def testClassifyKeywordsSingleAndMultiClass(self):
"""Tests simple classification with multiple labels for Keywords model."""
model = ClassificationModelKeywords()
samples = {
0: (["Pickachu"], numpy.array([0, 2, 2])),
1: (["Eevee"], numpy.array([2])),
2: (["Charmander"], numpy.array([0, 1, 1])),
3: (["Abra"], numpy.array([1])),
4: (["Squirtle"], numpy.array([1, 0, 1])),
}
patterns = model.encodeSamples(samples)
for i in xrange(len(samples)):
model.trainModel(i)
output = [model.testModel(i) for i in xrange(len(patterns))]
self.assertSequenceEqual(output[0].tolist(), [2, 0], "Incorrect output for first sample.")
self.assertSequenceEqual(output[1].tolist(), [2], "Incorrect output for second sample.")
self.assertSequenceEqual(output[2].tolist(), [1, 0], "Incorrect output for third sample.")
self.assertSequenceEqual(output[3].tolist(), [1], "Incorrect output for fourth sample.")
# Test the order of class labels doesn't matter when training.
self.assertTrue(numpy.allclose(output[2], output[4]), "Outputs for samples 2 and 4 should be identical.")
def testModelSaveAndLoad(self):
# Keywords model uses the base class implementations of save/load methods.
self.modelDir = "poke_model"
model = ClassificationModelKeywords(modelDir=self.modelDir,
verbosity=0)
samples = {
0: (["Pickachu"], numpy.array([0, 2, 2])),
1: (["Eevee"], numpy.array([2])),
2: (["Charmander"], numpy.array([0, 1, 1])),
3: (["Abra"], numpy.array([1])),
4: (["Squirtle"], numpy.array([1, 0, 1]))
}
patterns = model.encodeSamples(samples)
for i in xrange(len(samples)):
model.trainModel(i)
output = [model.testModel(i) for i in xrange(len(patterns))]
model.saveModel()
loadedModel = ClassificationModel(verbosity=0).loadModel(self.modelDir)
loadedModelOutput = [
loadedModel.testModel(i) for i in xrange(len(patterns))
]
for mClasses, lClasses in zip(output, loadedModelOutput):
self.assertSequenceEqual(
mClasses.tolist(), lClasses.tolist(), "Output "
"classifcations from loaded model don't match original model's."
)
def createModel(args):
"""
Return a classification model of the appropriate type. The model could be any
supported subclass of ClassficationModel based on args.
"""
if args.modelName == "htm":
# Instantiate the HTM model
model = ClassificationModelHTM(networkConfig=getNetworkConfig(
args.networkConfigPath),
inputFilePath=None,
retina=args.retina,
verbosity=args.verbosity,
numLabels=2,
prepData=False,
modelDir="tempdir")
elif args.modelName == "keywords":
# Instantiate the keywords model
model = ClassificationModelKeywords(verbosity=args.verbosity,
numLabels=2,
k=9,
modelDir="tempdir")
elif args.modelName == "docfp":
# Instantiate the document fingerprint model
model = ClassificationModelDocumentFingerprint(
verbosity=args.verbosity, retina=args.retina, numLabels=2, k=3)
else:
raise RuntimeError("Unknown model type: " + args.modelName)
return model
def testModelSaveAndLoad(self):
# Keywords model uses the base class implementations of save/load methods.
self.modelDir = "poke_model"
model = ClassificationModelKeywords(modelDir=self.modelDir, verbosity=0)
samples = {
0: (["Pickachu"], numpy.array([0, 2, 2])),
1: (["Eevee"], numpy.array([2])),
2: (["Charmander"], numpy.array([0, 1, 1])),
3: (["Abra"], numpy.array([1])),
4: (["Squirtle"], numpy.array([1, 0, 1])),
}
patterns = model.encodeSamples(samples)
for i in xrange(len(samples)):
model.trainModel(i)
output = [model.testModel(i) for i in xrange(len(patterns))]
model.saveModel()
loadedModel = ClassificationModel(verbosity=0).loadModel(self.modelDir)
loadedModelOutput = [loadedModel.testModel(i) for i in xrange(len(patterns))]
for mClasses, lClasses in zip(output, loadedModelOutput):
self.assertSequenceEqual(
mClasses.tolist(),
lClasses.tolist(),
"Output " "classifcations from loaded model don't match original model's.",
)
def testClassifyKeywordsSingleAndMultiClass(self):
"""Tests simple classification with multiple labels for Keywords model."""
model = ClassificationModelKeywords()
samples = {
0: (["Pickachu"], numpy.array([0, 2, 2])),
1: (["Eevee"], numpy.array([2])),
2: (["Charmander"], numpy.array([0, 1, 1])),
3: (["Abra"], numpy.array([1])),
4: (["Squirtle"], numpy.array([1, 0, 1]))
}
patterns = model.encodeSamples(samples)
for i in xrange(len(samples)):
model.trainModel(i)
output = [model.testModel(i) for i in xrange(len(patterns))]
self.assertSequenceEqual(output[0].tolist(), [2, 0],
"Incorrect output for first sample.")
self.assertSequenceEqual(output[1].tolist(), [2],
"Incorrect output for second sample.")
self.assertSequenceEqual(output[2].tolist(), [1, 0],
"Incorrect output for third sample.")
self.assertSequenceEqual(output[3].tolist(), [1],
"Incorrect output for fourth sample.")
# Test the order of class labels doesn't matter when training.
self.assertTrue(numpy.allclose(output[2], output[4]),
"Outputs for samples 2 and 4 should be identical.")
