def retrain(self, labeled_tweets: list):
labels = set()
for tweet in labeled_tweets:
if "labels" in tweet and len(tweet["labels"]) > 0:
labels.update([l for l in tweet["labels"] if not (l in self.labels_sent or l in self.labels_relevance)])
self.labels = list(labels)
assert "Irrelevant" not in self.labels, "Something went wrong"
self.lcc = ClassifierChain(SGDClassifier(max_iter=100, loss='log', random_state=1))
self.clrel = KNNClassifier()
self.clsent = KNNClassifier()
X, y, ys, yr = [], [], [], []
for tweet in labeled_tweets:
if "labels" in tweet and len(tweet["labels"]) > 0:
X.append(tweet["tweet"])
y.append(self._labels2array(tweet["labels"]))
sls = [l for l, v in tweet["labels"].items() if l in self.labels_sent and v]
if len(sls) == 1:
ys.append(self.labels_sent[sls[0]])
else:
ys.append(self.labels_sent["NEUTRAL"])
if self.labels_relevance[0] in tweet["labels"] and tweet["labels"][self.labels_relevance[0]]:
yr.append(1)
else:
yr.append(0)
X = np.array(self.tokenizer.transform(X).todense())
y = np.array(y)
ys = np.array(ys)
yr = np.array(yr)
self.clsent.fit(X, ys)
print("Trained Sentiment Classifier")
self.clrel.fit(X, yr)
print("Trained Relevance Classifier")
X2, y2 = [], []
for Xe, ye in zip(X, y):
if ye.sum() > 0:
X2.append(Xe)
y2.append(ye)
X = np.array(X2)
y = np.array(y2)
self.lcc.fit(X, y)
print("Trained Catecorical Classifier")
def __init__(self, texts: list):
self.tokenizer = TfidfVectorizer()
self.tokenizer.fit(texts)
self.labels_sent = {"POSITIVE": np.array([1, 0, 0]), "NEUTRAL": np.array([0, 1, 0]),
"NEGATIVE": np.array([0, 0, 1])}
self.labels_sent = {"POSITIVE": 0, "NEUTRAL": 1,
"NEGATIVE": 2}
self.reverse_sent = {0: {"POSITIVE": True, "NEUTRAL": False,
"NEGATIVE": False},
1: {"POSITIVE": False, "NEUTRAL": True,
"NEGATIVE": False},
2: {"POSITIVE": False, "NEUTRAL": False,
"NEGATIVE": True}}
self.labels_relevance = ["Irrelevant"]
self.labels = []
self.lcc = ClassifierChain(SGDClassifier(max_iter=100, loss='log', random_state=1))
self.clrel = KNNClassifier()
self.clsent = KNNClassifier()
def test_classifier_chains_all():
seed = 1
X, Y = make_logical(random_state=seed)
# CC
cc = ClassifierChain(SGDClassifier(max_iter=100, tol=1e-3, loss='log', random_state=seed))
cc.partial_fit(X, Y)
y_predicted = cc.predict(X)
y_expected = [[1, 0, 1], [1, 1, 0], [0, 0, 0], [1, 1, 0]]
assert np.alltrue(y_predicted == y_expected)
assert type(cc.predict_proba(X)) == np.ndarray
# RCC
rcc = ClassifierChain(SGDClassifier(max_iter=100, tol=1e-3, loss='log', random_state=seed), order='random',
random_state=seed)
rcc.partial_fit(X, Y)
y_predicted = rcc.predict(X)
y_expected = [[1, 0, 1], [1, 1, 0], [0, 0, 0], [1, 1, 0]]
assert np.alltrue(y_predicted == y_expected)
# MCC
mcc = MonteCarloClassifierChain(SGDClassifier(max_iter=100, tol=1e-3, loss='log', random_state=seed), M=1000)
