def do_predict(label, category, csvs_folder_path):
model_path = Utilities.construct_filepath(in_dir, [category, label],
".model")
text_clf_svm = pickle.load(open(model_path, "rb"))
# read in test csv data
test_csv_path = csvs_folder_path + category + const.TEST
csv_helper = CsvHelper()
csv_helper.set_csv(test_csv_path)
csv_helper.add_column(
label
) # create new column with title label (set on top) so that we can fil lin later
test_df = csv_helper.get_csv()
# Do prediction
data_df = test_df["title"]
id_df = test_df["itemid"]
predicted_multiclass = text_clf_svm.predict_proba(data_df)
pred_top2 = get_top_2_dict(text_clf_svm.classes_, predicted_multiclass,
id_df)
# write back to csv
write_to_csv(pred_top2, test_df, test_csv_path
) # top 2 is the top 2 predicted test_df is the original csv
def split(in_dir, out_dir, csv):
if not os.path.exists(out_dir):
os.makedirs(out_dir)
print "created folder :", out_dir
csv_helper = CsvHelper()
csv_helper.set_csv(csv)
print "[INFO] using csv file : ", csv
image_paths_df = csv_helper.get_single_column(IMAGE_PATH)
copy_images(image_paths_df, in_dir, out_dir)
"""
if __name__ == "__main__":
save_model = False
args = Utilities.process_arg("-save")
out_dir = "/content/drive/My Drive/models"
csvs_folder_path = "/content/drive/My Drive/yy_fashion.csv"
category = "fashion"
save = "true"
if save == "t" or save == "T" or save == "true":
save_model = True
train_csv_path = "/content/drive/My Drive/yy_fashion.csv"
csv_helper = CsvHelper()
csv_helper.set_csv(train_csv_path) # set csv file as train_csv_path
print("Set csv done")
'''
This part is for mass training like for each attribute/ label just train a model
for individual model tuning, set labels with a single label , like like 123
'''
csv_helper.set_all_headers_as_label(
) # set all the labels as label headers
labels = csv_helper.get_label_headers(
) # get lables i.e all column name besides image name, itemid and title
#labels = ["Colour_group"] # just doing
print("Predicting These classes : ", labels)
for label in labels:
do_training(label, csv_helper, save_model)
