def testCreateModel(self):
# Setup fake ImbuModels instance
imbu = ImbuModels(
cacheRoot="fake_cache_root",
dataPath=self.dataPath,
retina="en_associative",
apiKey=os.environ.get("CORTICAL_API_KEY")
)
# You must specify a known model type
self.assertRaises(TypeError, imbu.createModel, "random model type")
# Assert that you can create models of known types from scratch (i.e. with
# empty loadPath value)
for modelType in imbu.modelMappings:
# You _must_ specify loadPath and savePath. Assert that attempts to not
# specify will result in failure
self.assertRaises(TypeError, imbu.createModel, modelType)
self.assertRaises(TypeError, imbu.createModel, modelType, None)
self.assertRaises(ValueError, imbu.createModel, modelType, None, None)
# Attempt to create model using default arguments
model = imbu.createModel(
modelType, "", "", networkConfigName="imbu_sensor_knn.json"
)
# Assert model created was one of the expected types
self.assertTrue(isinstance(model, ClassificationModel) or
isinstance(model, ClassificationNetworkAPI))
def testCreateModel(self):
# Setup fake ImbuModels instance
imbu = ImbuModels(cacheRoot="fake_cache_root",
dataPath=self.dataPath,
retina="en_associative",
apiKey=os.environ.get("CORTICAL_API_KEY"))
# You must specify a known model type
self.assertRaises(TypeError, imbu.createModel, "random model type")
# Assert that you can create models of known types from scratch (i.e. with
# empty loadPath value)
for modelType in imbu.modelMappings:
# You _must_ specify loadPath and savePath. Assert that attempts to not
# specify will result in failure
self.assertRaises(TypeError, imbu.createModel, modelType)
self.assertRaises(TypeError, imbu.createModel, modelType, None)
self.assertRaises(ValueError, imbu.createModel, modelType, None,
None)
# Attempt to create model using default arguments
model = imbu.createModel(modelType,
"",
"",
networkConfigName="imbu_sensor_knn.json")
# Assert model created was one of the expected types
self.assertTrue(
isinstance(model, ClassificationModel)
or isinstance(model, ClassificationNetworkAPI))
def testMappingModelNamesToModelTypes(self):
imbu = ImbuModels(dataPath=self.dataPath)
for modelName, mappedName in imbu.modelMappings.iteritems():
self.assertEquals(mappedName, imbu._mapModelName(modelName),
"Incorrect mapping returned for model named '{}'".format(modelName))
self.assertRaises(ValueError, imbu._mapModelName, "fakeModel")
def testMappingModelNamesToModelTypes(self):
imbu = ImbuModels(cacheRoot="fake_cache_root", dataPath=self.dataPath)
for modelName, mappedName in imbu.modelMappings.iteritems():
self.assertEquals(
mappedName, imbu._mapModelName(modelName),
"Incorrect mapping returned for model named '{}'".format(
modelName))
self.assertRaises(ValueError, imbu._mapModelName, "fakeModel")
def testCacheDirProperty(self):
imbu = ImbuModels(dataPath=self.dataPath)
checkpointLocation = self._createTempModelCheckpoint()
model = imbu.createModel("HTMNetwork",
loadPath="",
savePath=checkpointLocation,
networkConfigName="imbu_sensor_knn.json")
# Test for default cache directory
encoder = model.getEncoder()
defaultCacheLocation = "nupic.research/htmresearch/encoders/CioCache"
self.assertIn(
defaultCacheLocation, getattr(encoder.client, "cacheDir"),
"Cio encoder cache dir is not the expected default location.")
