submit.py

import ast
import datetime as dt
import os
import time

import cv2
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import seaborn as sns
import tensorflow as tf
from keras.models import Model, load_model
from tensorflow import keras

from common import *

# os.environ['CUDA_VISIBLE_DEVICES'] = '-1'

start = dt.datetime.now()

model_path = './model/3/weights-001-0.942.hdf5'

print('Loading model at', model_path)

# Load previous model
trained_model = load_model(model_path, custom_objects={
    'top_3_accuracy': top_3_accuracy})

# # TTA hflip
# model = TTA_ModelWrapper(trained_model)

print('Loaded model. Predicting')
test = pd.read_csv(os.path.join(INPUT_DIR, 'test_simplified.csv'))


max_load_step = int(test.shape[0] / LOAD_SIZE) + 1
#max_load_step = 4

cats = list_all_categories()

id2cat = {k: cat.replace(' ', '_') for k, cat in enumerate(cats)}

test_predictions = None


for load_step in range(max_load_step):
    x_test = df_to_image_array_xd(test, size, load_step)
    new_predictions = model.predict(x_test, batch_size=128, verbose=1)
    test_predictions = new_predictions if test_predictions is None else np.concatenate(
        (test_predictions, new_predictions))

top3 = preds2catids(test_predictions)
top3cats = top3.replace(id2cat)

valid_df = pd.read_csv(os.path.join(
    DP_DIR, 'train_k{}.csv.gz'.format(NCSVS - 1)))


map3 = mapk(valid_df[['y']].values, top3.values)
print('Map3: {:.3f}'.format(map3))


test['word'] = top3cats['a'] + ' ' + top3cats['b'] + ' ' + top3cats['c']
submission = test[['key_id', 'word']]
submission.to_csv('gs_mn_submission_{}.csv'.format(
    int(map3 * 10**4)), index=False)


end = dt.datetime.now()
print('Latest run {}.\nTotal time {}s'.format(end, (end - start).seconds))