def test_zero_weight(self):
weights = CounterMap()
weights['dog'] = Counter({'warm': 2.0})
labels = set(weights.iterkeys())
logp = maxent.get_log_probabilities(self.features, weights, labels)
self.assertEqual(logp['dog'], 0.0)
def test_zero_weight(self):
weights = CounterMap()
weights['dog'] = Counter({'warm' : 2.0})
labels = set(weights.iterkeys())
logp = maxent.get_log_probabilities(self.features, weights, labels)
self.assertEqual(logp['dog'], 0.0)
def test_extraneous_label(self):
weights = CounterMap()
weights['dog'] = Counter({'warm': 2.0, 'fuzzy': 0.5})
labels = set(weights.iterkeys())
logp = maxent.get_log_probabilities(self.features, weights, labels)
self.assertEqual(logp['cat'], float('-inf'))
def test_extraneous_label(self):
weights = CounterMap()
weights['dog'] = Counter({'warm' : 2.0, 'fuzzy' : 0.5})
labels = set(weights.iterkeys())
logp = maxent.get_log_probabilities(self.features, weights, labels)
self.assertEqual(logp['cat'], float('-inf'))
def test_uneven_weights(self):
weights = CounterMap()
weights['dog'] = Counter({'warm': 2.0, 'fuzzy': 1.0})
weights['cat'] = Counter({'warm': 1.0, 'fuzzy': 1.0})
labels = set(weights.iterkeys())
logp = maxent.get_log_probabilities(self.features, weights, labels)
# construct scores
scores = Counter()
scores['dog'] = 2.0 * 1.0 + 1.0 * 1.0
scores['cat'] = 1.0 * 1.0 + 1.0 * 1.0
scores.log_normalize()
# check scores explicitly
self.assertAlmostEqual(scores['dog'], log(0.731), 3)
self.assertAlmostEqual(scores['cat'], log(0.269), 3)
# check that log probs is correct
self.assertEqual(logp['dog'], scores['dog'])
self.assertEqual(logp['cat'], scores['cat'])
def test_uneven_weights(self):
weights = CounterMap()
weights['dog'] = Counter({'warm' : 2.0, 'fuzzy' : 1.0})
weights['cat'] = Counter({'warm' : 1.0, 'fuzzy' : 1.0})
labels = set(weights.iterkeys())
logp = maxent.get_log_probabilities(self.features, weights, labels)
# construct scores
scores = Counter()
scores['dog'] = 2.0 * 1.0 + 1.0 * 1.0
scores['cat'] = 1.0 * 1.0 + 1.0 * 1.0
scores.log_normalize()
# check scores explicitly
self.assertAlmostEqual(scores['dog'], log(0.731), 3)
self.assertAlmostEqual(scores['cat'], log(0.269), 3)
# check that log probs is correct
self.assertEqual(logp['dog'], scores['dog'])
self.assertEqual(logp['cat'], scores['cat'])
class NaiveBayesClassifier: def train(self, labeled_data): self.feature_distribution = CounterMap() labels = set() for label, datum in labeled_data: labels.add(label) for feature in ngrams(datum, 3) self.feature_distribution[feature][label] += 1 for feature in self.feature_distribution.iterkeys(): self.feature_distribution[feature].default = 0.01 self.feature_distribution.normalize() self.feature_distribution.log() def label_distribution(self, datum): distribution = None for feature in ngrams(datum, 3): if distribution: distribution += self.feature_distribution[feature] else: distribution = copy(self.feature_distribution[feature]) distribution.log_normalize() return distribution def label(self, datum): distribution = None for feature in ngrams(datum, 3): if distribution: distribution += self.feature_distribution[feature] else: distribution = copy(self.feature_distribution[feature]) return distribution.arg_max()
class MaximumEntropyLogProbsTest(unittest.TestCase):
def setUp(self):
self.features = Counter((key, 1.0) for key in ['warm', 'fuzzy'])
self.weights = CounterMap()
self.weights['dog'] = Counter({'warm': 2.0, 'fuzzy': 0.5})
self.weights['cat'] = Counter({'warm': 0.5, 'fuzzy': 2.0})
self.labels = set(self.weights.iterkeys())
self.logp = maxent.get_log_probabilities(self.features, self.weights,
self.labels)
def test_fast_slow_equal(self):
slow_logp = maximumentropy.slow_log_probs(self.features, self.weights,
self.labels)
self.assertEqual(self.logp, slow_logp)
def test_logp_is_probability_distribution(self):
"""
Verify that all log probs are <= 0 and total probability is 1.0
"""
self.assertTrue(max(self.logp.itervalues()) <= 0.0)
self.assertAlmostEqual(sum(exp(val) for val in self.logp.itervalues()),
1.0)
def test_basic_values(self):
"""
Are the log probs as expected?