# Now explicitly set the cache directory
encoder.cacheDir = "fake_cache_root"
self.assertEqual("fake_cache_root", getattr(encoder.client,
"cacheDir"),
"Cio encoder cache dir did not set properly.")
def _exerciseModelLifecycle(self, modelType, queryTerm="unicorn",
networkConfigName="imbu_sensor_knn.json"):
# Setup fake ImbuModels instance
imbu = ImbuModels(
cacheRoot="fake_cache_root",
dataPath=self.dataPath,
retina="en_associative",
apiKey=os.environ.get("CORTICAL_API_KEY")
)
tmpDir = tempfile.mkdtemp()
self.addCleanup(shutil.rmtree, tmpDir)
checkpointLocation = os.path.join(tmpDir, "checkpoint")
# Create model with no load path. This should trigger the creation of a
# new intance, and train it. Specify save path so that the trained model
# gets saved
originalModel = imbu.createModel(modelType,
loadPath="",
savePath=checkpointLocation,
networkConfigName=networkConfigName)
# Make a query, keep around for later comparison
originalQueryResults = imbu.query(originalModel, queryTerm)
del originalModel
# Create model, specifying previous load path
model = imbu.createModel(modelType,
loadPath=checkpointLocation,
savePath=checkpointLocation,
networkConfigName=networkConfigName)
self.assertTrue(all(numpy.array_equal(original, new)
for (original, new)
in zip(originalQueryResults,
imbu.query(model, queryTerm))))
def testCacheSaveAndReload(self):
imbuTempDir = self._createTempDir()
originalCacheDir = os.path.join(imbuTempDir, "original_test_cache")
imbu = ImbuModels(cacheRoot=originalCacheDir,
dataPath=self.dataPath,
retina="en_associative",
apiKey=os.environ.get("CORTICAL_API_KEY"))
# Train with standard cache location, save
modelType = "HTMNetwork"
networkConfigName = "imbu_sensor_knn.json"
checkpointLocation = os.path.join(imbuTempDir, "checkpoint")
originalModel = imbu.createModel(modelType,
loadPath="",
savePath=checkpointLocation,
networkConfigName=networkConfigName)
# Physically move the cache to a new directory; we copy it b/c moving it
# would break the test cleanup method
encoder = originalModel.getEncoder()
newCacheDir = os.path.join(imbuTempDir, "new_test_cache")
shutil.copytree(originalCacheDir, newCacheDir)
self.addCleanup(shutil.rmtree, newCacheDir)
self.assertGreater(len(os.listdir(newCacheDir)), 0,
"The new cache directory is empty!")
# Load a new model and set cache location to the new directory
newModel = imbu.createModel(modelType,
loadPath=checkpointLocation,
savePath=checkpointLocation,
networkConfigName=networkConfigName)
newEncoder = newModel.getEncoder()
newEncoder.cacheDir = newCacheDir
newEncoderCacheDir = getattr(newEncoder, "cacheDir")
self.assertNotEquals(
newEncoderCacheDir, getattr(encoder, "cacheDir"),
"Old and new cache locations shouldn't be the same.")
del originalModel
# Run inference with old data, expecting no new caching
sizeOfCache = len(os.listdir(newEncoderCacheDir))
imbu.query(newModel, "unicorn")
self.assertEquals(sizeOfCache, len(os.listdir(newEncoderCacheDir)), "")
# Run inference with new data, adding to the new cache location
imbu.query(newModel, "brains")
self.assertEquals(sizeOfCache + 1, len(os.listdir(newEncoderCacheDir)),
"")
def testResultsFormatting(self):
imbu = ImbuModels(cacheRoot="fake_cache_root", dataPath=self.dataPath)
query = "Hello world!"
distanceArray = numpy.random.rand(35)
idList = range(27) + [x+1000 for x in xrange(8)]
# Test a model with windows
modelName = "HTM_sensor_simple_tp_knn"
results = imbu.formatResults(modelName, query, distanceArray, idList)
self.checkResultsFormatting(results, modelName, windowSize=10)
# Test a word-level model
modelName = "HTM_sensor_knn"
results = imbu.formatResults(modelName, query, distanceArray, idList)
self.checkResultsFormatting(results, modelName, windowSize=1)
# Test a document-level model
modelName = "CioDocumentFingerprint"
distanceArray = numpy.random.rand(2)
idList = range(1)
results = imbu.formatResults(modelName, query, distanceArray, idList)
self.checkResultsFormatting(results, modelName)
def testResultsFormatting(self):
imbu = ImbuModels(cacheRoot="fake_cache_root", dataPath=self.dataPath)
query = "Hello world!"
