Fix build (#939)

* fix broken build

* remove unneeded import

* remove print

* fix indent

* fix tests after h5py upgrade

* silence the linter

* linting

* more linter-silencing

* fix tensordot2 test

* fix string encoding

* fix torch version, change pylint indent to 2

* change yapf file

* nit

* remove print

* silence linter

* add comment

* fix number of epochs in the test

* fix convert_to_tensor in JaxBackend

* fix type tests for jax-arrays

Author: mganahl

.style.yapf

@@ -1,3 +1,4 @@
 [style]
 # TensorNetwork uses the yapf style
 based_on_style = yapf
+indent_width = 2

requirements_travis.txt

@@ -1,6 +1,6 @@
 tensorflow>=2.0.0
 pytest
-torch>=1.4.0
+torch==1.8.1 # TODO (mganahl): remove restriction once torch.tensordot bug is fixed (https://github.com/pytorch/pytorch/issues/65524)
 jax>=0.1.68
 jaxlib>=0.1.59
 pylint==2.5.3

tensornetwork/backends/jax/jax_backend.py

@@ -119,9 +119,10 @@ def sqrt(self, tensor: Tensor) -> Tensor:
     return jnp.sqrt(tensor)
 
   def convert_to_tensor(self, tensor: Tensor) -> Tensor:
-    if (not isinstance(tensor, jnp.ndarray) and not jnp.isscalar(tensor)):
-      raise TypeError("Expected a `jnp.array` or scalar. Got {}".format(
-          type(tensor)))
+    if (not isinstance(tensor, (np.ndarray, jnp.ndarray))
+        and not jnp.isscalar(tensor)):
+      raise TypeError(("Expected a `jnp.array`, `np.array` or scalar. "
+                       f"Got {type(tensor)}"))
     result = jnp.asarray(tensor)
     return result
 
@@ -320,7 +321,7 @@ def A(H,x):
                          "`dtype` have to be provided")
       initial_state = self.randn(shape, dtype)
 
-    if not isinstance(initial_state, jnp.ndarray):
+    if not isinstance(initial_state, (jnp.ndarray, np.ndarray)):
       raise TypeError("Expected a `jax.array`. Got {}".format(
           type(initial_state)))
 
@@ -435,7 +436,7 @@ def A(H,x):
                          "`dtype` have to be provided")
       initial_state = self.randn(shape, dtype)
 
-    if not isinstance(initial_state, jnp.ndarray):
+    if not isinstance(initial_state, (jnp.ndarray, np.ndarray)):
       raise TypeError("Expected a `jax.array`. Got {}".format(
           type(initial_state)))
 
@@ -555,7 +556,7 @@ def A(H,x):
                          "`dtype` have to be provided")
       initial_state = self.randn(shape, dtype)
 
-    if not isinstance(initial_state, jnp.ndarray):
+    if not isinstance(initial_state, (jnp.ndarray, np.ndarray)):
       raise TypeError("Expected a `jax.array`. Got {}".format(
           type(initial_state)))
     if A not in _CACHED_MATVECS:

tensornetwork/tn_keras/condenser.py

@@ -11,7 +11,7 @@
 
 
 # pytype: disable=module-attr
-@tf.keras.utils.register_keras_serializable(package='tensornetwork')
+@tf.keras.utils.register_keras_serializable(package='tensornetwork')# pylint: disable=no-member
 # pytype: enable=module-attr
 class DenseCondenser(Layer):
   """Condenser TN layer. Greatly reduces dimensionality of input.

tensornetwork/tn_keras/conv2d_mpo.py

@@ -10,7 +10,7 @@
 import math
 
 # pytype: disable=module-attr
-@tf.keras.utils.register_keras_serializable(package='tensornetwork')
+@tf.keras.utils.register_keras_serializable(package='tensornetwork')# pylint: disable=no-member
 # pytype: enable=module-attr
 class Conv2DMPO(Layer):
   """2D Convolutional Matrix Product Operator (MPO) TN layer.
@@ -197,7 +197,7 @@ def is_perfect_root(n, n_nodes):
     else:
       self.use_bias = None
 
-  def call(self, inputs: tf.Tensor, **kwargs) -> tf.Tensor: #pylint: disable=arguments-differ
+  def call(self, inputs: tf.Tensor) -> tf.Tensor: #pylint: disable=arguments-differ
 
     tn_nodes = [tn.Node(n, backend='tensorflow') for n in self.nodes]
     for i in range(len(tn_nodes) - 1):

tensornetwork/tn_keras/dense.py

@@ -9,7 +9,7 @@
 
 
 # pytype: disable=module-attr
-@tf.keras.utils.register_keras_serializable(package='tensornetwork')
+@tf.keras.utils.register_keras_serializable(package='tensornetwork')# pylint: disable=no-member
 # pytype: enable=module-attr
 class DenseDecomp(Layer):
   """TN layer comparable to Dense that carries out matrix multiplication

tensornetwork/tn_keras/entangler.py

@@ -11,7 +11,7 @@
 
 
 # pytype: disable=module-attr
-@tf.keras.utils.register_keras_serializable(package='tensornetwork')
+@tf.keras.utils.register_keras_serializable(package='tensornetwork')# pylint: disable=no-member
 # pytype: enable=module-attr
 class DenseEntangler(Layer):
   """Entangler TN layer. Allows for very large hidden layers.

tensornetwork/tn_keras/expander.py

@@ -11,7 +11,7 @@
 
 
 # pytype: disable=module-attr
-@tf.keras.utils.register_keras_serializable(package='tensornetwork')
+@tf.keras.utils.register_keras_serializable(package='tensornetwork') # pylint: disable=no-member
 # pytype: enable=module-attr
 class DenseExpander(Layer):
   """Expander TN layer. Greatly expands dimensionality of input.

tensornetwork/tn_keras/mpo.py

@@ -11,7 +11,7 @@
 
 
 # pytype: disable=module-attr
-@tf.keras.utils.register_keras_serializable(package='tensornetwork')
+@tf.keras.utils.register_keras_serializable(package='tensornetwork')# pylint: disable=no-member
 # pytype: enable=module-attr
 class DenseMPO(Layer):
   """Matrix Product Operator (MPO) TN layer.

tensornetwork/tn_keras/test_layer.py

@@ -113,8 +113,8 @@ def test_train(dummy_data, make_model):
                 loss='binary_crossentropy',
                 metrics=['accuracy'])
 
-  # Train the model for 10 epochs
-  history = model.fit(data, labels, epochs=50, batch_size=64)
+  # Train the model for 40 epochs
+  history = model.fit(data, labels, epochs=40, batch_size=64)
 
   # Check that loss decreases and accuracy increases
   assert history.history['loss'][0] > history.history['loss'][-1]