Training is very slow

$python examples/demo_mnist.py Using TensorFlow backend. [ 2018-12-14 13:34:10,577][cascade_classifier.fit_transform] X_groups_train.shape=[(60000, 1, 28, 28)],y_train.shape=(60000,),X_groups_test.shape=no_test,y_test.shape=no_test [ 2018-12-14 13:34:10,588][cascade_classifier.fit_transform] group_dims=[784] [ 2018-12-14 13:34:10,588][cascade_classifier.fit_transform] group_starts=[0] [ 2018-12-14 13:34:10,588][cascade_classifier.fit_transform] group_ends=[784] [ 2018-12-14 13:34:10,588][cascade_classifier.fit_transform] X_train.shape=(60000, 784),X_test.shape=(0, 784) [ 2018-12-14 13:34:10,645][cascade_classifier.fit_transform] [layer=0] look_indexs=[0], X_cur_train.shape=(60000, 784), X_cur_test.shape=(0, 784) [ 2018-12-14 13:34:27,575][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_0 - 5_folds.train_0.predict)=90.02% [ 2018-12-14 13:34:41,561][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_0 - 5_folds.train_1.predict)=90.07% [ 2018-12-14 13:34:55,516][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_0 - 5_folds.train_2.predict)=90.22% [ 2018-12-14 13:35:09,470][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_0 - 5_folds.train_3.predict)=90.11% [ 2018-12-14 13:35:23,300][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_0 - 5_folds.train_4.predict)=89.24% [ 2018-12-14 13:35:23,303][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_0 - 5_folds.train_cv.predict)=89.93% [ 2018-12-14 13:35:24,074][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_1 - 5_folds.train_0.predict)=94.64% [ 2018-12-14 13:35:24,841][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_1 - 5_folds.train_1.predict)=94.12% [ 2018-12-14 13:35:25,624][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_1 - 5_folds.train_2.predict)=93.92% [ 2018-12-14 13:35:26,382][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_1 - 5_folds.train_3.predict)=94.61% [ 2018-12-14 13:35:27,138][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_1 - 5_folds.train_4.predict)=94.35% [ 2018-12-14 13:35:27,144][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_1 - 5_folds.train_cv.predict)=94.33% [ 2018-12-14 13:35:27,924][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_2 - 5_folds.train_0.predict)=94.49% [ 2018-12-14 13:35:28,705][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_2 - 5_folds.train_1.predict)=94.85% [ 2018-12-14 13:35:29,484][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_2 - 5_folds.train_2.predict)=94.59% [ 2018-12-14 13:35:30,255][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_2 - 5_folds.train_3.predict)=94.96% [ 2018-12-14 13:35:31,042][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_2 - 5_folds.train_4.predict)=95.02% [ 2018-12-14 13:35:31,048][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_2 - 5_folds.train_cv.predict)=94.78% [ 2018-12-14 14:12:52,985][kfold_wrapper.log_eval_metrics] Accuracy(layer_0 - estimator_3 - 5_folds.train_0.predict)=91.26%

estimator0~estimator2 is fast， but estimator3 is very slow，how to fixed this problem？