mcc.partial_fit(X, Y)
y_predicted = mcc.predict(X)
y_expected = [[1, 0, 1], [1, 1, 0], [0, 0, 0], [1, 1, 0]]
assert np.alltrue(y_predicted == y_expected)
# PCC
pcc = ProbabilisticClassifierChain(SGDClassifier(max_iter=100, tol=1e-3, loss='log', random_state=seed))
pcc.partial_fit(X, Y)
y_predicted = pcc.predict(X)
y_expected = [[1, 0, 1], [1, 1, 0], [0, 0, 0], [1, 1, 0]]
assert np.alltrue(y_predicted == y_expected)
def test_classifier_chains():
seed = 112
stream = MultilabelGenerator(random_state=seed, n_targets=3, n_samples=5150)
stream.prepare_for_use()
estimator = SGDClassifier(random_state=seed, tol=1e-3, max_iter=10)
learner = ClassifierChain(base_estimator=estimator, random_state=seed)
X, y = stream.next_sample(150)
learner.partial_fit(X, y)
cnt = 0
max_samples = 5000
predictions = []
true_labels = []
wait_samples = 100
correct_predictions = 0
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])
true_labels.append(y[0])
if np.array_equal(y[0], predictions[-1]):
correct_predictions += 1
learner.partial_fit(X, y)
cnt += 1
if not sklearn_version.startswith("0.21"):
expected_predictions = [[0.0, 0.0, 1.0], [1.0, 0.0, 0.0], [1.0, 0.0, 1.0], [1.0, 1.0, 1.0], [1.0, 0.0, 1.0],
[1.0, 0.0, 0.0], [1.0, 0.0, 1.0], [1.0, 0.0, 1.0], [0.0, 0.0, 1.0], [0.0, 0.0, 0.0],
[1.0, 0.0, 0.0], [0.0, 1.0, 1.0], [0.0, 0.0, 1.0], [0.0, 0.0, 1.0], [0.0, 0.0, 1.0],
[0.0, 0.0, 1.0], [1.0, 1.0, 1.0], [1.0, 0.0, 1.0], [1.0, 1.0, 1.0], [0.0, 0.0, 0.0],
[0.0, 0.0, 1.0], [0.0, 0.0, 1.0], [0.0, 0.0, 1.0], [0.0, 1.0, 1.0], [0.0, 1.0, 1.0],
[1.0, 1.0, 1.0], [0.0, 0.0, 0.0], [0.0, 1.0, 0.0], [1.0, 1.0, 1.0], [1.0, 1.0, 0.0],
[0.0, 1.0, 0.0], [1.0, 0.0, 1.0], [1.0, 1.0, 1.0], [0.0, 0.0, 0.0], [0.0, 0.0, 0.0],
[1.0, 0.0, 0.0], [1.0, 1.0, 1.0], [0.0, 1.0, 1.0], [0.0, 0.0, 0.0], [1.0, 0.0, 1.0],
[0.0, 0.0, 1.0], [0.0, 0.0, 0.0], [1.0, 0.0, 1.0], [0.0, 0.0, 1.0], [0.0, 1.0, 0.0],
[0.0, 0.0, 0.0], [1.0, 1.0, 1.0], [0.0, 0.0, 0.0], [1.0, 1.0, 1.0]]
assert np.alltrue(np.array_equal(predictions, expected_predictions))
expected_correct_predictions = 21
assert correct_predictions == expected_correct_predictions
expected_info = "ClassifierChain(base_estimator=SGDClassifier(alpha=0.0001, average=False, class_weight=None,\n" \
" early_stopping=False, epsilon=0.1, eta0=0.0, fit_intercept=True,\n" \
" l1_ratio=0.15, learning_rate='optimal', loss='hinge', max_iter=10,\n" \
" n_iter=None, n_iter_no_change=5, n_jobs=None, penalty='l2',\n" \
" power_t=0.5, random_state=112, shuffle=True, tol=0.001,\n" \
" validation_fraction=0.1, verbose=0, warm_start=False),\n" \
" order=None, random_state=112)"
assert learner.get_info() == expected_info
else:
expected_predictions = [[0.0, 0.0, 1.0], [0.0, 0.0, 0.0], [1.0, 0.0, 1.0], [1.0, 0.0, 1.0], [0.0, 0.0, 1.0],
