def test_knn():
stream = SEAGenerator(random_state=1)
learner = KNNClassifier(n_neighbors=8, max_window_size=2000, leaf_size=40)
cnt = 0
max_samples = 5000
predictions = array('i')
correct_predictions = 0
wait_samples = 100
X_batch = []
y_batch = []
while cnt < max_samples:
X, y = stream.next_sample()
X_batch.append(X[0])
y_batch.append(y[0])
# 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
expected_predictions = array('i', [
1, 1, 1, 0, 1, 1, 0, 0, 0, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 1, 0, 0, 1, 0,
1
])
assert np.alltrue(predictions == expected_predictions)
expected_correct_predictions = 49
assert correct_predictions == expected_correct_predictions
expected_info = "KNNClassifier(leaf_size=40, max_window_size=2000, " \
"metric='euclidean', n_neighbors=8)"
info = " ".join([line.strip() for line in learner.get_info().split()])
assert info == expected_info
learner.reset()
info = " ".join([line.strip() for line in learner.get_info().split()])
assert info == expected_info
X_batch = np.array(X_batch)
y_batch = np.array(y_batch)
learner.fit(X_batch[:4500], y_batch[:4500], classes=[0, 1])
predictions = learner.predict(X_batch[4501:4550])
expected_predictions = array('i', [
1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 0, 1, 0, 1, 1,
1, 1, 0, 0, 1, 0, 0, 1, 1, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 1,
0
])
assert np.alltrue(predictions == expected_predictions)
correct_predictions = sum(predictions == y_batch[4501:4550])
expected_correct_predictions = 49
assert correct_predictions == expected_correct_predictions
assert type(learner.predict(X)) == np.ndarray
assert type(learner.predict_proba(X)) == np.ndarray
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
