add +x bit to run.sh, change tf -> tf.compat.v1, freeze requirements …

…for tf.contrib

PiperOrigin-RevId: 276586556

attribution/integrated_gradients.py

@@ -289,7 +289,7 @@ def AddBOWIntegratedGradientsOps(graph,
     assert len(embedding_lookup.get_shape()) == 3
     assert len(embedding.get_shape()) == 2
   with graph.as_default():
-    num_evals = tf.placeholder_with_default(
+    num_evals = tf.compat.v1.placeholder_with_default(
         tf.constant(50, name='num_evals'), shape=())
     attribution_dims_map = {embedding: [1] for embedding in embedding_list}
     attribution_hooks = AddIntegratedGradientsOps(

attribution/integrated_gradients_test.py

@@ -28,11 +28,12 @@ class AttributionTest(tf.test.TestCase):
 
   def testAddIntegratedGradientsOps(self):
     with tf.Graph().as_default() as graph:
-      var1 = tf.get_variable(name='var1', initializer=[[[1., 2., 3.]]])
+      var1 = tf.compat.v1.get_variable(
+          name='var1', initializer=[[[1., 2., 3.]]])
       input_tensor = tf.placeholder(shape=[None, None, 3], dtype=tf.float32)
       x = tf.multiply(input_tensor, [[[1.]]])
       var1_times_x = tf.multiply(var1, x)
-      var2 = tf.get_variable(
+      var2 = tf.compat.v1.get_variable(
           name='var2', initializer=[[4., 5.], [6., 7.], [4., 3.]])
       matmul = tf.einsum('ijk,kl->ijl', var1_times_x, var2)
       output_tensor = tf.reduce_sum(matmul, [1, 2], name='out')

bitempered_loss/requirements.txt

@@ -1,3 +1,3 @@
-tensorflow >=1.11.0
+tensorflow>=1.11.0,<2.0.0
 numpy>=1.13.1
-absl>=0.1.6
+absl-py>=0.1.6

dql_grasping/requirements.txt

@@ -1,6 +1,6 @@
 absl-py>=0.5.0
 numpy>=1.13.3
-tensorflow>=1.11.0
+tensorflow>=1.11.0,<2.0.0
 gin-config
 pybullet
 Pillow==5.3.0

evanet/requirements.txt

@@ -1,4 +1,4 @@
 absl-py>=0.5.0
 numpy>=1.13.3
-tensorflow>=1.11.0
+tensorflow>=1.11.0,<2.0.0
 protobuf>=3.9.0

explaining_risk_increase/requirements.txt

@@ -1 +1 @@
-tensorflow>=1.12.0
+tensorflow>=1.12.0,<2.0.0

fat/run.sh

@@ -20,5 +20,5 @@ virtualenv -p python3 .
 source ./bin/activate
 
 pip install tensorflow
-pip install -r fat_bert_nq/requirements.txt
+pip install -r fat/fat_bert_nq/requirements.txt
 python3 -m fat.fat_bert_nq.run_nq_test

large_margin/margin_loss.py

@@ -88,7 +88,7 @@ def large_margin(  # pylint: disable=invalid-name
     use_approximation=True,
     worst_case_loss=True,
     layers_weights=None,
-    loss_collection=tf.GraphKeys.LOSSES):
+    loss_collection=tf.compat.v1.GraphKeys.LOSSES):
   """Creates a large margin loss.
 
   Args:

m_theory/dim4/so8_supergravity_extrema/code/extrema.py

@@ -26,11 +26,6 @@
 import pprint
 import time
 
-# Setting up `mpmath` global default precision at initialization time.
-# Uses value from the environment variable `MPMATH_DPS`, or 100 if unset.
-mpmath.mp.dps = int(os.getenv('MPMATH_DPS', '100'))
-
-
 from dim4.so8_supergravity_extrema.code import distillation
 from dim4.so8_supergravity_extrema.code import scalar_sector_mpmath
 from dim4.so8_supergravity_extrema.code import scalar_sector_tensorflow
@@ -39,6 +34,10 @@
 import mpmath
 import numpy
 
