def test_simple_duplicates(self):
dupe = Datapoint(phraseid="a", sentenceid="b", phrase="b a", sentiment="1")
# Train has a lot of "2" sentiments
train = [Datapoint(phraseid=str(i),
sentenceid=str(i),
phrase="a b",
sentiment="2") for i in range(10)]
train.append(dupe)
test = [Datapoint(*dupe)]
predictor = PhraseSentimentPredictor(duplicates=True)
predictor.fit(train)
predicted = predictor.predict(test)[0]
self.assertEqual(predicted, "1")
def test_simple_predict(self):
train, test = corpus.make_train_test_split("inhaler")
predictor = PhraseSentimentPredictor()
predictor.fit(train)
predictions = predictor.predict(test)
# Same amount of predictions than input values
self.assertEqual(len(predictions), len(test))
# Predicted labels where seen during training
train_labels = set(x.sentiment for x in train)
predicted_labels = set(predictions)
self.assertEqual(predicted_labels - train_labels, set())
def test_simple_error_matrix(self):
train, test = corpus.make_train_test_split("reflektor", proportion=0.4)
predictor = PhraseSentimentPredictor()
predictor.fit(train)
error = predictor.error_matrix(test)
for real, predicted in error.keys():
self.assertNotEqual(real, predicted)
score = predictor.score(test)
assert score > 0, "Test is valid only if score is more than 0"
N = float(len(test))
wrong = sum(len(xs) for xs in error.values())
self.assertEqual((N - wrong) / N, score)
def test_simple_predict(self):
train, test = corpus.make_train_test_split("inhaler")
predictor = PhraseSentimentPredictor()
predictor.fit(train)
predictions = predictor.predict(test)
# Same amount of predictions than input values
self.assertEqual(len(predictions), len(test))
# Predicted labels where seen during training
train_labels = set(x.sentiment for x in train)
predicted_labels = set(predictions)
self.assertEqual(predicted_labels - train_labels, set())
def test_simple_duplicates(self):
dupe = Datapoint(phraseid="a",
sentenceid="b",
phrase="b a",
sentiment="1")
# Train has a lot of "2" sentiments
train = [
Datapoint(phraseid=str(i),
sentenceid=str(i),
phrase="a b",
sentiment="2") for i in range(10)
]
train.append(dupe)
test = [Datapoint(*dupe)]
predictor = PhraseSentimentPredictor(duplicates=True)
predictor.fit(train)
predicted = predictor.predict(test)[0]
self.assertEqual(predicted, "1")
def test_simple_error_matrix(self):
train, test = corpus.make_train_test_split("reflektor", proportion=0.4)
predictor = PhraseSentimentPredictor()
predictor.fit(train)
error = predictor.error_matrix(test)
for real, predicted in error.keys():
self.assertNotEqual(real, predicted)
score = predictor.score(test)
assert score > 0, "Test is valid only if score is more than 0"
N = float(len(test))
wrong = sum(len(xs) for xs in error.values())
self.assertEqual((N - wrong) / N, score)
if __name__ == "__main__":
import argparse
import json
import csv
import sys
from samr.corpus import iter_corpus, iter_test_corpus
from samr.predictor import PhraseSentimentPredictor
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("filename")
config = parser.parse_args()
config = json.load(open(config.filename))
t1=int(round(time.time() *1000))
predictor = PhraseSentimentPredictor(**config)
predictor.fit(list(iter_corpus()))
t2=int(round(time.time() *1000))
test = list(iter_test_corpus())
prediction = predictor.predict2(test)
t3=int(round(time.time() *1000))
print t2-t1, t3-t2
writer = csv.writer(sys.stdout)
writer.writerow(("PhraseId", "Sentiment"))
for datapoint, sentiment in zip(test, prediction):
writer.writerow((datapoint.phraseid, sentiment))
value = float(value)
except ValueError:
pass
new[key] = value
return new
if __name__ == "__main__":
import argparse
import json
import csv
import sys
from samr.corpus import iter_corpus, iter_test_corpus
from samr.predictor import PhraseSentimentPredictor
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("filename")
config = parser.parse_args()
config = json.load(open(config.filename))
predictor = PhraseSentimentPredictor(**config)
predictor.fit(list(iter_corpus()))
test = list(iter_test_corpus())
prediction = predictor.predict(test)
writer = csv.writer(sys.stdout)
writer.writerow(("PhraseId", "Sentiment"))
for datapoint, sentiment in zip(test, prediction):
writer.writerow((datapoint.phraseid, sentiment))
def test_fit_returns_self(self):
train, _ = corpus.make_train_test_split("defiant order")
predictor = PhraseSentimentPredictor()
s = predictor.fit(train)
self.assertEqual(predictor, s)
def test_fit_returns_self(self):
train, _ = corpus.make_train_test_split("defiant order")
predictor = PhraseSentimentPredictor()
s = predictor.fit(train)
self.assertEqual(predictor, s)
def report(self, score):
new = time.time()
self.i += 1
print("individual {}-th fold score={}% took {} seconds".format(
self.i, score * 100, new - self.last))
self.last = new
if __name__ == "__main__":
import argparse
import json
from samr.evaluation import cross_validation
from samr.predictor import PhraseSentimentPredictor
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("filename")
config = parser.parse_args()
config = json.load(open(config.filename))
factory = lambda: PhraseSentimentPredictor(**config)
factory() # Run once to check config is ok
report = PrintPartialCV()
result = cross_validation(factory,
seed="robot rock",
callback=report.report)
print("10-fold cross validation score {}%".format(result * 100))
def runThroughSetup(self, **kwargs):
predictor = PhraseSentimentPredictor(**kwargs)
predictor.fit(self.train[:self.samples])
return str(predictor.score(self.test))
