Update config file for training on mnist 14*14.

experiments/mnist/mnist_0_scaled_14_linf_lognormal_gpu.yaml

@@ -20,12 +20,13 @@ experiments:
     trial_config:
       logging:
         images: true
-        "image_shape": [10, 10]
+        "image_shape": [14, 14]
       dataset: &dataset
         __object__: src.explib.datasets.MnistSplit
         scale: true
         digit: 0
         device: *device
+        scale_factor: 2
       epochs: &epochs 200000
       patience: &patience 40
       batch_size: &batch_size
@@ -35,7 +36,7 @@ experiments:
           __class__: torch.optim.Adam
         params:
           lr:
-            __eval__: tune.loguniform(1e-7, 1e-4)
+            __eval__: tune.loguniform(1e-6, 1e-4)
           weight_decay: 0.0
       model_cfg:
         type:
@@ -52,73 +53,21 @@ experiments:
           coupling_layers: &coupling_layers
             __eval__: tune.choice([i for i in range(3, 4)])
           coupling_nn_layers: &coupling_nn_layers
-            __eval__: "tune.choice([[w] * l for l in [1, 2, 3] for w in [100, 200, 300]])" # tune.choice([[c*32, c*16, c*8, c*16, c*32] for c in [1, 2, 3, 4]] + [[c*64, c*32, c*64]  for c in range(1,5)] + [[c*128] * 2 for c in range(1,5)] + [[c*256] for c in range(1,5)])
+            __eval__: "tune.choice([[w] * l for l in [1, 2, 3] for w in [196, 392]])" # tune.choice([[c*32, c*16, c*8, c*16, c*32] for c in [1, 2, 3, 4]] + [[c*64, c*32, c*64]  for c in range(1,5)] + [[c*128] * 2 for c in range(1,5)] + [[c*256] for c in range(1,5)])
           nonlinearity: &nonlinearity
             __eval__: tune.choice([torch.nn.ReLU()])
-          split_dim: 50
+          split_dim: 98
           base_distribution:
             __object__: src.veriflow.distributions.RadialDistribution
             device: *device
             p:
               __eval__: math.inf
             loc:
-              __eval__: torch.zeros(100).to("cuda")
+              __eval__: torch.zeros(196).to("cuda")
             norm_distribution:
               __object__: pyro.distributions.LogNormal
               loc:
                 __eval__: (1.2 * torch.ones(1)).to("cuda")
               scale:
                 __eval__: (0.5 * torch.ones(1)).to("cuda")
           use_lu: false
-  - &mnist_logNormal_linf_loc_12_scale_05_medium_sized
-    __overwrites__: *mnist_logNormal_linf_loc_1_scale_05_medium_sized
-    name: mnist_logNormal_linf_loc_12_scale_05_medium_sized
-    trial_config:
-      model_cfg:
-        params:
-          base_distribution:
-            norm_distribution:
-              __object__: pyro.distributions.LogNormal
-              loc:
-                __eval__: (1.2 * torch.ones(1)).to("cuda")
-              scale:
-                __eval__: (0.5 * torch.ones(1)).to("cuda")
-  - &mnist_logNormal_linf_loc_08_scale_05_medium_sized
-    __overwrites__: *mnist_logNormal_linf_loc_1_scale_05_medium_sized
-    name: mnist_logNormal_linf_loc_08_scale_05_medium_sized
-    trial_config:
-      model_cfg:
-        params:
-          base_distribution:
-            norm_distribution:
-              __object__: pyro.distributions.LogNormal
-              loc:
-                __eval__: (0.8 * torch.ones(1)).to("cuda")
-              scale:
-                __eval__: (0.5 * torch.ones(1)).to("cuda")
-  - &mnist_logNormal_linf_loc_1_scale_03_medium_sized
-    __overwrites__: *mnist_logNormal_linf_loc_1_scale_05_medium_sized
-    name: mnist_logNormal_linf_loc_1_scale_03_medium_sized
-    trial_config:
-      model_cfg:
-        params:
-          base_distribution:
-            norm_distribution:
-              __object__: pyro.distributions.LogNormal
-              loc:
-                __eval__: (1 * torch.ones(1)).to("cuda")
-              scale:
-                __eval__: (0.3 * torch.ones(1)).to("cuda")
-  - &mnist_logNormal_linf_loc_1_scale_07_medium_sized
-    __overwrites__: *mnist_logNormal_linf_loc_1_scale_05_medium_sized
-    name: mnist_logNormal_linf_loc_1_scale_07_medium_sized
-    trial_config:
-      model_cfg:
-        params:
-          base_distribution:
-            norm_distribution:
-              __object__: pyro.distributions.LogNormal
-              loc:
-                __eval__: (1 * torch.ones(1)).to("cuda")
-              scale:
-                __eval__: (0.7 * torch.ones(1)).to("cuda")