haved seq length, double batch size, now like in paper

LCAS · Jun 24, 2019 · 616493f · 616493f
1 parent 10ad6f0
commit 616493f
Show file tree

Hide file tree

Showing 2 changed files with 6 additions and 6 deletions.
diff --git a/rkn/pendulum_image_imputation.py b/rkn/pendulum_image_imputation.py
@@ -63,8 +63,8 @@ def build_decoder_hidden(self):
                 seed=0,
                 pendulum_params=pend_params)
 
-train_obs, train_obs_valid, train_targets = generate_imputation_data_set(data, 1000, 150, seed=42)
-test_obs, test_obs_valid, test_targets = generate_imputation_data_set(data, 250, 150, seed=23541)
+train_obs, train_obs_valid, train_targets = generate_imputation_data_set(data, 2000, 75, seed=42)
+test_obs, test_obs_valid, test_targets = generate_imputation_data_set(data, 1000, 75, seed=23541)
 
 # Build Model
 rkn = PendulumImageImputationRKN(observation_shape=train_obs.shape[-3:], latent_observation_dim=15,
@@ -74,7 +74,7 @@ def build_decoder_hidden(self):
 
 # Train Model
 rkn.fit((train_obs, train_obs_valid),
-        train_targets, batch_size=25, epochs=1000,
+        train_targets, batch_size=50, epochs=500,
         validation_data=((test_obs, test_obs_valid), test_targets))
 
 

diff --git a/rkn/pendulum_state_estimation.py b/rkn/pendulum_state_estimation.py
@@ -47,16 +47,16 @@ def build_var_decoder_hidden(self):
                 seed=0,
                 pendulum_params=pend_params)
 
-train_obs, train_targets = generate_pendulum_filter_dataset(data, 1000, 150, 42)
-test_obs, test_targets = generate_pendulum_filter_dataset(data, 250, 150, 12312)
+train_obs, train_targets = generate_pendulum_filter_dataset(data, 2000, 75, np.random.randint(100000000))
+test_obs, test_targets = generate_pendulum_filter_dataset(data, 1000, 75, np.random.randint(10000000))
 
 # Build Model
 rkn = PendulumStateEstemRKN(observation_shape=train_obs.shape[-3:], latent_observation_dim=15, output_dim=2, num_basis=15,
                             bandwidth=3, never_invalid=True)
 rkn.compile(optimizer=k.optimizers.Adam(clipnorm=5.0), loss=rkn.gaussian_nll, metrics=[rkn.rmse])
 
 # Train Model
-rkn.fit(train_obs, train_targets, batch_size=25, epochs=1000, validation_data=(test_obs, test_targets))
+rkn.fit(train_obs, train_targets, batch_size=50, epochs=500, validation_data=(test_obs, test_targets), verbose=2)