def main():
config = Config()
args = add_arguments(config)
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu_id
# sample the dataset
config.parse_arg(args)
dset = Dataset(config)
print('------------------------------------------------------------')
print('Pivot word discovery:')
dset.build()
config.vocab_size = len(dset.word2id)
print('------------------------------------------------------------')
print('Pivot classifier:')
dset.classify()
print('------------------------------------------------------------')
print('Precision-recall histogram:')
dset.get_prec_recl()
print('------------------------------------------------------------')
print('Storing the pivot outputs')
dset.store_pivots()
# the logistic classifier
if (args.classifier == 'ff'):
classifier = FFClassifier(config)
x_train, y_train = dset.to_bow_numpy('train')
classifier.train(x_train, y_train)
x_dev, y_dev = dset.to_bow_numpy('dev')
classifier.test(x_dev, y_dev)
x_test, y_test = dset.to_bow_numpy('test')
classifier.test(x_test, y_test)
elif (args.classifier == 'cnn'):
cnn = CNNClassifier(config)
x_train, y_train = dset.to_sent_numpy('train')
cnn.train(x_train, y_train)
x_dev, y_dev = dset.to_sent_numpy('dev')
cnn.test(x_dev, y_dev)
x_test, y_test = dset.to_sent_numpy('test')
cnn.test(x_test, y_test)
else:
pass
# correlation between the pivot words and logistic classifier words
return
def main():
# configuration
config = Config()
config.parse_arg(FLAGS)
config.setup_path()
config.print_arg()
# dataset
if (config.dataset == 'wikibio'):
dset = DatasetTable2text(config)
dset.load()
config.key_size = len(dset.key2id)
else:
dset = Dataset(config)
dset.build()
config.vocab_size = len(dset.word2id)
config.dec_start_id = dset.word2id["_GOO"]
config.dec_end_id = dset.word2id["_EOS"]
config.pad_id = dset.pad_id
config.stop_words = dset.stop_words
config.id2wordemb = dset.id2wordemb
# model
if (config.model_name == "transformer_bow"):
Model = TransformerBow
elif (config.model_name == "seq2seq"):
if (config.dataset == 'wikibio'): Model = Seq2seqData2text
else: Model = Seq2seq
elif (config.model_name == "bow_seq2seq"): Model = BowSeq2seq
elif (config.model_name == "vae"): Model = Vae
elif (config.model_name == "hierarchical_vae"): Model = Hierarchical_Vae
elif (config.model_name == "latent_bow"):
if (config.dataset == 'wikibio'): Model = LatentBowData2text
else: Model = LatentBow
elif (config.model_name == "lm"): Model = LM
else:
msg = "the model name shoule be in ['transformer_bow', 'seq2seq', 'vae', 'hierarchical_vae', 'latent_low', 'lm'], "
msg += "current name: %s" % config.model_name
raise Exception(msg)
model = Model(config)
with tf.variable_scope(config.model_name):
model.build()
# controller
controller = Controller(config)
if (config.model_name != "lm"):
if ("lm" in controller.eval_metrics_list):
controller.build_lm(LM, config)
controller.train(model, dset)
return
def main():
config = Config()
args = add_arguments(config)
config.parse_arg(args)
dset = Dataset(config)
dset.build()
# print('debug:')
# print(dset.id2word[1])
config.vocab_size = len(dset.word2id)
# read the transfered sentences
transfer_analysis = PivotTransferAnalysis(config)
if(config.model == 'cmu'):
transfer_analysis.pipeline_w_cmu(dset)
else:
transfer_analysis.pipeline(dset)
return
def main():
config = json.load(open("config.json"))
dset = Dataset(config)
dset.build()
return
