Changed code to give option to leave out reconstructions for ANN data.

ANN_Data_Generator.py

@@ -31,7 +31,8 @@ class TrainingDataGenerator:
 
     """
     def __init__(self, initial_conditions, left_bound=-1, right_bound=1,
-                 balance=0.5, stencil_length=3, directory='test_data'):
+                 balance=0.5, stencil_length=3, directory='test_data',
+                 add_reconstructions=True):
         """Initializes TrainingDataGenerator.
 
         Parameters
@@ -49,11 +50,15 @@ class TrainingDataGenerator:
         directory : str, optional
             Path to directory in which training data is saved.
             Default: 'test_data'.
+        add_reconstructions: bool, optional
+            Flag whether reconstructions of the middle cell are included.
+            Default: True.
 
         """
         self._balance = balance
         self._left_bound = left_bound
         self._right_bound = right_bound
+        self._add_reconstructions = add_reconstructions
 
         # Set stencil length
         if stencil_length % 2 == 0:
@@ -174,7 +179,10 @@ class TrainingDataGenerator:
         print('Samples to complete:', num_samples)
         tic = time.perf_counter()
 
-        input_data = np.zeros((num_samples, self._stencil_length+2))
+        num_datapoints = self._stencil_length
+        if self._add_reconstructions:
+            num_datapoints += 2
+        input_data = np.zeros((num_samples, num_datapoints))
         num_init_cond = len(initial_conditions)
         count = 0
         for i in range(num_samples):
@@ -202,7 +210,8 @@ class TrainingDataGenerator:
                 right_bound=right_bound, quadrature='Gauss',
                 quadrature_config={'num_eval_points': polynomial_degree+1})
             input_data[i] = dg_scheme.build_training_data(
-                adjustment, self._stencil_length, initial_condition)
+                adjustment, self._stencil_length, self._add_reconstructions,
+                initial_condition)
 
             count += 1
             if count % 1000 == 0:

DG_Approximation.py

@@ -8,6 +8,9 @@ TODO: Adapt TCD from Soraya
     (Dropbox->...->TEST_troubled-cell-detector->Troubled_Cell_Detector)
 TODO: Add verbose output
 TODO: Improve file naming (e.g. use '.' instead of '__')
+TODO: Give option to leave out reconstructions for ANN data -> Done
+TODO: Force input_size for each ANN model to be stencil length
+TODO: Combine ANN workflows
 
 Critical, but not urgent:
 TODO: Use cfl_number for updating, not just time
@@ -384,6 +387,8 @@ class DGScheme:
             Extent of adjustment of each evaluation point in x-direction.
         stencil_length : int
             Size of training data array.
+        add_reconstructions: bool
+            Flag whether reconstructions of the middle cell are included.
         initial_condition : InitialCondition object, optional
             Initial condition used for calculation.
             Default: None (i.e. instance variable).
@@ -399,5 +404,6 @@ class DGScheme:
             initial_condition = self._init_cond
         projection = self._do_initial_projection(initial_condition, adjustment)
 
-        return self._detector.calculate_cell_average_and_reconstructions(
-            projection[:, 1:-1], stencil_length)
+        return self._detector.calculate_cell_average(projection[:, 1:-1],
+                                                     stencil_length,
+                                                     add_reconstructions)

Troubled_Cell_Detector.py

@@ -132,13 +132,13 @@ class TroubledCellDetector:
         """
         pass
 
-    def calculate_cell_average_and_reconstructions(self, projection,
-                                                   stencil_length):
-        """Calculates cell averages and reconstructions for a given projection.
+    def calculate_cell_average(self, projection, stencil_length,
+                               add_reconstructions=True):
+        """Calculates cell averages for a given projection.
 
         Calculate the cell averages of all cells in a projection.
-        Reconstructions are only calculated for the middle cell and added
-        left and right to it, respectively.
+        If desired, reconstructions are calculated for the middle cell
+        and added left and right to it, respectively.
 
         Parameters
         ----------
@@ -146,16 +146,21 @@ class TroubledCellDetector:
             Matrix of projection for each polynomial degree.
         stencil_length : int
             Size of data array.
+        add_reconstructions: bool, optional
+            Flag whether reconstructions of the middle cell are included.
+            Default: True.
 
         Returns
         -------
         ndarray
-            Matrix containing cell averages and reconstructions for initial
+            Matrix containing cell averages (and reconstructions) for initial
             projection.
 
