def main(): dl = IMDBDataLoader() create_logging() estimator = tf.estimator.Estimator(model_fn) estimator.train(dl.train_input_fn()) y_pred = np.fromiter(estimator.predict(dl.predict_input_fn()), np.int32) tf.logging.info('\n' + classification_report(dl.y_test, y_pred))
def main(): create_logging() tf.logging.info('\n'+pprint.pformat(args.__dict__)) dl = DataLoader() if not os.path.exists(MODEL_PATH): os.makedirs(MODEL_PATH) prev_models = os.listdir(MODEL_PATH) if prev_models is not None: for f in prev_models: os.remove(os.path.join(MODEL_PATH, f)) estimator = tf.estimator.Estimator(model_fn, model_dir=MODEL_PATH, config=tf.estimator.RunConfig(keep_checkpoint_max=1)) y_true = get_val_labels() for _ in range(args.n_epochs): estimator.train(lambda: dl.train_input_fn()) y_pred = list(estimator.predict(lambda: dl.val_input_fn())) tf.logging.info('\nVal Log Loss: %.3f\n' % log_loss( np.asarray(y_true, np.float64), np.asarray(y_pred, np.float64), labels=[0, 1])) submit_arr = np.asarray(list(estimator.predict(lambda: dl.predict_input_fn()))) print(submit_arr.shape) submit = pd.DataFrame() submit['y_pre'] = submit_arr submit.to_csv(SUBMIT_PATH, index=False)
def main(): create_logging() sess = tf.Session() dl = IMDBDataLoader(sess) model = Model(dl) trainer = Trainer(sess, model, dl) trainer.train()
def main(): create_logging() sess = tf.Session() vocab = IMDBVocab() dl = VAEDataLoader(sess, vocab) model = VAE(dl, vocab) tf.logging.info('\n' + pprint.pformat(tf.trainable_variables())) trainer = VAETrainer(sess, model, dl, vocab) trainer.train()
def main(): create_logging() sess = tf.Session() vocab = IMDBVocab() dl = WakeSleepDataLoader(sess, vocab) model = WakeSleepController(dl, vocab) tf.logging.info('\n' + pprint.pformat(tf.trainable_variables())) trainer = WakeSleepTrainer(sess, model, dl, vocab) model.load(sess, args.vae_ckpt_dir) trainer.train()
def main(): create_logging() tf.logging.info('\n' + pprint.pformat(args.__dict__)) sess = tf.Session() vocab = IMDBVocab() discri_dl = DiscriminatorDataLoader(sess, vocab) wake_sleep_dl = WakeSleepDataLoader(sess, vocab) model = WakeSleepController(discri_dl, wake_sleep_dl, vocab) tf.logging.info('\n' + pprint.pformat(tf.trainable_variables())) trainer = WakeSleepTrainer(sess, model, discri_dl, wake_sleep_dl, vocab) model.load(sess, args.vae_ckpt_dir) trainer.train()
def main(): create_logging() dl = DataLoader() estimator = tf.estimator.Estimator(model_fn) y_true = get_val_labels() for _ in range(args.n_epochs): estimator.train(lambda: dl.train_input_fn()) y_pred = list(estimator.predict(lambda: dl.val_input_fn())) tf.logging.info('\nVal Log Loss: %.3f\n' % log_loss(y_true, y_pred, eps=1e-15)) submit_arr = np.asarray( list(estimator.predict(lambda: dl.predict_input_fn()))) print(submit_arr.shape) submit = pd.DataFrame() submit['y_pre'] = submit_arr submit.to_csv('./submit_siamese_rnn.csv', index=False)