def create_config(): c = Config("", "-m", "test") c.update({"verbose": 2, "timeout": 1, "embedding_layer_dim": 1, "ner_dim": 1, "action_dim": 1, "lemma_dim": 1, "max_words_external": 3, "word_dim_external": 1, "word_dim": 1, "max_words": 3, "max_lemmas": 3, "max_tags": 3, "max_pos": 3, "max_deps": 3, "max_edge_labels": 3, "max_puncts": 3, "max_action_types": 3, "max_ner_types": 3, "edge_label_dim": 1, "tag_dim": 1, "pos_dim": 1, "dep_dim": 1, "optimizer": "sgd", "output_dim": 1, "layer_dim": 2, "layers": 3, "lstm_layer_dim": 2, "lstm_layers": 3, "max_action_ratio": 10, "update_word_vectors": False, "copy_shared": None}) c.update_hyperparams(shared={"lstm_layer_dim": 2, "lstm_layers": 1}, ucca={"word_dim": 2}, amr={"max_node_labels": 3, "max_node_categories": 3, "node_label_dim": 1, "node_category_dim": 1}) return c
def __init__(self): config = Config() setting = Settings(*('implicit')) config.update(setting.dict()) config.set_format("ucca") self.feature_extractor = DenseFeatureExtractor( OrderedDict(), indexed=config.args.classifier != 'mlp', hierarchical=False, node_dropout=config.args.node_dropout, omit_features=config.args.omit_features) self.sess = tf.Session() saver = tf.train.import_meta_graph(glob('env_r_model-*.meta')[0]) saver.restore(self.sess, tf.train.latest_checkpoint('./')) graph = tf.get_default_graph() self.x = graph.get_tensor_by_name("Placeholder:0") self.y = graph.get_tensor_by_name("dense_2/BiasAdd:0") self.length = None
def config(): c = Config("", "-m", "test") c.update({"no_node_labels": True, "evaluate": True, "minibatch_size": 50}) c.update_hyperparams(shared={"layer_dim": 50}) return c
def produce_oracle(filename, feature_extractor): passage = load_passage(filename) sys.stdout.write('.') sys.stdout.flush() #store_sequence_to = "data/oracles/%s/%s.txt" % (cat, basename(filename))#, setting.suffix()) #with open(store_sequence_to, "w", encoding="utf-8") as f: # for i, action in enumerate(gen_actions(passage, feature_extractor)): # pass#print(action, file=f) for _ in gen_actions(passage, feature_extractor): pass if __name__ == "__main__": config = Config() setting = Settings(*('implicit')) config.update(setting.dict()) config.set_format("ucca") feature_extractor = DenseFeatureExtractor( OrderedDict(), indexed=config.args.classifier != 'mlp', hierarchical=False, node_dropout=config.args.node_dropout, omit_features=config.args.omit_features) filenames = passage_files(sys.argv[1]) for filename in filenames: #TODO: solve the problem of "KILLED" while wring file. Use 100 files temporarily before solving this. produce_oracle(filename, feature_extractor) # dump envTrainingData to a file for further learning in rewardNN.py json_str = json.dumps(envTrainingData) + "\n" json_bytes = json_str.encode('utf-8')