Just fyi, added skip-troughs master branch and ran the trainer on python 2.7 on ubuntu 17.04 Got an error 'SequenceTransformer instance has no attribute 'filtered_ent_counts'. Did not have time to look into it but it seems the example is not working out of the box. Maybe not enough example data?

python train_generation.py --train_seqs example_stories.csv --save_filepath example_model/

Using Theano backend. loaded transformer with 16 words from example_model//transformer.pkl Created model RNNLM : {'n_timesteps': 15, 'verbose': 1, 'sample_words': None, 'n_pos_nodes': 100, 'n_pos_tags': 59, 'filepath': 'example_model/', 'n_feature_nodes': 100, 'batch_size': 25, 'lr': 0.001, 'n_hidden_nodes': 500, 'decay': 1e-06, 'n_pos_embedding_nodes': 25, 'embeddings': None, 'n_hidden_layers': 2, 'lexicon_size': None, 'clipvalue': 5.0, 'n_embedding_nodes': 300, 'optimizer': 'Adam', 'use_features': False, 'use_pos': False} training epoch 1/10... training RNNLM on 10 sequences with batch size = 25 Saved RNNLM to example_model/ loss: 1.945 (0.543m) generating sequences... created predictor model decoding generated sequences... Traceback (most recent call last): File "train_generation.py", line 73, in train_model(train_seqs_file=args.train_seqs, model=model, n_epochs=args.n_epochs) File "train_generation.py", line 41, in train_model temp=1.0, n_sents_per_seq=1, detokenize=True, adapt_ents=True) File "/home/a/narrative-prediction-master/models/pipeline.py", line 68, in predict capitalize_ents=capitalize_ents, adapt_ents=adapt_ents) File "/home/a/narrative-prediction-master/models/transformer.py", line 557, in decode_num_seqs seq = adapt_tok_seq_ents(seq, ents=ents[seq_idx], sub_ent_probs=ent_counts_to_probs(self.filtered_ent_counts)) AttributeError: SequenceTransformer instance has no attribute 'filtered_ent_counts'

roemmele / narrative-prediction

SequenceTransformer instance has no attribute 'filtered_ent_counts' #1

python train_generation.py --train_seqs example_stories.csv --save_filepath example_model/