Parameter space noise for DQN and DDPG (openai#75)

* Export param noise

* Update documentation

* Final finishing touches

Author: matthiasplappert

README.md

@@ -15,3 +15,4 @@ pip install baselines
 - [DQN](baselines/deepq)
 - [PPO](baselines/pposgd)
 - [TRPO](baselines/trpo_mpi)
+- [DDPG](baselines/ddpg)

baselines/common/azure_utils.py

@@ -3,7 +3,10 @@
 import zipfile
 
 from azure.common import AzureMissingResourceHttpError
-from azure.storage.blob import BlobService
+try:
+    from azure.storage.blob import BlobService
+except ImportError:
+    from azure.storage.blob import BlockBlobService as BlobService
 from shutil import unpack_archive
 from threading import Event
 
@@ -114,18 +117,23 @@ def progress_callback(current, total):
             arcpath = os.path.join(td, "archive.zip")
             for backup_blob_name in [blob_name, blob_name + '.backup']:
                 try:
-                    blob_size = self._service.get_blob_properties(
+                    properties = self._service.get_blob_properties(
                         blob_name=backup_blob_name,
                         container_name=self._container_name
-                    )['content-length']
+                    )
+                    if hasattr(properties, 'properties'):
+                        # Annoyingly, Azure has changed the API and this now returns a blob
+                        # instead of it's properties with up-to-date azure package.
+                        blob_size = properties.properties.content_length
+                    else:
+                        blob_size = properties['content-length']
                     if int(blob_size) > 0:
                         self._service.get_blob_to_path(
                             container_name=self._container_name,
                             blob_name=backup_blob_name,
                             file_path=arcpath,
                             max_connections=4,
-                            progress_callback=progress_callback,
-                            max_retries=10)
+                            progress_callback=progress_callback)
                         unpack_archive(arcpath, dest_path)
                         download_done.wait()
                         return True

baselines/common/misc_util.py

@@ -237,8 +237,9 @@ def boolean_flag(parser, name, default=False, help=None):
     help: str
         help string for the flag
     """
-    parser.add_argument("--" + name, action="store_true", default=default, help=help)
-    parser.add_argument("--no-" + name, action="store_false", dest=name)
+    dest = name.replace('-', '_')
+    parser.add_argument("--" + name, action="store_true", default=default, dest=dest, help=help)
+    parser.add_argument("--no-" + name, action="store_false", dest=dest)
 
 
 def get_wrapper_by_name(env, classname):

baselines/common/mpi_fork.py

@@ -1,6 +1,6 @@
 import os, subprocess, sys
 
-def mpi_fork(n):
+def mpi_fork(n, bind_to_core=False):
     """Re-launches the current script with workers
     Returns "parent" for original parent, "child" for MPI children
     """
@@ -13,7 +13,11 @@ def mpi_fork(n):
             OMP_NUM_THREADS="1",
             IN_MPI="1"
         )
-        subprocess.check_call(["mpirun", "-np", str(n), sys.executable] + sys.argv, env=env)
+        args = ["mpirun", "-np", str(n)]
+        if bind_to_core:
+            args += ["-bind-to", "core"]
+        args += [sys.executable] + sys.argv
+        subprocess.check_call(args, env=env)
         return "parent"
     else:
         return "child"