change test set seed

TSP/tsp_utils.py

@@ -120,10 +120,10 @@ def __init__(self,
 
         self.n_nodes = args['n_nodes']
         self.input_dim = args['input_dim']
-        self.input_data = tf.placeholder(tf.float32,\
+        self.input_data = tf.compat.v1.placeholder(tf.float32,\
             shape=[None,self.n_nodes,args['input_dim']])
         self.input_pnt = self.input_data
-        self.batch_size = tf.shape(self.input_data)[0] 
+        self.batch_size = tf.shape(input=self.input_data)[0] 
 
     def reset(self,beam_width=1):
         '''
@@ -198,6 +198,6 @@ def reward_func(sample_solution=None):
     # get the reward based on the route lengths
 
 
-    route_lens_decoded = tf.reduce_sum(tf.pow(tf.reduce_sum(tf.pow(\
-        (sample_solution_tilted - sample_solution) ,2), 2) , .5), 0)
+    route_lens_decoded = tf.reduce_sum(input_tensor=tf.pow(tf.reduce_sum(input_tensor=tf.pow(\
+        (sample_solution_tilted - sample_solution) ,2), axis=2) , .5), axis=0)
     return route_lens_decoded 

VRP/vrp_attention.py

@@ -6,19 +6,19 @@ def __init__(self, dim, use_tanh=False, C=10,_name='Attention',_scope=''):
         self.use_tanh = use_tanh
         self._scope = _scope
 
-        with tf.variable_scope(_scope+_name):
+        with tf.compat.v1.variable_scope(_scope+_name):
             # self.v: is a variable with shape [1 x dim]
-            self.v = tf.get_variable('v',[1,dim],
-                       initializer=tf.contrib.layers.xavier_initializer())
+            self.v = tf.compat.v1.get_variable('v',[1,dim],
+                       initializer=tf.compat.v1.keras.initializers.VarianceScaling(scale=1.0, mode="fan_avg", distribution="uniform"))
             self.v = tf.expand_dims(self.v,2)
 
-        self.emb_d = tf.layers.Conv1D(dim,1,_scope=_scope+_name+'/emb_d' ) #conv1d
-        self.emb_ld = tf.layers.Conv1D(dim,1,_scope=_scope+_name+'/emb_ld' ) #conv1d_2
+        self.emb_d = tf.compat.v1.layers.Conv1D(dim,1,_scope=_scope+_name+'/emb_d' ) #conv1d
+        self.emb_ld = tf.compat.v1.layers.Conv1D(dim,1,_scope=_scope+_name+'/emb_ld' ) #conv1d_2
 
-        self.project_d = tf.layers.Conv1D(dim,1,_scope=_scope+_name+'/proj_d' ) #conv1d_1
-        self.project_ld = tf.layers.Conv1D(dim,1,_scope=_scope+_name+'/proj_ld' ) #conv1d_3
-        self.project_query = tf.layers.Dense(dim,_scope=_scope+_name+'/proj_q' ) #
-        self.project_ref = tf.layers.Conv1D(dim,1,_scope=_scope+_name+'/proj_ref' ) #conv1d_4
+        self.project_d = tf.compat.v1.layers.Conv1D(dim,1,_scope=_scope+_name+'/proj_d' ) #conv1d_1
+        self.project_ld = tf.compat.v1.layers.Conv1D(dim,1,_scope=_scope+_name+'/proj_ld' ) #conv1d_3
+        self.project_query = tf.compat.v1.layers.Dense(dim,_scope=_scope+_name+'/proj_q' ) #
+        self.project_ref = tf.compat.v1.layers.Conv1D(dim,1,_scope=_scope+_name+'/proj_ref' ) #conv1d_4
 
 
         self.C = C  # tanh exploration parameter
@@ -40,7 +40,7 @@ def __call__(self, query, ref, env):
         # get the current demand and load values from environment
         demand = env.demand
         load = env.load
-        max_time = tf.shape(demand)[1]
+        max_time = tf.shape(input=demand)[1]
 
         # embed demand and project it
         # emb_d:[batch_size x max_time x dim ]
@@ -61,7 +61,7 @@ def __call__(self, query, ref, env):
         expanded_q = tf.tile(tf.expand_dims(q,1),[1,max_time,1])
 
         # v_view:[batch_size x dim x 1]
-        v_view = tf.tile( self.v, [tf.shape(e)[0],1,1]) 
+        v_view = tf.tile( self.v, [tf.shape(input=e)[0],1,1]) 
 