         """
         cell_averages = calculate_approximate_solution(
             projection, [0], 0, self._basis.get_basis_vector())
+
+        if add_reconstructions:
             left_reconstructions = calculate_approximate_solution(
                 projection, [-1], self._polynomial_degree,
                 self._basis.get_basis_vector())
@@ -169,6 +174,7 @@ class TroubledCellDetector:
                                          cell_averages[:, middle_idx],
                                          right_reconstructions[:, middle_idx],
                                          cell_averages[:, middle_idx+1:]))))
+        return np.array(list(map(np.float64, cell_averages)))
 
     def plot_results(self, projection, troubled_cell_history, time_history):
         """Plots results and troubled cells of a projection.
@@ -283,9 +289,13 @@ class ArtificialNeuralNetwork(TroubledCellDetector):
         super()._reset(config)
 
         self._stencil_len = config.pop('stencil_len', 3)
+        self._add_reconstructions = config.pop('add_reconstructions', True)
         self._model = config.pop('model', 'ThreeLayerReLu')
+        num_datapoints = self._stencil_len
+        if self._add_reconstructions:
+            num_datapoints += 2
         self._model_config = config.pop('model_config', {
-            'input_size': self._stencil_len+2, 'first_hidden_size': 8,
+            'input_size': num_datapoints, 'first_hidden_size': 8,
             'second_hidden_size': 4, 'output_size': 2,
             'activation_function': 'Softmax', 'activation_config': {'dim': 1}})
         model_state = config.pop('model_state', 'Snakemake-Test/trained '
@@ -322,9 +332,9 @@ class ArtificialNeuralNetwork(TroubledCellDetector):
 
         # Calculate input data depending on stencil length
         input_data = torch.from_numpy(
-            np.vstack([self.calculate_cell_average_and_reconstructions(
+            np.vstack([self.calculate_cell_average(
                 projection[:, cell-num_ghost_cells:cell+num_ghost_cells+1],
-                self._stencil_len)
+                self._stencil_len, self._add_reconstructions)
                 for cell in range(num_ghost_cells,
                                   len(projection[0])-num_ghost_cells)]))
 

config.yaml

-data_dir: 'model-Mar01'
+data_dir: 'model-Soraya_Mar02'
 random_seed: 1234
 
 # Parameter for Approximation with Troubled Cell Detection
 Approximation:
-  plot_dir: 'fig-Mar01'
+  plot_dir: 'fig-Soraya_Mar02'
 
   schemes:
     Separation_Test:
@@ -21,6 +21,7 @@ Approximation:
       detector_config:
         fold_len: 16
         whisker_len: 3
+        add_reconstructions: False
         model_state: 'Adam.model.pt'
 
       init_cond: 'Sine'
@@ -51,6 +52,7 @@ ANN_Data:
   smooth_troubled_balance: 0.5
 
   stencil_length: 3
+  add_reconstructions : False
 
   # Initial Conditions for Training Data
   functions:
@@ -85,7 +87,7 @@ ANN_Training:
       threshold: 1.0e-5
       batch_size: 500
       model: ThreeLayerReLu
-      model_config: {}
+      model_config: {input_size: 3}
       loss_function: BCELoss
       optimizer: Adam
     SGD:
@@ -93,7 +95,7 @@ ANN_Training:
       threshold: 1.0e-5
       batch_size: 500
       model: ThreeLayerReLu
-      model_config: {}
+      model_config: {input_size: 3}
       loss_function: BCELoss
       optimizer: SGD
 

workflows/ANN_data.smk

@@ -22,6 +22,7 @@ rule generate_data:
         balance = config['smooth_troubled_balance'],
         stencil_length = config['stencil_length'],
         sample_number = config['sample_number'],
+        reconstruction_flag = config['add_reconstructions'],
         functions = expand('{FUNCTION}', FUNCTION=config['functions'])
     log:
         DIR+'/log/generate_data.log'
@@ -41,6 +42,7 @@ rule generate_data:
                 initial_conditions=initial_conditions,
                 left_bound=params.left_bound, right_bound=params.right_bound,
                 balance=params.balance,
-                stencil_length=params.stencil_length, directory=DIR)
+                stencil_length=params.stencil_length, directory=DIR,
+                add_reconstructions=params.reconstruction_flag)
             data = generator.build_training_data(
                 num_samples=params.sample_number)
\ No newline at end of file