def main():
parser = create_parser()
opts, args = parser.parse_args()
if len(args) != 3:
parser.error("invalid number of arguments")
training_sqlite3, test_sqlite3, process_language = args
print("Arguments: %s %s %s" %
(training_sqlite3, test_sqlite3, process_language))
#cwd = os.getcwd()
#print "Working directory: %s" % (cwd)
temporary_dir = opts.temporary_dir if opts.temporary_dir else P.dirname(
training_sqlite3)
if not P.isdir(temporary_dir):
parser.error("error: temporary directory %s does not exist" %
temporary_dir)
def log(message):
print("[%s] %s" % (datetime.datetime.now().isoformat(), message))
nlp = spacy.load(process_language)
#
# Training section
#
log("Preparing SQLite training database")
training_sqlite = P.join(temporary_dir, training_sqlite3)
prep_sqlite(training_sqlite)
log("Calculating dimensions")
calc_dim(nlp, training_sqlite, 0, False)
log("Excluding dimensions")
#if process_language == 'en':
# exclude_stopwords(training_sqlite)
#else :
# exclude_stopwords_spacy(training_sqlite, process_language)
exclude_stopwords_spacy(nlp, training_sqlite, process_language)
exclude_non_alpha_partial(training_sqlite)
exclude_unigrams_shorter_than(training_sqlite, 3)
exclude_ngrams_shorter_than(training_sqlite, 1)
log("Pruning excluded dimensions")
prune(training_sqlite)
log("Indexing training database")
index(training_sqlite, nlp)
log("Running mRMR algorithm to select features")
mrmr(training_sqlite, temporary_dir)
log("Pruning excluded dimensions (again)")
prune(training_sqlite)
log("Outputting training samples to temporary data file")
training_samples = P.join(temporary_dir, "training-samples.dat")
svmvec(training_sqlite, training_samples)
log("Training classifier")
classifier = P.join(temporary_dir, "classifier.svm")
learn(training_sqlite, training_samples, classifier)
#
# Test section
#
log("Preparing SQLite test database")
test_sqlite = P.join(temporary_dir, test_sqlite3)
prep_sqlite(test_sqlite)
log("Copying dimensions from training database to test database")
copy_dim(training_sqlite, test_sqlite)
log("Indexing test database")
index(test_sqlite, nlp)
log("Outputting test samples to temporary data file")
test_samples = P.join(temporary_dir, "test-samples.dat")
svmvec(test_sqlite, test_samples)
log("Classifying test samples")
classify(test_sqlite, test_samples, classifier, False, temporary_dir)
def main():
parser = create_parser()
opts, args = parser.parse_args()
if len(args) != 3:
parser.error("invalid number of arguments")
mode, training_mdb, test_mdb = args
if mode not in ("rank", "classify"):
parser.error("invalid mode: %s" % mode)
temporary_dir = opts.temporary_dir if opts.temporary_dir else P.dirname(training_mdb)
if not P.isdir(temporary_dir):
parser.error("error: temporary directory %s does not exist" % `temporary_dir`)
def log(message):
print "[%s] %s" % (datetime.datetime.now().isoformat(), message)
#
# Training section
#
log("Converting training database from MS Access to SQLite")
training_sqlite = P.join(temporary_dir, P.splitext(P.basename(training_mdb))[0]+".sqlite3")
mdb2sqlite(training_mdb, training_sqlite, zero_score=False)
log("Calculating dimensions")
calc_dim(training_sqlite)
log("Excluding dimensions")
exclude_stopwords(training_sqlite)
exclude_non_alpha_partial(training_sqlite)
exclude_shorter_than(training_sqlite, 3)
log("Pruning excluded dimensions")
prune(training_sqlite)
log("Indexing training database")
index(training_sqlite, IndexingOptions())
log("Running mRMR algorithm to select features")
mrmr(training_sqlite, temporary_dir)
log("Pruning excluded dimensions (again)")
prune(training_sqlite)
log("Outputting training samples to temporary data file")
training_samples = P.join(temporary_dir, "training-samples.dat")
svmvec(training_sqlite, training_samples)
log("Selecting best value for C parameter")
best_c = parameter_selection(training_samples)
log("Training classifier")
classifier = P.join(temporary_dir, "classifier.svm.dlib")
trainer(training_samples, classifier, best_c)
#
# Ranking section
#
log("Converting test database from MS Access to SQLite")
test_sqlite = P.join(temporary_dir, P.splitext(P.basename(test_mdb))[0]+".sqlite3")
mdb2sqlite(test_mdb, test_sqlite, zero_score=True)
log("Copying dimensions from training database to test database")
copy_dim(training_sqlite, test_sqlite)
log("Indexing test database")
index(test_sqlite, IndexingOptions())
log("Outputting test samples to temporary data file")
test_samples = P.join(temporary_dir, "test-samples.dat")
svmvec(test_sqlite, test_samples)
if mode == "rank":
log("Ranking test samples")
ranker(training_samples, test_samples, best_c)
elif mode == "classify":
log("Classifying test samples")
dlib_classifier(test_sqlite, test_samples, classifier)
#
# This won't work on OS/X since the ODBC driver is read-only :(
#
log("Copying scores to test database (MS Access)")
copy_scores(test_sqlite, test_mdb)
else:
assert False