def main(_):
project_name = 'DUAL'
ft_params = DualParams(project_name)
ft_params.post_process()
mode = ft_params.params['mode']
model_fn = DUALModel(ft_params.params)
input_fn = DUALBaseData(ft_params.params)
if mode == 'train':
print('train&eval')
estimator_solver.train(input_fn=input_fn,
model_fn=model_fn,
params=ft_params.params)
elif mode == 'eval':
print('eval')
estimator_solver.evaluate(input_fn=input_fn,
model_fn=model_fn,
params=ft_params.params)
elif mode == 'infer':
print('predict')
estimator_solver.predict(input_fn=input_fn,
model_fn=model_fn,
params=ft_params.params)
else:
raise ValueError('invalid mode: {}'.format(mode))
model_dir = ft_params.params['model_dir']
utils.stat_eval_results(model_dir, 'eval_result.txt')
def main(_):
project_name = 'ADEN_ft'
seq_params = MixedSeqMultiModalFtParams(project_name)
seq_params.post_process()
mode = seq_params.params['mode']
model_fn = MixedSeqMultiModalFtModel(seq_params.params)
input_fn = MixedSeqMultiModalData(seq_params.params)
if mode == 'train':
print('train&eval')
estimator_solver.train(input_fn=input_fn,
model_fn=model_fn,
params=seq_params.params)
elif mode == 'eval':
print('eval')
estimator_solver.evaluate(input_fn=input_fn,
model_fn=model_fn,
params=seq_params.params)
elif mode == 'infer':
print('predict')
estimator_solver.predict(input_fn=input_fn,
model_fn=model_fn,
params=seq_params.params)
else:
raise ValueError('invalid mode: {}'.format(mode))
print('------------ evaluate ------------')
tf.logging.info('------------ evaluate ------------')
estimator_solver.evaluate(input_fn=input_fn,
model_fn=model_fn,
params=seq_params.params)
model_dir = seq_params.params['model_dir']
utils.stat_eval_results(model_dir, 'eval_result.txt')