filenames = os.listdir(CONFIGS_MODEL_DIR)
filepaths = [
os.path.join(CONFIGS_MODEL_DIR, filename) for filename in filenames
]
mod_times = [(os.path.getmtime(x), i) for i, x in enumerate(filepaths)]
cfg_path = filepaths[max(mod_times)[1]]
print("Using config from: {}".format(cfg_path))
cfg = yaml.safe_load(open(cfg_path, "r"))
print("Config:")
[print("\t{}: {}".format(k, v)) for k, v in cfg.items()]
print("*******")
texts, das = get_multi_reference_training_variables()
da_embedder = DAEmbeddingSeq2SeqExtractor(das)
# This is a very lazy move
texts_flat, _ = get_training_variables()
text_embedder = TokEmbeddingSeq2SeqExtractor(texts_flat)
if cfg['output_type'] != 'pair':
reranker = TrainableReranker(da_embedder, text_embedder, cfg_path)
else:
reranker = PairwiseReranker(da_embedder, text_embedder, cfg_path)
if reranker.load_model():
print("WARNING THE TRAINING START POINT IS AN ALREADY TRAINED MODEL")
if cfg["train"]:
print("Training")
text_seqs, da_seqs, scores, log_probs = get_scores_ordered_beam(
cfg,
da_embedder,
import pickle from base_models import TGEN_Model from embedding_extractor import TokEmbeddingSeq2SeqExtractor, DAEmbeddingSeq2SeqExtractor from utils import get_training_variables, get_test_das import numpy as np texts, das = get_training_variables() text_embedder = TokEmbeddingSeq2SeqExtractor(texts) da_embedder = DAEmbeddingSeq2SeqExtractor(das) das_test = get_test_das() beam_path = 'output_files/saved_beams/16_vanilla_2_3.pickle' beam_path_2 = 'output_files/saved_beams/t2p_vanilla_3.pickle' beams = pickle.load(file=open(beam_path, 'rb')) beams_2 = pickle.load(file=open(beam_path_2, 'rb')) models = TGEN_Model(da_embedder, text_embedder, 'new_configs/model_configs/seq2seq_all_data.yaml') models.load_models() da_emb = da_embedder.get_embeddings([das_test[0]])[0] inf_enc_out = models.encoder_model.predict(np.array([da_emb])) enc_outs = inf_enc_out[0] enc_last_state = inf_enc_out[1:] print(beams[0][0][1] == beams_2[0][0][1]) print(sum(models.get_prob_sequence(enc_outs, beams[0][0][1], enc_last_state))) print(beams[0][0][0], ' '.join(text_embedder.reverse_embedding(beams[0][0][1]))) print() print(sum(models.get_prob_sequence(enc_outs, beams_2[0][0][1], enc_last_state))) print(beams_2[0][0][0], ' '.join(text_embedder.reverse_embedding(beams_2[0][0][1])))