minor change: set kernel params earlier for nu method

lasgroup · Aug 24, 2023 · 6402dc8 · 6402dc8
1 parent 0a4c59e
commit 6402dc8
Show file tree

Hide file tree

Showing 4 changed files with 16 additions and 13 deletions.
diff --git a/experiments/meta_learning_exp/sweep_meta_learning_exp.py b/experiments/meta_learning_exp/sweep_meta_learning_exp.py
@@ -16,7 +16,7 @@
         'bandwidth': {'distribution': 'log_uniform_10', 'min': 0., 'max': 2.},
     },
     'NP': {
-        'num_iter_meta_train': {'values': [40000]},
+        'num_iter_meta_train': {'values': [100000]},
         'latent_dim': {'values': [64, 128]},
         'hidden_dim': {'values': [32, 64]},
         'lr': {'distribution': 'log_uniform_10', 'min': -4., 'max': -3},

diff --git a/experiments/regression_exp/inspect_results.py b/experiments/regression_exp/inspect_results.py
@@ -93,19 +93,21 @@ def main(args, drop_nan=False):
 
     print('Models:', set(df_agg['model']))
 
-    different_method_plot(df_agg, metric='nll')
-    different_method_plot(df_agg, metric='rmse')
+    different_method_plot(df_agg, metric='nll', filter_std_higher_than=50.)
+    different_method_plot(df_agg, metric='rmse', filter_std_higher_than=0.2)
 
-    df_method = df_agg[(df_agg['model'] == 'BNN_FSVGD_SimPrior_gp')]
+    df_method = df_agg[(df_agg['model'] == 'BNN_FSVGD_SimPrior_nu-method')]
+    #df_method = df_method[(df_method['num_train_steps'] == 80000)]
 
     #df_method = df_method[df_method['bandwidth_score_estim'] > 1.0]
 
     metric = 'nll'
     for param in ['num_f_samples', 'bandwidth_score_estim', 'bandwidth_svgd', 'num_measurement_points']:
         plt.scatter(df_method[param], df_method[(metric, 'mean')])
         plt.xlabel(param)
-        plt.xscale('log')
+        #plt.xscale('log')
         plt.ylabel(metric)
+        plt.ylim(-15, 20)
         plt.show()
 
     QUANTILE_BASED_CI = True
@@ -157,7 +159,7 @@ def main(args, drop_nan=False):
 
 if __name__ == '__main__':
     parser = argparse.ArgumentParser(description='Inspect results of a regression experiment.')
-    parser.add_argument('--exp_name', type=str, default='aug22')
-    parser.add_argument('--data_source', type=str, default='racecar')
+    parser.add_argument('--exp_name', type=str, default='june09')
+    parser.add_argument('--data_source', type=str, default='pendulum')
     args = parser.parse_args()
     main(args)
diff --git a/experiments/regression_exp/sweep_regression_exp.py b/experiments/regression_exp/sweep_regression_exp.py
@@ -34,22 +34,23 @@
     },
     'BNN_FSVGD_SimPrior_nu-method': {
         'bandwidth_svgd': {'distribution': 'log_uniform_10', 'min': -1.0, 'max': 0.0},
-        'num_train_steps': {'values': [40000]},
+        'num_train_steps': {'values': [30000, 40000, 50000]},
         'num_measurement_points': {'values': [16, 32]},
         'num_f_samples': {'values': [512]},
-        'bandwidth_score_estim': {'distribution': 'log_uniform_10', 'min': 0.0, 'max': 0.5},
+        #'bandwidth_score_estim': {'distribution': 'log_uniform_10', 'min': -0.5, 'max': 0.5},
+        'bandwidth_score_estim': {'distribution': 'uniform', 'min': 0.5, 'max': 4.0},
     },
     'BNN_FSVGD_SimPrior_gp+nu-method': {
         'bandwidth_svgd': {'distribution': 'log_uniform_10', 'min': -1.0, 'max': 0.0},
-        'num_train_steps': {'values': [40000]},
+        'num_train_steps': {'values': [80000]},
         'num_measurement_points': {'values': [16, 32]},
         'num_f_samples': {'values': [512]},
         'switch_score_estimator_frac': {'values': [0.6667]},
         'bandwidth_score_estim': {'distribution': 'log_uniform_10', 'min': 0.0, 'max': 0.5},
     },
     'BNN_FSVGD_SimPrior_kde': {
         'bandwidth_svgd': {'distribution': 'log_uniform', 'min': -2., 'max': 2.},
-        'num_train_steps': {'values': [20000]},
+        'num_train_steps': {'values': [40000]},
         'num_measurement_points': {'values': [16, 32]},
         'num_f_samples': {'values': [512, 1024, 2056]},
     },

diff --git a/sim_transfer/score_estimation/nu_method.py b/sim_transfer/score_estimation/nu_method.py
@@ -26,9 +26,9 @@ def __init__(self,
             num_iter = jnp.array(1.0 / jnp.sqrt(lam)).astype(jnp.int32) + 1
 
         if kernel_type == 'curl_free_imq':
-            self._kernel = CurlFreeIMQKernel()
+            self._kernel = CurlFreeIMQKernel(kernel_hyperparams=bandwidth)
         elif kernel_type == 'curl_free_se':
-            self._kernel = CurlFreeSEKernel()
+            self._kernel = CurlFreeSEKernel(kernel_hyperparams=bandwidth)
         else:
             raise NotImplementedError(f'Kernel type {kernel_type} is not implemented.')