def read_train(f):
"""
Reads train data
:param f: csv file containing rows of image number - whale type pairs
:rtype : numpy array of image number - whale id pairs and whale types numpy array
"""
reader = csv.reader(f)
header = next(reader)
assert (header[0], header[1]) == ('Image', 'whaleID')
whale_ids = set()
image_ids_whale_ids = []
for image_name, whale_name in reader:
if not (image_name.startswith('w_') and image_name.endswith('.jpg')):
continue
assert whale_name.startswith('whale_')
image_id = ImagesReader.get_image_id(image_name)
whale_id = whale_name.split("_")[1]
whale_ids.add(whale_id)
image_ids_whale_ids.append((image_id, whale_id))
return np.array(image_ids_whale_ids), whale_ids
def main():
with open(TRAIN_FILENAME) as train_data_file:
image_ids_whale_ids, whale_ids = utility.read_train(train_data_file)
images_reader = ImagesReader(IMAGES_DIR)
images_reader.pre_process(image_processors.region_crop_gray_downscale, rewrite=False, threads=1)
all_train_images_ids = image_ids_whale_ids[:, 0]
unique_train_images_ids = set(all_train_images_ids)
all_images_ids = set(images_reader.image_ids)
result_images_ids = all_images_ids.difference(unique_train_images_ids)
train_image_id_whale_id = dict(image_ids_whale_ids)
print('Reading train data\n')
x_train = np.asarray([images_reader.read_image_vector(image_id)
for image_id in all_train_images_ids])
y_train = np.asarray([train_image_id_whale_id[image_id] for image_id in all_train_images_ids])
features_cnt = len(x_train[0])
num_targets = len(set(y_train))
clf = simple_cnn.CNN(features_cnt, num_targets,
num_epochs=10,
fresh_start=False,
dump_dir="network_weights/",
filename_to_dump="net.w")
# clf = SVC(probability=True)
print('Fitting\n')
clf.fit(x_train, y_train)
print('Reading test data\n')
x_test = np.array([images_reader.read_image_vector(image_id)
for image_id in result_images_ids])
print('Predicting\n')
y_predicted = clf.predict_proba(x_test)
print('Writing submission')
with open(SUBMISSION_FILENAME, 'w') as submission_file:
utility.write_submission(whale_ids, result_images_ids, y_predicted, submission_file)
return 0
def write_submission(whale_ids, image_ids, whale_probs, submission_file):
"""
Writes image_ids_whale_probabilities to submission_file
:param submission_file: file to write the submission to
:param whale_types: array of whale types to be written to csv header
:param image_ids_whale_probabilities: array of pairs image_id - whale probabilities array
"""
assert len(image_ids) == 6925 and len(whale_probs) == 6925
submission_file.write(",".join(["Image"] + list('whale_'+str(whale_id) for whale_id in sorted(whale_ids))))
submission_file.write("\n")
for image_id, whale_probs in zip(image_ids, whale_probs):
submission_file.write(",".join([ImagesReader.get_image_name(image_id)] + list(str(prob) for prob in whale_probs)))
submission_file.write("\n")