[1.0, 0.0, 0.0], [1.0, 0.0, 1.0], [1.0, 0.0, 1.0], [0.0, 0.0, 1.0], [0.0, 0.0, 0.0],
[1.0, 0.0, 1.0], [0.0, 0.0, 1.0], [0.0, 0.0, 1.0], [0.0, 0.0, 1.0], [0.0, 0.0, 1.0],
[0.0, 0.0, 1.0], [1.0, 0.0, 1.0], [0.0, 0.0, 0.0], [1.0, 0.0, 1.0], [0.0, 0.0, 0.0],
[0.0, 1.0, 1.0], [0.0, 1.0, 1.0], [0.0, 0.0, 1.0], [0.0, 1.0, 1.0], [0.0, 1.0, 1.0],
[0.0, 1.0, 1.0], [0.0, 1.0, 0.0], [0.0, 1.0, 0.0], [1.0, 1.0, 1.0], [0.0, 1.0, 0.0],
[0.0, 1.0, 1.0], [1.0, 0.0, 1.0], [0.0, 1.0, 1.0], [0.0, 0.0, 0.0], [0.0, 0.0, 0.0],
[1.0, 0.0, 0.0], [1.0, 1.0, 1.0], [0.0, 1.0, 1.0], [0.0, 0.0, 0.0], [1.0, 0.0, 1.0],
[0.0, 0.0, 1.0], [0.0, 0.0, 0.0], [0.0, 0.0, 0.0], [0.0, 0.0, 1.0], [0.0, 1.0, 0.0],
[0.0, 0.0, 0.0], [1.0, 1.0, 1.0], [0.0, 0.0, 0.0], [1.0, 1.0, 1.0]]
assert np.alltrue(np.array_equal(predictions, expected_predictions))
expected_correct_predictions = 26
assert correct_predictions == expected_correct_predictions
expected_info = "ClassifierChain(base_estimator=SGDClassifier(alpha=0.0001, average=False, class_weight=None,\n" \
" early_stopping=False, epsilon=0.1, eta0=0.0, fit_intercept=True,\n" \
" l1_ratio=0.15, learning_rate='optimal', loss='hinge', max_iter=10,\n" \
" n_iter_no_change=5, n_jobs=None, penalty='l2', power_t=0.5,\n" \
" random_state=112, shuffle=True, tol=0.001,\n" \
" validation_fraction=0.1, verbose=0, warm_start=False),\n" \
" order=None, random_state=112)"
assert learner.get_info() == expected_info
assert type(learner.predict(X)) == np.ndarray
def test_classifier_chains():
seed = 112
stream = MultilabelGenerator(random_state=seed,
n_targets=3,
n_samples=5150)
estimator = SGDClassifier(random_state=seed, max_iter=10)
learner = ClassifierChain(base_estimator=estimator, random_state=seed)
X, y = get_next_n_samples(stream, 150)
learner.partial_fit(X, y)
cnt = 0
max_samples = 5000
predictions = []
true_labels = []
wait_samples = 100
correct_predictions = 0
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])
true_labels.append(y[0])
if np.array_equal(y[0], predictions[-1]):
correct_predictions += 1
learner.partial_fit(X, y)
cnt += 1
if not sklearn_version.startswith("0.21"):
expected_predictions = [[0., 0., 1.], [0., 0., 0.], [1., 0., 1.],
[1., 0., 1.], [0., 0., 1.], [1., 0., 0.],
[1., 0., 1.], [1., 0., 1.], [0., 0., 1.],
[0., 0., 0.], [1., 0., 1.], [0., 0., 1.],
[0., 0., 1.], [0., 0., 1.], [0., 0., 1.],
[0., 0., 1.], [1., 0., 1.], [0., 0., 0.],
[1., 0., 1.], [0., 0., 0.], [0., 1., 1.],
[0., 1., 1.], [0., 0., 1.], [0., 1., 1.],
[0., 1., 1.], [0., 1., 1.], [0., 1., 0.],
[0., 1., 0.], [1., 1., 1.], [0., 1., 0.],
[0., 1., 1.], [1., 0., 1.], [0., 1., 1.],
[0., 0., 0.], [0., 0., 0.], [1., 0., 0.],
[1., 1., 1.], [0., 1., 1.], [0., 0., 0.],
[1., 0., 1.], [0., 0., 1.], [0., 0., 0.],
[0., 0., 0.], [0., 0., 1.], [0., 1., 0.],
[0., 0., 0.], [1., 1., 1.], [0., 0., 0.],
[1., 1., 1.]]