         # u : [batch_size x max_time x dim] * [batch_size x dim x 1] = 
         #       [batch_size x max_time]
@@ -82,17 +82,17 @@ def __init__(self, dim, use_tanh=False, C=10,_name='Attention',_scope=''):
         self.use_tanh = use_tanh
         self._scope = _scope
 
-        with tf.variable_scope(_scope+_name):
+        with tf.compat.v1.variable_scope(_scope+_name):
             # self.v: is a variable with shape [1 x dim]
-            self.v = tf.get_variable('v',[1,dim],
-                       initializer=tf.contrib.layers.xavier_initializer())
+            self.v = tf.compat.v1.get_variable('v',[1,dim],
+                       initializer=tf.compat.v1.keras.initializers.VarianceScaling(scale=1.0, mode="fan_avg", distribution="uniform"))
             self.v = tf.expand_dims(self.v,2)
 
-        self.emb_d = tf.layers.Conv1D(dim,1,_scope=_scope+_name +'/emb_d') #conv1d
-        self.project_d = tf.layers.Conv1D(dim,1,_scope=_scope+_name +'/proj_d') #conv1d_1
+        self.emb_d = tf.compat.v1.layers.Conv1D(dim,1,_scope=_scope+_name +'/emb_d') #conv1d
+        self.project_d = tf.compat.v1.layers.Conv1D(dim,1,_scope=_scope+_name +'/proj_d') #conv1d_1
 
-        self.project_query = tf.layers.Dense(dim,_scope=_scope+_name +'/proj_q') #
-        self.project_ref = tf.layers.Conv1D(dim,1,_scope=_scope+_name +'/proj_e') #conv1d_2
+        self.project_query = tf.compat.v1.layers.Dense(dim,_scope=_scope+_name +'/proj_q') #
+        self.project_ref = tf.compat.v1.layers.Conv1D(dim,1,_scope=_scope+_name +'/proj_e') #conv1d_2
 
         self.C = C  # tanh exploration parameter
         self.tanh = tf.nn.tanh
@@ -119,7 +119,7 @@ def __call__(self, query, ref, env):
         """
         # we need the first demand value for the critic
         demand = env.input_data[:,:,-1]
-        max_time = tf.shape(demand)[1]
+        max_time = tf.shape(input=demand)[1]
 
         # embed demand and project it
         # emb_d:[batch_size x max_time x dim ]
@@ -134,7 +134,7 @@ def __call__(self, query, ref, env):
         expanded_q = tf.tile(tf.expand_dims(q,1),[1,max_time,1])
 
         # v_view:[batch_size x dim x 1]
-        v_view = tf.tile( self.v, [tf.shape(e)[0],1,1]) 
+        v_view = tf.tile( self.v, [tf.shape(input=e)[0],1,1]) 
 
         # u : [batch_size x max_time x dim] * [batch_size x dim x 1] = 
         #       [batch_size x max_time]

VRP/vrp_utils.py

@@ -144,12 +144,12 @@ def __init__(self,
         self.n_nodes = args['n_nodes']
         self.n_cust = args['n_cust']
         self.input_dim = args['input_dim']
-        self.input_data = tf.placeholder(tf.float32,\
+        self.input_data = tf.compat.v1.placeholder(tf.float32,\
             shape=[None,self.n_nodes,self.input_dim])
 
         self.input_pnt = self.input_data[:,:,:2]
         self.demand = self.input_data[:,:,-1]
-        self.batch_size = tf.shape(self.input_pnt)[0] 
+        self.batch_size = tf.shape(input=self.input_pnt)[0] 
 
     def reset(self,beam_width=1):
         '''
@@ -220,7 +220,7 @@ def step(self,
         d_sat = tf.minimum(tf.gather_nd(self.demand,batched_idx), self.load)
 
         # update the demand
-        d_scatter = tf.scatter_nd(batched_idx, d_sat, tf.cast(tf.shape(self.demand),tf.int64))
+        d_scatter = tf.scatter_nd(batched_idx, d_sat, tf.cast(tf.shape(input=self.demand),tf.int64))
         self.demand = tf.subtract(self.demand, d_scatter)
 