distanceArray = numpy.random.rand(35)
idList = range(27) + [x + 1000 for x in xrange(8)]
# Test a model with windows
modelName = "HTM_sensor_simple_tp_knn"
results = imbu.formatResults(modelName, query, distanceArray, idList)
self.checkResultsFormatting(results, modelName, windowSize=10)
# Test a word-level model
modelName = "HTM_sensor_knn"
results = imbu.formatResults(modelName, query, distanceArray, idList)
self.checkResultsFormatting(results, modelName, windowSize=1)
# Test a document-level model
modelName = "CioDocumentFingerprint"
distanceArray = numpy.random.rand(2)
idList = range(1)
results = imbu.formatResults(modelName, query, distanceArray, idList)
self.checkResultsFormatting(results, modelName)
def testCacheDirProperty(self):
imbu = ImbuModels(dataPath=self.dataPath)
checkpointLocation = self._createTempModelCheckpoint()
model = imbu.createModel("HTMNetwork",
loadPath="",
savePath=checkpointLocation,
networkConfigName="imbu_sensor_knn.json")
# Test for default cache directory
encoder = model.getEncoder()
defaultCacheLocation = "nupic.research/htmresearch/encoders/CioCache"
self.assertIn(
defaultCacheLocation,
getattr(encoder.client, "cacheDir"),
"Cio encoder cache dir is not the expected default location.")
# Now explicitly set the cache directory
encoder.cacheDir = "fake_cache_root"
self.assertEquals(
"fake_cache_root",
getattr(encoder.client, "cacheDir"),
"Cio encoder cache dir did not set properly.")
def testCacheSaveAndReload(self):
imbuTempDir = self._createTempDir()
originalCacheDir = os.path.join(imbuTempDir, "original_test_cache")
imbu = ImbuModels(
cacheRoot=originalCacheDir,
dataPath=self.dataPath,
retina="en_associative",
apiKey=os.environ.get("CORTICAL_API_KEY")
)
# Train with standard cache location, save
modelType = "HTMNetwork"
networkConfigName = "imbu_sensor_knn.json"
checkpointLocation = os.path.join(imbuTempDir, "checkpoint")
originalModel = imbu.createModel(modelType,
loadPath="",
savePath=checkpointLocation,
networkConfigName=networkConfigName)
# Physically move the cache to a new directory; we copy it b/c moving it
# would break the test cleanup method
encoder = originalModel.getEncoder()
newCacheDir = os.path.join(imbuTempDir, "new_test_cache")
shutil.copytree(originalCacheDir, newCacheDir)
self.addCleanup(shutil.rmtree, newCacheDir)
self.assertGreater(len(os.listdir(newCacheDir)), 0,
"The new cache directory is empty!")
# Load a new model and set cache location to the new directory
newModel = imbu.createModel(modelType,
loadPath=checkpointLocation,
savePath=checkpointLocation,
networkConfigName=networkConfigName)
newEncoder = newModel.getEncoder()
newEncoder.cacheDir = newCacheDir
newEncoderCacheDir = getattr(newEncoder, "cacheDir")
self.assertNotEquals(
newEncoderCacheDir,
getattr(encoder, "cacheDir"),
"Old and new cache locations shouldn't be the same.")