assert np.alltrue(np.array_equal(predictions, expected_predictions))
expected_correct_predictions = 26
assert correct_predictions == expected_correct_predictions
expected_info = "ClassifierChain(base_estimator=SGDClassifier(max_iter=10, " \
"random_state=112), order=None, random_state=112)"
info = " ".join([line.strip() for line in learner.get_info().split()])
assert info == expected_info
else:
expected_predictions = [[0.0, 0.0, 1.0], [0.0, 0.0, 0.0],
[1.0, 0.0, 1.0], [1.0, 0.0, 1.0],
[0.0, 0.0, 1.0], [1.0, 0.0, 0.0],
[1.0, 0.0, 1.0], [1.0, 0.0, 1.0],
[0.0, 0.0, 1.0], [0.0, 0.0, 0.0],
[1.0, 0.0, 1.0], [0.0, 0.0, 1.0],
[0.0, 0.0, 1.0], [0.0, 0.0, 1.0],
[0.0, 0.0, 1.0], [0.0, 0.0, 1.0],
[1.0, 0.0, 1.0], [0.0, 0.0, 0.0],
[1.0, 0.0, 1.0], [0.0, 0.0, 0.0],
[0.0, 1.0, 1.0], [0.0, 1.0, 1.0],
[0.0, 0.0, 1.0], [0.0, 1.0, 1.0],
[0.0, 1.0, 1.0], [0.0, 1.0, 1.0],
[0.0, 1.0, 0.0], [0.0, 1.0, 0.0],
[1.0, 1.0, 1.0], [0.0, 1.0, 0.0],
[0.0, 1.0, 1.0], [1.0, 0.0, 1.0],
[0.0, 1.0, 1.0], [0.0, 0.0, 0.0],
[0.0, 0.0, 0.0], [1.0, 0.0, 0.0],
[1.0, 1.0, 1.0], [0.0, 1.0, 1.0],
[0.0, 0.0, 0.0], [1.0, 0.0, 1.0],
[0.0, 0.0, 1.0], [0.0, 0.0, 0.0],
[0.0, 0.0, 0.0], [0.0, 0.0, 1.0],
[0.0, 1.0, 0.0], [0.0, 0.0, 0.0],
[1.0, 1.0, 1.0], [0.0, 0.0, 0.0],
[1.0, 1.0, 1.0]]
assert np.alltrue(np.array_equal(predictions, expected_predictions))
expected_correct_predictions = 26
assert correct_predictions == expected_correct_predictions
expected_info = "ClassifierChain(base_estimator=SGDClassifier(max_iter=10, " \
"random_state=112), order=None, random_state=112)"
info = " ".join([line.strip() for line in learner.get_info().split()])
assert info == expected_info
assert type(learner.predict(X)) == np.ndarray
n_targets=data_stream.n_targets
),
"ensemble": False
},
"br_nb": {
"name": "Binary Relevance - Naive Bayes",
"model": lambda data_stream: MultiOutputLearner(
NaiveBayes(),
n_targets=data_stream.n_targets
),
"ensemble": False
},
"cc": {
"name": "Classifier Chain - Perceptron",
"model": lambda data_stream: ClassifierChain(
Perceptron(),
n_targets=data_stream.n_targets
),
"ensemble": False
},
"cc_ht": {
"name": "Binary Relevance - HoeffdingTreeClassifier",
"model": lambda data_stream: ClassifierChain(
PerceptronMask(),
n_targets=data_stream.n_targets
),
"ensemble": False
},
"cc_nb": {
"name": "Classifier Chain - Naive Bayes",
"model": lambda data_stream: ClassifierChain(
NaiveBayes(),
class LabelPredict:
def __init__(self, texts: list):
self.tokenizer = TfidfVectorizer()
self.tokenizer.fit(texts)
self.labels_sent = {"POSITIVE": np.array([1, 0, 0]), "NEUTRAL": np.array([0, 1, 0]),
"NEGATIVE": np.array([0, 0, 1])}
self.labels_sent = {"POSITIVE": 0, "NEUTRAL": 1,
"NEGATIVE": 2}
self.reverse_sent = {0: {"POSITIVE": True, "NEUTRAL": False,
"NEGATIVE": False},
1: {"POSITIVE": False, "NEUTRAL": True,
"NEGATIVE": False},
2: {"POSITIVE": False, "NEUTRAL": False,
"NEGATIVE": True}}
self.labels_relevance = ["Irrelevant"]
self.labels = []
self.lcc = ClassifierChain(SGDClassifier(max_iter=100, loss='log', random_state=1))