+# Setting up `mpmath` global default precision at initialization time.
+# Uses value from the environment variable `MPMATH_DPS`, or 100 if unset.
+mpmath.mp.dps = int(os.getenv('MPMATH_DPS', '100'))
+
 
 def scan_for_solutions(seed, scale, num_iterations, output_basename):
   """Scans for critical points (with TensorFlow)."""

robust_loss/adaptive.py

@@ -58,7 +58,7 @@ def _construct_scale(x, scale_lo, scale_init, float_dtype):
   else:
     # Otherwise we construct a "latent" scale variable and define `scale`
     # As an affine function of a softplus on that latent variable.
-    latent_scale = tf.get_variable(
+    latent_scale = tf.compat.v1.get_variable(
         'LatentScale', initializer=tf.zeros((1, x.shape[1]), float_dtype))
     scale = util.affine_softplus(latent_scale, lo=scale_lo, ref=scale_init)
   return scale
@@ -176,7 +176,7 @@ def lossfun(x,
         alpha_init = (alpha_lo + alpha_hi) / 2.
       latent_alpha_init = util.inv_affine_sigmoid(
           alpha_init, lo=alpha_lo, hi=alpha_hi)
-      latent_alpha = tf.get_variable(
+      latent_alpha = tf.compat.v1.get_variable(
           'LatentAlpha',
           initializer=tf.fill((1, x.shape[1]),
                               tf.cast(latent_alpha_init, dtype=float_dtype)))
@@ -233,7 +233,7 @@ def lossfun_students(x, scale_lo=1e-5, scale_init=1.):
   float_dtype = x.dtype
   assert_ops = [tf.Assert(tf.equal(tf.rank(x), 2), [tf.rank(x)])]
   with tf.control_dependencies(assert_ops):
-    log_df = tf.get_variable(
+    log_df = tf.compat.v1.get_variable(
         name='LogDf', initializer=tf.zeros((1, x.shape[1]), float_dtype))
     scale = _construct_scale(x, scale_lo, scale_init, float_dtype)
     loss = util.students_t_nll(x, tf.math.exp(log_df), scale)

robust_loss/adaptive_test.py

@@ -150,7 +150,7 @@ def testInitialAlphaAndScaleAreCorrect(self, float_dtype):
         true_alpha_init = (alpha_lo + alpha_hi) / 2.
       scale_init = float_dtype(np.random.uniform() + 0.5)
       scale_lo = float_dtype(np.random.uniform() * 0.1)
-      with tf.variable_scope('trial_' + str(i)):
+      with tf.compat.v1.variable_scope('trial_' + str(i)):
         _, alpha, scale = adaptive.lossfun(
             tf.constant(np.zeros((10, 10), float_dtype)),
             alpha_lo=alpha_lo,
@@ -179,7 +179,7 @@ def testFixedAlphaAndScaleAreCorrect(self, float_dtype):
       mu = tf.Variable(
           tf.zeros(tf.shape(samples)[1], float_dtype), name='DummyMu')
       x = samples - mu[tf.newaxis, :]
-      with tf.variable_scope('trial_' + str(i)):
+      with tf.compat.v1.variable_scope('trial_' + str(i)):
         loss, alpha, scale = adaptive.lossfun(
             x,
             alpha_lo=alpha_lo,

robust_loss/wavelet.py

@@ -145,7 +145,7 @@ def pad_reflecting(x, padding_below, padding_above, axis):
   i = tf.range(-padding_below, n + padding_above, dtype=tf.int32)
   # `j` contains the indices of the input tensor corresponding to the output
   # padded tensor.
-  i_mod = tf.mod(i, tf.maximum(1, 2 * (n - 1)))
+  i_mod = tf.math.mod(i, tf.maximum(1, 2 * (n - 1)))
   j = tf.minimum(2 * (n - 1) - i_mod, i_mod)
   return tf.gather(x, j, axis=axis)
 

towards_gan_benchmarks/requirements.txt

@@ -1,4 +1,4 @@
 absl-py>=0.7.0
 numpy>=1.15.2
 scipy>=1.2.0
-tensorflow>=1.10.0
+tensorflow>=1.10.0,<2.0.0