del originalModel
# Run inference with old data, expecting no new caching
sizeOfCache = len(os.listdir(newEncoderCacheDir))
imbu.query(newModel, "unicorn")
self.assertEquals(sizeOfCache, len(os.listdir(newEncoderCacheDir)), "")
# Run inference with new data, adding to the new cache location
imbu.query(newModel, "brains")
self.assertEquals(sizeOfCache + 1, len(os.listdir(newEncoderCacheDir)), "")
def _exerciseModelLifecycle(self,
modelType,
queryTerm="food",
networkConfigName="imbu_sensor_knn.json"):
# Setup fake ImbuModels instance
imbu = ImbuModels(cacheRoot="fake_cache_root",
dataPath=self.dataPath,
retina="en_associative",
apiKey=os.environ.get("CORTICAL_API_KEY"))
tmpDir = tempfile.mkdtemp()
self.addCleanup(shutil.rmtree, tmpDir)
checkpointLocation = os.path.join(tmpDir, "checkpoint")
# Create model with no load path. This should trigger the creation of a
# new intance, and train it. Specify save path so that the trained model
# gets saved
originalModel = imbu.createModel(modelType,
loadPath="",
savePath=checkpointLocation,
networkConfigName=networkConfigName)
# Make a query, keep around for later comparison
originalQueryResults = imbu.query(originalModel, queryTerm)
del originalModel
# Create model, specifying previous load path
model = imbu.createModel(modelType,
loadPath=checkpointLocation,
savePath=checkpointLocation,
networkConfigName=networkConfigName)
self.assertTrue(
all(
numpy.array_equal(original, new) for (original, new) in zip(
originalQueryResults, imbu.query(model, queryTerm))))
# There is no longer training in imbu web app. User must specify loadPath
if "IMBU_LOAD_PATH_PREFIX" in os.environ:
_IMBU_LOAD_PATH_PREFIX = os.environ["IMBU_LOAD_PATH_PREFIX"]
else:
raise KeyError("Required IMBU_LOAD_PATH_PREFIX missing from environment")
g_models = {} # Global model cache
imbu = ImbuModels(
cacheRoot=os.environ.get("MODEL_CACHE_DIR", os.getcwd()),
modelSimilarityMetric=None,
dataPath=os.environ.get("IMBU_DATA",
pkg_resources.resource_filename(__name__,
"data.csv")),
retina=os.environ["IMBU_RETINA_ID"],
apiKey=os.environ["CORTICAL_API_KEY"]
)
def addStandardHeaders(contentType="application/json; charset=UTF-8"):
"""
Add Standard HTTP Headers ("Content-Type", "Server") to the response.
Here is an example of the headers added by this method using the default
values::
Content-Type: application/json; charset=UTF-8
Server: Imbu x.y.z
:param content_type: The value for the "Content-Type" header.
(default "application/json; charset=UTF-8")
# No training in Imbu web app, user must specify loadPath
if "IMBU_LOAD_PATH_PREFIX" in os.environ:
_IMBU_LOAD_PATH_PREFIX = os.environ["IMBU_LOAD_PATH_PREFIX"]
else:
raise KeyError("Required IMBU_LOAD_PATH_PREFIX missing from environment")
g_imbus = {} # Global ImbuModels cache
g_models = {} # Global NLP model cache
for datasetName in os.listdir(_IMBU_LOAD_PATH_PREFIX):
datasetPath = os.path.join(_IMBU_LOAD_PATH_PREFIX, datasetName)
if os.path.isdir(datasetPath) and "egg" not in datasetPath:
# Create an imbu instance for each dataset
imbu = ImbuModels(cacheRoot=os.environ.get("MODEL_CACHE_DIR",
os.getcwd()),
modelSimilarityMetric=None,
dataPath=os.path.join(datasetPath, "data.csv"),
retina=os.environ["IMBU_RETINA_ID"],
apiKey=os.environ["CORTICAL_API_KEY"])
g_imbus.update(((datasetName, imbu), ))
# Init the dict for this dataset's models
g_models[datasetName] = {}
def addStandardHeaders(contentType="application/json; charset=UTF-8"):
"""
Add Standard HTTP Headers ("Content-Type", "Server") to the response.
Here is an example of the headers added by this method using the default
values::
Content-Type: application/json; charset=UTF-8
Server: Imbu x.y.z
:param content_type: The value for the "Content-Type" header.
def _setupFakeImbuModelsInstance(self, retina="en_associative"):
return ImbuModels(cacheRoot="fake_cache_root",
dataPath=self.dataPath,
retina=retina,
apiKey=os.environ.get("CORTICAL_API_KEY"))