"""
self.assertAlmostEqual(exp(self.logp['cat']), 0.5)
self.assertAlmostEqual(exp(self.logp['dog']), 0.5)
def test_single_label(self):
weights = CounterMap()
weights['dog'] = Counter({'warm': 2.0, 'fuzzy': 0.5})
labels = set(weights.iterkeys())
logp = maxent.get_log_probabilities(self.features, weights, labels)
self.assertEqual(logp['dog'], 0.0)
def test_extraneous_label(self):
weights = CounterMap()
weights['dog'] = Counter({'warm': 2.0, 'fuzzy': 0.5})
labels = set(weights.iterkeys())
logp = maxent.get_log_probabilities(self.features, weights, labels)
self.assertEqual(logp['cat'], float('-inf'))
def test_zero_weight(self):
weights = CounterMap()
weights['dog'] = Counter({'warm': 2.0})
labels = set(weights.iterkeys())
logp = maxent.get_log_probabilities(self.features, weights, labels)
self.assertEqual(logp['dog'], 0.0)
def test_uneven_weights(self):
weights = CounterMap()
weights['dog'] = Counter({'warm': 2.0, 'fuzzy': 1.0})
weights['cat'] = Counter({'warm': 1.0, 'fuzzy': 1.0})
labels = set(weights.iterkeys())
logp = maxent.get_log_probabilities(self.features, weights, labels)
# construct scores
scores = Counter()
scores['dog'] = 2.0 * 1.0 + 1.0 * 1.0
scores['cat'] = 1.0 * 1.0 + 1.0 * 1.0
scores.log_normalize()
# check scores explicitly
self.assertAlmostEqual(scores['dog'], log(0.731), 3)
self.assertAlmostEqual(scores['cat'], log(0.269), 3)
# check that log probs is correct
self.assertEqual(logp['dog'], scores['dog'])
self.assertEqual(logp['cat'], scores['cat'])
def test_performance(self):
"""
C api should be faster than python API (this is potentialy flakey, depending on system load patterns)
"""
start = time.time()
for i in xrange(100000):
test = maximumentropy.slow_log_probs(self.features, self.weights,
self.labels)
slow_time = time.time() - start
start = time.time()
for i in xrange(100000):
test = maxent.get_log_probabilities(self.features, self.weights,
self.labels)
fast_time = time.time() - start
self.assertTrue(fast_time < slow_time)
class MaximumEntropyLogProbsTest(unittest.TestCase):
def setUp(self):
self.features = Counter((key, 1.0) for key in ['warm', 'fuzzy'])
self.weights = CounterMap()
self.weights['dog'] = Counter({'warm' : 2.0, 'fuzzy' : 0.5})
self.weights['cat'] = Counter({'warm' : 0.5, 'fuzzy' : 2.0})
self.labels = set(self.weights.iterkeys())
self.logp = maxent.get_log_probabilities(self.features, self.weights, self.labels)
def test_fast_slow_equal(self):
slow_logp = maximumentropy.slow_log_probs(self.features, self.weights, self.labels)
self.assertEqual(self.logp, slow_logp)
def test_logp_is_probability_distribution(self):
"""
Verify that all log probs are <= 0 and total probability is 1.0
"""
self.assertTrue(max(self.logp.itervalues()) <= 0.0)
self.assertAlmostEqual(sum(exp(val) for val in self.logp.itervalues()), 1.0)
def test_basic_values(self):
"""
Are the log probs as expected?
"""
self.assertAlmostEqual(exp(self.logp['cat']), 0.5)
self.assertAlmostEqual(exp(self.logp['dog']), 0.5)
def test_single_label(self):
weights = CounterMap()
weights['dog'] = Counter({'warm' : 2.0, 'fuzzy' : 0.5})
labels = set(weights.iterkeys())
logp = maxent.get_log_probabilities(self.features, weights, labels)
self.assertEqual(logp['dog'], 0.0)
def test_extraneous_label(self):
weights = CounterMap()
weights['dog'] = Counter({'warm' : 2.0, 'fuzzy' : 0.5})
labels = set(weights.iterkeys())
logp = maxent.get_log_probabilities(self.features, weights, labels)
self.assertEqual(logp['cat'], float('-inf'))
def test_zero_weight(self):
weights = CounterMap()
weights['dog'] = Counter({'warm' : 2.0})
labels = set(weights.iterkeys())
logp = maxent.get_log_probabilities(self.features, weights, labels)
self.assertEqual(logp['dog'], 0.0)
def test_uneven_weights(self):
weights = CounterMap()
weights['dog'] = Counter({'warm' : 2.0, 'fuzzy' : 1.0})
weights['cat'] = Counter({'warm' : 1.0, 'fuzzy' : 1.0})
labels = set(weights.iterkeys())
logp = maxent.get_log_probabilities(self.features, weights, labels)
# construct scores
scores = Counter()
scores['dog'] = 2.0 * 1.0 + 1.0 * 1.0
scores['cat'] = 1.0 * 1.0 + 1.0 * 1.0
scores.log_normalize()
# check scores explicitly
self.assertAlmostEqual(scores['dog'], log(0.731), 3)
self.assertAlmostEqual(scores['cat'], log(0.269), 3)
# check that log probs is correct
self.assertEqual(logp['dog'], scores['dog'])
self.assertEqual(logp['cat'], scores['cat'])
def test_performance(self):
"""
C api should be faster than python API (this is potentialy flakey, depending on system load patterns)
"""
start = time.time()
for i in xrange(100000):
test = maximumentropy.slow_log_probs(self.features, self.weights, self.labels)
slow_time = time.time() - start
start = time.time()
for i in xrange(100000):
test = maxent.get_log_probabilities(self.features, self.weights, self.labels)
fast_time = time.time() - start
self.assertTrue(fast_time < slow_time)
