merged master

baselines/common/mpi_adam_optimizer.py

@@ -1,6 +1,11 @@
 import numpy as np
 import tensorflow as tf
-from mpi4py import MPI
+from baselines.common import tf_util as U
+from baselines.common.tests.test_with_mpi import with_mpi
+try:
+    from mpi4py import MPI
+except ImportError:
+    MPI = None
 
 class MpiAdamOptimizer(tf.train.AdamOptimizer):
     """Adam optimizer that averages gradients across mpi processes."""
@@ -13,34 +18,60 @@ class MpiAdamOptimizer(tf.train.AdamOptimizer):
         flat_grad = tf.concat([tf.reshape(g, (-1,)) for g, v in grads_and_vars], axis=0)
         shapes = [v.shape.as_list() for g, v in grads_and_vars]
         sizes = [int(np.prod(s)) for s in shapes]
-
         num_tasks = self.comm.Get_size()
         buf = np.zeros(sum(sizes), np.float32)
-
-        sess = tf.get_default_session()
-        assert sess is not None
         countholder = [0] # Counts how many times _collect_grads has been called
         stat = tf.reduce_sum(grads_and_vars[0][1]) # sum of first variable
-        def _collect_grads(flat_grad):
+        def _collect_grads(flat_grad, np_stat):
             self.comm.Allreduce(flat_grad, buf, op=MPI.SUM)
             np.divide(buf, float(num_tasks), out=buf)
             if countholder[0] % 100 == 0:
-                check_synced(sess, self.comm, stat)
+                check_synced(np_stat, self.comm)
             countholder[0] += 1
             return buf
 
-        avg_flat_grad = tf.py_func(_collect_grads, [flat_grad], tf.float32)
+        avg_flat_grad = tf.py_func(_collect_grads, [flat_grad, stat], tf.float32)
         avg_flat_grad.set_shape(flat_grad.shape)
         avg_grads = tf.split(avg_flat_grad, sizes, axis=0)
         avg_grads_and_vars = [(tf.reshape(g, v.shape), v)
                     for g, (_, v) in zip(avg_grads, grads_and_vars)]
         return avg_grads_and_vars
 
-def check_synced(sess, comm, tfstat):
+def check_synced(localval, comm=None):
     """
-    Check that 'tfstat' evaluates to the same thing on every MPI worker
+    It's common to forget to initialize your variables to the same values, or
+    (less commonly) if you update them in some other way than adam, to get them out of sync.
+    This function checks that variables on all MPI workers are the same, and raises
+    an AssertionError otherwise
+
+    Arguments:
+        comm: MPI communicator
+        localval: list of local variables (list of variables on current worker to be compared with the other workers)
     """
-    localval = sess.run(tfstat)
+    comm = comm or MPI.COMM_WORLD
     vals = comm.gather(localval)
     if comm.rank == 0:
         assert all(val==vals[0] for val in vals[1:])
+
+
+@with_mpi(timeout=5)
+def test_nonfreeze():
+    np.random.seed(0)
+    tf.set_random_seed(0)
+
+    a = tf.Variable(np.random.randn(3).astype('float32'))
+    b = tf.Variable(np.random.randn(2,5).astype('float32'))
+    loss = tf.reduce_sum(tf.square(a)) + tf.reduce_sum(tf.sin(b))
+
+    stepsize = 1e-2
+    # for some reason the session config with inter_op_parallelism_threads was causing
+    # nested sess.run calls to freeze
+    config = tf.ConfigProto(inter_op_parallelism_threads=1)
+    sess = U.get_session(config=config)
+    update_op = MpiAdamOptimizer(comm=MPI.COMM_WORLD, learning_rate=stepsize).minimize(loss)
+    sess.run(tf.global_variables_initializer())
+    losslist_ref = []
+    for i in range(100):
+        l,_ = sess.run([loss, update_op])
+        print(i, l)
+        losslist_ref.append(l)

baselines/ppo2/ppo2.py

@@ -160,7 +160,6 @@ def learn(*, network, env, total_timesteps, eval_env = None, seed=None, nsteps=2
             envsperbatch = nenvs // nminibatches
             envinds = np.arange(nenvs)
             flatinds = np.arange(nenvs * nsteps).reshape(nenvs, nsteps)
-            envsperbatch = nbatch_train // nsteps
             for _ in range(noptepochs):
                 np.random.shuffle(envinds)
                 for start in range(0, nenvs, envsperbatch):