         # update load
@@ -239,7 +239,7 @@ def step(self,
 
         self.mask += tf.concat( [tf.tile(tf.expand_dims(tf.cast(tf.equal(self.load,0),
             tf.float32),1), [1,self.n_cust]),                      
-            tf.expand_dims(tf.multiply(tf.cast(tf.greater(tf.reduce_sum(self.demand,1),0),tf.float32),
+            tf.expand_dims(tf.multiply(tf.cast(tf.greater(tf.reduce_sum(input_tensor=self.demand,axis=1),0),tf.float32),
                              tf.squeeze( tf.cast(tf.equal(idx,self.n_cust),tf.float32))),1)],1)
 
         state = State(load=self.load,
@@ -283,7 +283,7 @@ def reward_func(sample_solution):
     # get the reward based on the route lengths
 
 
-    route_lens_decoded = tf.reduce_sum(tf.pow(tf.reduce_sum(tf.pow(\
-        (sample_solution_tilted - sample_solution) ,2), 2) , .5), 0)
+    route_lens_decoded = tf.reduce_sum(input_tensor=tf.pow(tf.reduce_sum(input_tensor=tf.pow(\
+        (sample_solution_tilted - sample_solution) ,2), axis=2) , .5), axis=0)
     return route_lens_decoded 
 

main.py

@@ -10,6 +10,8 @@
 from shared.decode_step import RNNDecodeStep
 from model.attention_agent import RLAgent
 
+tf.compat.v1.disable_eager_execution()
+
 def load_task_specific_components(task):
     '''
     This function load task-specific libraries
@@ -36,9 +38,9 @@ def load_task_specific_components(task):
     return DataGenerator, Env, reward_func, AttentionActor, AttentionCritic
 
 def main(args, prt):
-    config = tf.ConfigProto()
+    config = tf.compat.v1.ConfigProto()
     config.gpu_options.allow_growth = True
-    sess = tf.Session(config=config)
+    sess = tf.compat.v1.Session(config=config)
 
     # load task specific classes
     DataGenerator, Env, reward_func, AttentionActor, AttentionCritic = \
@@ -97,7 +99,7 @@ def main(args, prt):
     if random_seed is not None and random_seed > 0:
         prt.print_out("# Set random seed to %d" % random_seed)
         np.random.seed(random_seed)
-        tf.set_random_seed(random_seed)
-    tf.reset_default_graph()
+        tf.compat.v1.set_random_seed(random_seed)
+    tf.compat.v1.reset_default_graph()
 
     main(args, prt)

misc_utils.py

@@ -91,11 +91,11 @@ class Logger(object):
 
     def __init__(self, log_dir):
         """Create a summary writer logging to log_dir."""
-        self.writer = tf.summary.FileWriter(log_dir)
+        self.writer = tf.compat.v1.summary.FileWriter(log_dir)
 
     def scalar_summary(self, tag, value, step):
         """Log a scalar variable."""
-        summary = tf.Summary(value=[tf.Summary.Value(tag=tag, simple_value=value)])
+        summary = tf.compat.v1.Summary(value=[tf.compat.v1.Summary.Value(tag=tag, simple_value=value)])
         self.writer.add_summary(summary, step)
 
     def image_summary(self, tag, images, step):
@@ -111,14 +111,14 @@ def image_summary(self, tag, images, step):
             scipy.misc.toimage(img).save(s, format="png")
 
             # Create an Image object
-            img_sum = tf.Summary.Image(encoded_image_string=s.getvalue(),
+            img_sum = tf.compat.v1.Summary.Image(encoded_image_string=s.getvalue(),
                                        height=img.shape[0],
                                        width=img.shape[1])
             # Create a Summary value
-            img_summaries.append(tf.Summary.Value(tag='%s/%d' % (tag, i), image=img_sum))
+            img_summaries.append(tf.compat.v1.Summary.Value(tag='%s/%d' % (tag, i), image=img_sum))
 
         # Create and write Summary
-        summary = tf.Summary(value=img_summaries)
+        summary = tf.compat.v1.Summary(value=img_summaries)
         self.writer.add_summary(summary, step)
 
     def histo_summary(self, tag, values, step, bins=1000):
@@ -128,7 +128,7 @@ def histo_summary(self, tag, values, step, bins=1000):
         counts, bin_edges = np.histogram(values, bins=bins)
 
         # Fill the fields of the histogram proto
-        hist = tf.HistogramProto()
+        hist = tf.compat.v1.HistogramProto()
         hist.min = float(np.min(values))
         hist.max = float(np.max(values))
         hist.num = int(np.prod(values.shape))
@@ -145,7 +145,7 @@ def histo_summary(self, tag, values, step, bins=1000):
             hist.bucket.append(c)
 
