bugfixes, add dtype support

params/calo_inn.yaml

@@ -1,36 +1,42 @@
 run_name: calo_inn
+dtype: float64
+
 
 #Dataset
 loader_module: calo_inn
 loader_params: 
-  geant_file: /remote/gpu06/favaro/calo_inn/datasets/cls_data/train_cls_piplus.hdf5
-  generated_file: /remote/gpu06/favaro/calo_inn/datasets/train_piplus.hdf5
+  geant_file: /remote/gpu06/favaro/discriminator-metric/data/calo_cls_geant/full_cls_eplus.hdf5
+  generated_file: /remote/gpu06/favaro/discriminator-metric/data/calo_bay_samples/samples_eplus.hdf5
   add_log_energy: True
   add_log_layer_ens: True
   add_logit_step: False
+  add_cut: 0.0
   train_split: 0.6
   test_split: 0.2
 
 # Model
 activation: leaky_relu
-negative_slope: 0.1
-dropout: 0.1
-layers: 5
-hidden_size: 256
+negative_slope: 0.2
+dropout: 0.0
+layers: 2
+hidden_size: 512
 
 # Training
 bayesian: False
+prior_prec: 0.01
+std_init: -9.0
+
 lr: 1.e-3
 betas: [0.9, 0.99]
 weight_decay: 0.0
-epochs: 50
-batch_size: 1024
+epochs: 150
+batch_size: 512
 lr_scheduler: reduce_on_plateau
 lr_decay_factor: 0.1
 lr_patience: 5
 checkpoint_interval: 5
 
 # Evaluation
-#bayesian_samples: 2
+bayesian_samples: 5
 #lower_cluster_thresholds: [0.01, 0.1]
 #upper_cluster_thresholds: [0.9, 0.99]

src/__main__.py

@@ -34,6 +34,15 @@ def main():
     print("Using device " + ("GPU" if use_cuda else "CPU"))
     device = torch.device("cuda:0" if use_cuda else "cpu")
 
+    dtype = params.get('dtype', 'float32')
+    if dtype=='float64':
+        torch.set_default_dtype(torch.float64)
+    elif dtype=='float16':
+        torch.set_default_dtype(torch.float16)
+    elif dtype=='float32':
+        torch.set_default_dtype(torch.float32)
+    print("Using dtype {}".format(dtype))
+
     print("Loading data")
     loader = import_module(f"src.loaders.{params['loader_module']}")
     datasets = loader.load(params["loader_params"])

src/loaders/calo_inn.py

@@ -86,7 +86,7 @@ def create_data(data_path, dataset_list, **kwargs):
 
     if kwargs['add_log_energy']:
         data = np.concatenate((data, np.log10(en_test*10).reshape(-1, 1)), axis=1)
-    data = np.nan_to_num(data, posinf=0, neginf=0)
+    #data = np.nan_to_num(data, posinf=0, neginf=0)
 
     en0_t = np.log10(en0_t + 1e-8) + 2.
     en1_t = np.log10(en1_t + 1e-8) +2.

src/train.py

@@ -64,7 +64,7 @@ def init_data_loaders(self):
         """
         make_loader = lambda data, mode: torch.utils.data.DataLoader(
             dataset = torch.utils.data.TensorDataset(
-                torch.tensor(data, device=self.device)
+                torch.tensor(data, device=self.device, dtype=torch.get_default_dtype())
             ),
             batch_size = self.params["batch_size"],
             shuffle = mode == "train",