Ensured repeatable output through seeds and deterministic models.

ANN_Training.py

@@ -71,6 +71,13 @@ class ModelTrainer:
         self._num_epochs = config.pop('num_epochs', 1000)
         self._threshold = config.pop('threshold', 1e-5)
 
+        # Set random seed
+        seed = config.pop('random_seed', None)
+        if seed is not None:
+            np.random.seed(seed)
+            torch.manual_seed(seed)
+            torch.use_deterministic_algorithms(True)
+
         model = config.pop('model', 'ThreeLayerReLu')
         model_config = config.pop('model_config', {})
         loss_function = config.pop('loss_function', 'BCELoss')

Snakefile

-import numpy as np
-
 configfile: 'config.yaml'
 
 DIR = 'workflows'
-if config['random_seed'] is not None:
-    np.random.seed(config['random_seed'])
 
 module_config = {'data_dir': config['data_dir'],
                  'random_seed': config['random_seed']}

workflows/ANN_training.smk

 import sys
-import numpy as np
 
 import ANN_Training
 from ANN_Training import *
@@ -9,8 +8,6 @@ configfile: 'config.yaml'
 
 DIR = config['data_dir']
 MODELS = config['models']
-if config['random_seed'] is not None:
-    np.random.seed(config['random_seed'])
 
 rule all:
     input:
@@ -30,15 +27,9 @@ rule plot_test_results:
     log:
         DIR+'/log/plot_test_results.log'
     run:
-        models = {}
         with open(str(log), 'w') as logfile:
             sys.stdout = logfile
             sys.stderr = logfile
-            for model in MODELS:
-                trainer= ANN_Training.ModelTrainer(
-                    {'model_name': model, 'dir': DIR, 'model_dir': DIR,
-                     **MODELS[model]})
-                models[model] = trainer
             plot_evaluation_results(evaluation_file=input.json_file,
                 directory=DIR, colors=params.colors)
 
@@ -52,7 +43,8 @@ rule test_model:
         protected(DIR+'/model evaluation/'+'_'.join(MODELS.keys())+'.json')
     params:
         num_iterations = config['num_iterations'],
-        compare_normalization = config['compare_normalization']
+        compare_normalization = config['compare_normalization'],
+        random_seed = config['random_seed']
     log:
         DIR+'/log/test_model.log'
     run:
@@ -63,6 +55,7 @@ rule test_model:
             for model in MODELS:
                 trainer= ANN_Training.ModelTrainer(
                     {'model_name': model, 'dir': DIR, 'model_dir': DIR,
+                     'random_seed': params.random_seed,
                      **MODELS[model]})
                 models[model] = trainer
             evaluate_models(models=models, directory=DIR,
@@ -77,6 +70,8 @@ rule train_model:
     output:
         protected(DIR+'/trained models/{model}.model.pt'),
         protected(DIR+'/trained models/{model}.loss.pt')
+    params:
+        random_seed = config['random_seed']
     log:
         DIR+'/log/train_model_{model}.log'
     run:
@@ -84,6 +79,7 @@ rule train_model:
             sys.stdout = logfile
             training_data = read_training_data(DIR)
             trainer= ANN_Training.ModelTrainer(
-                config={**MODELS[wildcards.model]})
+                config={'random_seed': params.random_seed,
+                        **MODELS[wildcards.model]})
             trainer.epoch_training(dataset=training_data)
             trainer.save_model(directory=DIR, model_name=wildcards.model)

workflows/approximation.smk

 import sys
 import time
-import numpy as np
 
 import Initial_Condition
 import Quadrature
 from DG_Approximation import DGScheme
 from Plotting import plot_approximation_results
-from projection_utils import Mesh
 
 configfile: 'config.yaml'
 
 PLOTS = ['error', 'exact_and_approx', 'semilog_error', 'shock_tube']
 DIR = config['data_dir']
 SCHEMES = config['schemes']
-if config['random_seed'] is not None:
-    np.random.seed(config['random_seed'])
 
 rule all:
     input: