Simplify numpy_to_input, removing unused no_reshape and setting gridp…

…oints to None in shape always

n3fit/src/n3fit/backends/keras_backend/operations.py

@@ -114,9 +114,7 @@ def batchit(x, batch_dimension=0, **kwarg):
 # layer generation
 def numpy_to_input(
     numpy_array: np.ndarray,
-    no_reshape: bool = False,
     name: str = None,
-    custom_shape: tuple = None,
 ):
     """
     Takes a numpy array and generates a Input layer.
@@ -126,26 +124,17 @@ def numpy_to_input(
     Parameters
     ----------
         numpy_array: np.ndarray
-        no_reshape: bool
-            if true, don't add batch dimension, take the first dimension of the array as the batch
         name: bool
             name to give to the layer
-        custom_shape: tuple
-            To specify a more general shape with None values
     """
-    if no_reshape:
-        batched_array = numpy_array
-        batch_size = numpy_array.shape[0]
-        shape = numpy_array.shape[1:]
-    else:
-        batched_array = np.expand_dims(numpy_array, 0)
-        batch_size = 1
-        shape = numpy_array.shape
-    if custom_shape is not None:
-        shape = custom_shape
+    batched_array = np.expand_dims(numpy_array, 0)
+    batch_size = 1
+    shape = list(numpy_array.shape)
+    # set the number of gridpoints to None, otherwise shapes don't show in model.summary
+    shape[0] = None
+
     input_layer = Input(batch_size=batch_size, shape=shape, name=name)
     input_layer.tensor_content = batched_array
-    input_layer.original_shape = no_reshape
     return input_layer
 
 

n3fit/src/n3fit/model_gen.py

@@ -577,9 +577,7 @@ def pdfNN_layer_generator(
             input_x_eq_1 = scaler(input_x_eq_1)[0]
         # the layer that subtracts 1 from the NN output
         subtract_one_layer = Lambda(op.op_subtract, name='subtract_one')
-        layer_x_eq_1 = op.numpy_to_input(
-            np.array(input_x_eq_1).reshape(1, 1), name='x_ones', custom_shape=(None, 1)
-        )  # Just to make shapes consistent
+        layer_x_eq_1 = op.numpy_to_input(np.array(input_x_eq_1).reshape(1, 1), name='x_eq_1')
         model_input["layer_x_eq_1"] = layer_x_eq_1
 
     # the layer that multiplies the NN output by the preprocessing factor

n3fit/src/n3fit/msr.py

@@ -59,12 +59,8 @@ def generate_msr_model_and_grid(
     if scaler:
         xgrid_integration = scaler(xgrid_integration)
 
-    # Turn into input layer. Specify a custom shape here using None,
-    # so shapes will display properly in the model summary
-    grid_shape = 2 if scaler is not None else 1
-    xgrid_integration = op.numpy_to_input(
-        xgrid_integration, name="integration_grid", custom_shape=(None, grid_shape)
-    )
+    # Turn into input layer.
+    xgrid_integration = op.numpy_to_input(xgrid_integration, name="integration_grid")
 
     # 1c Get the original grid
     if scaler: