remove TensorInput and SimplePolyNN

astroNN/models/__init__.py

@@ -18,7 +18,7 @@
     ApokascEncoderDecoder,
     StarNet2017,
 )
-from astroNN.models.misc_models import Cifar10CNN, MNIST_BCNN, SimplePolyNN
+from astroNN.models.misc_models import Cifar10CNN, MNIST_BCNN
 from astroNN.nn.losses import losses_lookup
 from astroNN.nn.utilities import Normalizer
 from astroNN.shared.dict_tools import dict_list_to_dict_np, list_to_dict
@@ -37,7 +37,6 @@
     "StarNet2017",
     "Cifar10CNN",
     "MNIST_BCNN",
-    "SimplePolyNN",
 ]
 
 optimizers = keras.optimizers

astroNN/models/apogee_models.py

@@ -12,8 +12,6 @@
     MCDropout,
     BoolMask,
     StopGrad,
-    KLDivergenceLayer,
-    TensorInput,
     VAESampling,
 )
 from astroNN.nn.losses import (

docs/source/neuralnets/layers.rst

@@ -414,52 +414,3 @@ It can be used with keras or tensorflow.keras, you just have to import the funct
         stopped_grad_layer = BoolMask(mask=....)(...)
         # some layers ...
         return model
-
-
-TensorInput Layer
------------------------
-
-.. autoclass:: astroNN.nn.layers.TensorInput
-    :members: call, get_config
-
-
-`TensorInput` takes tensorflow tensor as layer initialization and return the tensor.
-
-`TensorInput` can be imported by
-
-.. code-block:: python
-    :linenos:
-    
-    from astroNN.nn.layers import TensorInput
-
-For example, if you want to generate random tensor as other layers input and do not want it to register it as model input, you can
-
-.. code-block:: python
-    :linenos:
-    
-    from astroNN.nn.layers import TensorInput
-    # we use zeros loss just to demonstrate StopGrad works and no error backprop from StopGrad layer
-    from astroNN.nn.losses import zeros_loss
-    import numpy as np
-    from astroNN.shared.nn_tools import cpu_fallback
-    import tensorflow as tf
-    import keras
-
-    cpu_fallback()  # force tf to use CPU
-
-    Input = keras.layers.Input
-    Dense = keras.layers.Dense
-    concatenate = keras.layers.concatenate
-    Model = keras.models.Model
-
-    # Data preparation
-    random_xdata = np.random.normal(0, 1, (100, 7514))
-    random_ydata = np.random.normal(0, 1, (100, 25))
-    input1 = Input(shape=[7514])
-    input2 = TensorInput(tensor=tf.random.normal(mean=0., stddev=1., shape=tf.shape(input1)))([])
-    output = Dense(25, name='dense')(concatenate([input1, input2]))
-    model = Model(inputs=input1, outputs=output)
-    model.compile(optimizer=keras.optimizers.SGD(lr=0.1),
-                  loss='mse')
-    print(model.input_names)
-    >>> ['input_1']  # only input_1 as input_2 is not really an input we requiring user to input

tests/test_layers.py

@@ -250,22 +250,5 @@ def test_FastMCInference(self):
         # make sure accelerated model has no variance (uncertainty) on deterministic model prediction
         self.assertAlmostEqual(np.sum(sy[:, :, 1]), 0.0)
 
-    def test_TensorInput(self):
-        print("==========BoolMask tests==========")
-        from astroNN.nn.layers import TensorInput
-
-        input1 = Input(shape=[7514], name="input")
-        input2 = TensorInput(
-            tensor=keras.backend.random.normal(
-                mean=0.0, stddev=1.0, shape=keras.backend.shape(input1)
-            )
-        )([])
-        output = Dense(25, name="dense")(concatenate([input1, input2]))
-        model = Model(inputs=input1, outputs=output)
-        model.compile(optimizer="adam", loss="mse")
-
-        self.assertEqual(len(model.input_names), 1)
-
-
 if __name__ == "__main__":
     unittest.main()