best at-parameters

-------------------------------------------------------
training_noise 0.0, inference_noise 0.1
-------------------------------------------------------
at_opt_accessor = lambda out, state: state[0]

at = active_tuning.ActiveTuning(model=model_at, initial_model_state=lstm_state_at,
                  initial_model_output=x_t.clone(), tuning_length=8,
                  tf_rate = 0.0,
                  opt_accessor=at_opt_accessor,
                  bias_correction=False,
                  tuning_cycles=10, eta=0.008, beta1=0.9, beta2=0.99/0.999,
                  epsilon=1e-8)

-------------------------------------------------------
training_noise 0.0, inference_noise 0.2
-------------------------------------------------------

at_opt_accessor = lambda out, state: state[0]

at = active_tuning.ActiveTuning(model=model_at, initial_model_state=lstm_state_at,
                  initial_model_output=x_t.clone(), tuning_length=8,
                  tf_rate = 0.0,
                  opt_accessor=at_opt_accessor,
                  bias_correction=False,
                  tuning_cycles=10, eta=0.005, beta1=0.9, beta2=0.9,
                  epsilon=1e-8)

rmse teacher forcing :  0.20235786
rmse active tuning   :  0.20433469

-------------------------------------------------------
training_noise 0.0, inference_noise 0.5
-------------------------------------------------------

        at_opt_accessor = lambda out, state: state[0]

        at = active_tuning.ActiveTuning(model=model_at, initial_model_state=lstm_state_at,
                          initial_model_output=x_t.clone(), tuning_length=8,
                          tf_rate = 0.0,
                          opt_accessor=at_opt_accessor,
                          bias_correction=False,
                          tuning_cycles=10, eta=0.005, beta1=0.9, beta2=0.999,
                          epsilon=1e-8)

rmse (at): 0.3733811


at_opt_accessor = lambda out, state: state[0]

at = active_tuning.ActiveTuning(model=model_at, initial_model_state=lstm_state_at,
                  initial_model_output=x_t.clone(), tuning_length=8,
                  tf_rate = 0.0,
                  opt_accessor=at_opt_accessor,
                  bias_correction=False,
                  tuning_cycles=10, eta=0.002, beta1=0.0, beta2=0.9,
                  epsilon=1e-8)

rmse teacher forcing :  0.40659916
rmse active tuning   :  0.39893943

-------------------------------------------------------
training_noise 0.0, inference_noise 1.0
-------------------------------------------------------

at_opt_accessor = lambda out, state: state[0]

at = active_tuning.ActiveTuning(model=model_at, initial_model_state=lstm_state_at,
                  initial_model_output=x_t.clone(), tuning_length=8,
                  tf_rate = 0.0,
                  opt_accessor=at_opt_accessor,
                  bias_correction=False,
                  tuning_cycles=10, eta=0.001, beta1=0.0, beta2=0.9,
                  epsilon=1e-8)

rmse teacher forcing :  0.5443207
rmse active tuning   :  0.65555054



-------------------------------------------------------
training_noise 0.05, inference_noise 0.1
-------------------------------------------------------

at_opt_accessor = lambda out, state: state[0]

at = active_tuning.ActiveTuning(model=model_at, initial_model_state=lstm_state_at,
                  initial_model_output=x_t.clone(), tuning_length=8,
                  tf_rate = 0.0,
                  opt_accessor=at_opt_accessor,
                  bias_correction=False,
                  tuning_cycles=10, eta=0.008, beta1=0.9, beta2=0.9,
                  epsilon=1e-8)

rmse teacher forcing :  0.13756949
rmse active tuning   :  0.09435017


-------------------------------------------------------
training_noise 0.05, inference_noise 0.5
-------------------------------------------------------

at_opt_accessor = lambda out, state: state[0]

at = active_tuning.ActiveTuning(model=model_at, initial_model_state=lstm_state_at,
                  initial_model_output=x_t.clone(), tuning_length=8,
                  tf_rate = 0.0,
                  opt_accessor=at_opt_accessor,
                  bias_correction=False,
                  tuning_cycles=10, eta=0.002, beta1=0.9, beta2=0.9,
                  epsilon=1e-8)

rmse teacher forcing :  0.41027328
rmse active tuning   :  0.392077

