annotations,
self.model_3_with_annotations
]
self.logger.log(
'Predicting with models trained only on the training set...')
self.stream = FileStream('data/validation.csv')
num_samples = self.stream.n_remaining_samples()
for model in models_train:
self.stream.restart()
y_pred_train = self.predict(model, self.stream)
cr = classification_report(y_true, y_pred_train, labels=labels)
cm = confusion_matrix(y_true, y_pred_train, labels=labels)
print(cr)
pprint.pprint(cm)
self.logger.log(
'Predicting with models trained with the annotations added...')
for model in models_with_annotations:
self.stream.restart()
y_pred_with_annotations = self.predict(model, self.stream)
cr = classification_report(y_true,
y_pred_with_annotations,
labels=labels)
cm = confusion_matrix(y_true,
y_pred_with_annotations,
labels=labels)
print(cr)
pprint.pprint(cm)
def predict(self, model, stream):
y_pred = []
count = 0
while stream.has_more_samples():
X, y = stream.next_sample()
y_pred.append(model.predict(X))
if (count % 50000) == 0:
print('Predictions so far:', count)
count = count + 1
return y_pred
