from keras.layers.recurrent import LSTM from keras.layers import activations, Wrapper from keras.layers import Input, Embedding, Flatten, Dropout, Lambda, concatenate, Dense if __name__ == "__main__": os.environ["CUDA_VISIBLE_DEVICES"] = "0" config = OrderedDict() config.MAX_WINDOW_SIZE = 10 config.MAX_MENTION_LENGTH = 10 config.EMBEDDING_TRAINABLE = False config.WORD_EMBEDDING_DIM = 300 #first #config.ENTITY_EMBEDDING_DIM = 300 #second config.MAX_ENTITY_DESC_LENGTH = 150 #no config.MENTION_CONTEXT_LATENT_SIZE = 50 config.LSTM_SIZE = 300 config.DROPOUT = 0.3 config.ACTIVATION_FUNCTION = 'tanh' config.batch_size = 1024 config.num_of_neg = 1 #the number of negative sample of each senetence config.start_epochs = 0 # the epoch of start config.epochs = 5 #the number of iteration config.batch_epochs = 1 #the number of batch of evaluate times # reforcement learning config config.updaterate = 1 config.num_epoch = 5 config.sampletimes = 1 config.negative_sample = 5 config.context_length = config.MAX_WINDOW_SIZE + config.MAX_MENTION_LENGTH
from keras.layers import activations, Wrapper from keras.layers import Input, Embedding, Flatten, Dropout, Lambda, concatenate, Dense if __name__ == "__main__": os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID" os.environ["CUDA_VISIBLE_DEVICES"] = "1" config = OrderedDict() config.MAX_WINDOW_SIZE = 15 config.MAX_MENTION_LENGTH = 5 config.EMBEDDING_TRAINABLE = False config.WORD_EMBEDDING_DIM = 100 #first config.ENTITY_EMBEDDING_DIM = 100 #second config.MAX_ENTITY_DESC_LENGTH = 100 #no config.MENTION_CONTEXT_LATENT_SIZE = 200 config.LSTM_SIZE = 100 config.DROPOUT = 0.3 config.ACTIVATION_FUNCTION = 'tanh' context_length = config.MAX_WINDOW_SIZE + config.MAX_MENTION_LENGTH batch_size = 256 start_epochs = 0 epochs = 35 batch_epochs = 5 word_index, entity_indices, word_ebd, entity_ebd = tools.load_matrices() save_path = './model/origin_rl_model.ckpt' save_path2 = './model/origin_rl_entity_model.ckpt' test_dataset_2013 = './data/2013_prepare_filled.txt'