self.clrel = KNNClassifier()
self.clsent = KNNClassifier()
def _labels2array(self, labeldict: dict):
target = []
for label in self.labels:
if label in labeldict and labeldict[label] == True:
target.append(1)
else:
target.append(0)
return np.array(target)
def retrain(self, labeled_tweets: list):
labels = set()
for tweet in labeled_tweets:
if "labels" in tweet and len(tweet["labels"]) > 0:
labels.update([l for l in tweet["labels"] if not (l in self.labels_sent or l in self.labels_relevance)])
self.labels = list(labels)
assert "Irrelevant" not in self.labels, "Something went wrong"
self.lcc = ClassifierChain(SGDClassifier(max_iter=100, loss='log', random_state=1))
self.clrel = KNNClassifier()
self.clsent = KNNClassifier()
X, y, ys, yr = [], [], [], []
for tweet in labeled_tweets:
if "labels" in tweet and len(tweet["labels"]) > 0:
X.append(tweet["tweet"])
y.append(self._labels2array(tweet["labels"]))
sls = [l for l, v in tweet["labels"].items() if l in self.labels_sent and v]
if len(sls) == 1:
ys.append(self.labels_sent[sls[0]])
else:
ys.append(self.labels_sent["NEUTRAL"])
if self.labels_relevance[0] in tweet["labels"] and tweet["labels"][self.labels_relevance[0]]:
yr.append(1)
else:
yr.append(0)
X = np.array(self.tokenizer.transform(X).todense())
y = np.array(y)
ys = np.array(ys)
yr = np.array(yr)
self.clsent.fit(X, ys)
print("Trained Sentiment Classifier")
self.clrel.fit(X, yr)
print("Trained Relevance Classifier")
X2, y2 = [], []
for Xe, ye in zip(X, y):
if ye.sum() > 0:
X2.append(Xe)
y2.append(ye)
X = np.array(X2)
y = np.array(y2)
self.lcc.fit(X, y)
print("Trained Catecorical Classifier")
def predict(self, text: str):
X = np.array(self.tokenizer.transform([text]).todense()).reshape((1, -1))
predicted = self.lcc.predict(X)
labels_add = {label: bool(value) for label, value in zip(self.labels, predicted.flatten())}
sent_pred = self.clsent.predict(X)
labels_add.update(self.reverse_sent[sent_pred.flatten()[0]])
assert "POSITIVE" in labels_add, "Klassifikation nicht eindeutig"
if self.clrel.predict(X) == np.array([1]):
labels_add[self.labels_relevance[0]] = True
else:
labels_add[self.labels_relevance[0]] = False
return labels_add
def train_item(self, tweet):
text = tweet["tweet"]
labeldict = tweet["labels"]
for l in labeldict:
if l not in self.labels and l not in self.labels_relevance and l not in self.labels_sent:
print("RETRAIN!")
return False
y = self._labels2array(labeldict).reshape((1, -1))
X = np.array(self.tokenizer.transform([text]).todense()).reshape((1, -1))
sls = [l for l, v in labeldict.items() if l in self.labels_sent and v]
if len(sls) == 1:
ys = self.labels_sent[sls[0]]
else:
ys = self.labels_sent["NEUTRAL"]
ys = np.array([ys])
if self.labels_relevance[0] in labeldict and labeldict[self.labels_relevance[0]]:
yr = np.array([1])
else:
yr = np.array([0])
if y.sum() > 0:
self.lcc.partial_fit(X, y)
if yr.sum() > 0:
self.clrel.partial_fit(X, yr)
if ys.sum() > 0:
self.clsent.partial_fit(X, ys)
return True