def _runWekaCommand(self, command, verbose=False):
"""Set the CLASSPATH necessary to use Weka, then runs a shell `command`."""
if not settings.WEKA_PATH[self.WEKA_VERSION]:
raise ImproperlyConfigured(
"'WEKA_PATH' is not set in settings.py!")
set_path = "export CLASSPATH=$CLASSPATH:{}; ".format(
settings.WEKA_PATH[self.WEKA_VERSION])
command = set_path + command
logger.debug("Running in Shell:\n{}".format(command))
if verbose:
logger.info("Running in Shell:\n{}".format(command))
subprocess.check_output(command, shell=True)
def _prepareArff(self, reactions, whitelistHeaders, verbose=False):
"""Write an *.arff file using the provided queryset of reactions."""
logger.debug("Preparing ARFF file...")
filename = "{}_{}.arff".format(self.statsModel.pk, uuid.uuid4())
filepath = os.path.join(settings.TMP_DIR, filename)
# uber paranoid making sure we don't race condition
while os.path.isfile(filepath):
filename = "{}_{}.arff".format(self.statsModel.pk, uuid.uuid4())
filepath = os.path.join(settings.TMP_DIR, filename)
if verbose:
logger.info("Writing arff to {}".format(filepath))
with open(filepath, "w") as f:
reactions.toArff(f,
expanded=True,
whitelistHeaders=whitelistHeaders)
return filepath
def train(self, verbose=False):
"""Train the weka model."""
reactions = self.statsModel.trainingSet.reactions.all()
descriptorHeaders = [
d.csvHeader
for d in chain(self.statsModel.container.descriptors,
self.statsModel.container.outcomeDescriptors)
]
filePath = self.statsModel.outputFile.name
if not self.statsModel.inputFile.name:
self.statsModel.inputFile = self._prepareArff(
reactions, descriptorHeaders, verbose)
self.statsModel.save(update_fields=['inputFile'])
elif not os.path.isfile(self.statsModel.inputFile.name):
if self.invalid:
raise RuntimeError(
'Could not find statsModel arff file and model is invalid')
else:
raise warning.warn(
'Could not find statsModel arff file, but model is valid, so recreating'
)
self.statsModel.inputFile.name = self._prepareArff(
reactions, descriptorHeaders, verbose)
self.statsModel.save(update_fields=['inputFile'])
elif verbose:
logger.info("Using existing arff file.")
arff_file = self.statsModel.inputFile.name
# Currently, we support only one "response" variable.
response = list(self.statsModel.container.outcomeDescriptors)[0]
headers = [
h for h in reactions.expandedCsvHeaders() if h in descriptorHeaders
]
response_index = headers.index(response.csvHeader) + 1
if self.BCR:
cost_matrix_string = self.BCR_cost_matrix(reactions, response)
command = "java weka.classifiers.meta.CostSensitiveClassifier -cost-matrix {} -W {} -t {} -d {} -p 0 -c {} -- {}".format(
cost_matrix_string, self.wekaCommand, arff_file, filePath,
response_index, self.wekaTrainOptions)
else:
command = "java {} -t {} -d {} -p 0 -c {} {}".format(
self.wekaCommand, arff_file, filePath, response_index,
self.wekaTrainOptions)
self._runWekaCommand(command, verbose=verbose)
def predict(self, reactions, verbose=False):
"""Create the predictions for these reactions for the model."""
descriptorHeaders = [
d.csvHeader
for d in chain(self.statsModel.container.descriptors,
self.statsModel.container.outcomeDescriptors)
]
arff_file = self._prepareArff(reactions,
descriptorHeaders,
verbose=verbose)
model_file = self.statsModel.outputFile.name
results_file = "{}_{}.out".format(self.statsModel.pk, uuid.uuid4())
results_path = os.path.join(settings.TMP_DIR, results_file)
# Currently, we support only one "response" variable.
headers = [
h for h in reactions.expandedCsvHeaders() if h in descriptorHeaders
]
response = list(self.statsModel.container.outcomeDescriptors)[0]
response_index = headers.index(response.csvHeader) + 1
command = "java {} -T {} -l {} -p 0 -c {} 1> {}".format(
self.wekaCommand, arff_file, model_file, response_index,
results_path)
if verbose:
logger.info("Writing results to {}".format(results_path))
self._runWekaCommand(command, verbose=verbose)
if isinstance(response, rxnDescriptors.BoolRxnDescriptor):
typeConversionFunction = booleanConversion
elif isinstance(response, rxnDescriptors.OrdRxnDescriptor):
typeConversionFunction = ordConversion
elif isinstance(response, rxnDescriptors.NumRxnDescriptor):
typeConversionFunction = numConversion
elif isinstance(response, rxnDescriptors.CatRxnDescriptor):
typeConversionFunction = str
else:
raise TypeError("Response descriptor is of invalid type {}".format(
type(response)))
results = tuple((reaction, result) for reaction, result in zip(
reactions,
self._readWekaOutputFile(results_path, typeConversionFunction)))
return {response: results}