         # Create and write Summary
-        summary = tf.Summary(value=[tf.Summary.Value(tag=tag, histo=hist)])
+        summary = tf.compat.v1.Summary(value=[tf.compat.v1.Summary.Value(tag=tag, histo=hist)])
         self.writer.add_summary(summary, step)
         self.writer.flush()
 
@@ -158,12 +158,12 @@ def _single_cell(unit_type, num_units, forget_bias, dropout, prt,
     # Cell Type
     if unit_type == "lstm":
         prt.print_out("  LSTM, forget_bias=%g" % forget_bias, new_line=False)
-        single_cell = tf.contrib.rnn.BasicLSTMCell(
+        single_cell = tf.compat.v1.nn.rnn_cell.BasicLSTMCell(
                 num_units,
                 forget_bias=forget_bias)
     elif unit_type == "gru":
         prt.print_out("  GRU", new_line=False)
-        single_cell = tf.contrib.rnn.GRUCell(num_units)
+        single_cell = tf.compat.v1.nn.rnn_cell.GRUCell(num_units)
     else:
         raise ValueError("Unknown unit type %s!" % unit_type)
 
@@ -250,15 +250,15 @@ def create_rnn_cell(unit_type, num_units, num_layers, num_residual_layers,
     if len(cell_list) == 1:  # Single layer.
         return cell_list[0]
     else:  # Multi layers
-        return tf.contrib.rnn.MultiRNNCell(cell_list)
+        return tf.compat.v1.nn.rnn_cell.MultiRNNCell(cell_list)
 
 def gradient_clip(gradients, params, max_gradient_norm):
     """Clipping gradients of a model."""
     clipped_gradients, gradient_norm = tf.clip_by_global_norm(
             gradients, max_gradient_norm)
-    gradient_norm_summary = [tf.summary.scalar("grad_norm", gradient_norm)]
+    gradient_norm_summary = [tf.compat.v1.summary.scalar("grad_norm", gradient_norm)]
     gradient_norm_summary.append(
-            tf.summary.scalar("clipped_gradient", tf.global_norm(clipped_gradients)))
+            tf.compat.v1.summary.scalar("clipped_gradient", tf.linalg.global_norm(clipped_gradients)))
 
     return clipped_gradients, gradient_norm_summary
 
@@ -274,7 +274,7 @@ def create_or_load_model(model, model_dir, session, out_dir, name):
     else:
         utils.print_out("  created %s model with fresh parameters, time %.2fs." %
                                         (name, time.time() - start_time))
-        session.run(tf.global_variables_initializer())
+        session.run(tf.compat.v1.global_variables_initializer())
 
     global_step = model.global_step.eval(session=session)
     return model, global_step
@@ -290,14 +290,14 @@ def add_summary(summary_writer, global_step, tag, value):
     """Add a new summary to the current summary_writer.
     Useful to log things that are not part of the training graph, e.g., tag=BLEU.
     """
-    summary = tf.Summary(value=[tf.Summary.Value(tag=tag, simple_value=value)])
+    summary = tf.compat.v1.Summary(value=[tf.compat.v1.Summary.Value(tag=tag, simple_value=value)])
     summary_writer.add_summary(summary, global_step)
 
 
 def get_config_proto(log_device_placement=False, allow_soft_placement=True):
     # GPU options:
     # https://www.tensorflow.org/versions/r0.10/how_tos/using_gpu/index.html
-    config_proto = tf.ConfigProto(
+    config_proto = tf.compat.v1.ConfigProto(
             log_device_placement=log_device_placement,
             allow_soft_placement=allow_soft_placement)
     config_proto.gpu_options.allow_growth = True
@@ -311,7 +311,7 @@ def debug_tensor(s, msg=None, summarize=10):
     """Print the shape and value of a tensor at test time. Return a new tensor."""
     if not msg:
         msg = s.name
-    return tf.Print(s, [tf.shape(s), s], msg + " ", summarize=summarize)
+    return tf.compat.v1.Print(s, [tf.shape(input=s), s], msg + " ", summarize=summarize)
 
 def tf_print(tensor, transform=None):
 
@@ -321,7 +321,7 @@ def print_tensor(x):
         # but adding a transformation of some kind usually makes the output more digestible
         print(x if transform is None else transform(x))
         return x
-    log_op = tf.py_func(print_tensor, [tensor], [tensor.dtype])[0]
+    log_op = tf.compat.v1.py_func(print_tensor, [tensor], [tensor.dtype])[0]
     with tf.control_dependencies([log_op]):
         res = tf.identity(tensor)