more fixes

a2c_discrete.py

@@ -1,50 +1,15 @@
 import tr_helpers
-import wrappers
 import tensorflow as tf
 import numpy as np
 import collections
 import time
 from collections import deque, OrderedDict
 from tensorboardX import SummaryWriter
-from tensorflow_utils import TensorFlowVariables
 import gym
 import vecenv
 from tf_moving_mean_std import MovingMeanStd
 
 
-class RewardsMap:
-    def __init__(self, num_actors, width = 40):
-        self.width = width
-        self.height = width
-        self.num_actors = num_actors
-        self.array = np.zeros((num_actors, width, width))
-        self.pos = np.zeros((num_actors, 2))
-
-    def update(self, v, dones):
-        v_scale = 8.0 / 10.0 
-        pos_shift = self.width // 2
-        self.pos = self.pos + v * v_scale
-
-        coords = self.pos.astype(np.int32) + pos_shift
-        coords = np.clip(coords, 0, self.width-1)
-        rewards = np.zeros(self.num_actors)
-        print(self.pos[0])
-        print(coords[0])
-        print(v[0])
-        for i in range(self.num_actors):
-            if (self.array[i, coords[i, 0], coords[i, 1]]) == 0:
-                self.array[i, coords[i, 0], coords[i, 1]] = 1
-            else:
-                rewards[i] = -1.0
-            self.array[i, :] = self.array[i, :] * (1.0 - dones[i])
-            self.pos[i, :] = self.pos[i, :] * (1.0 - dones[i])
-
-        #self.array = self.array * (1.0 - dones)
-        #self.pos = self.pos * (1.0 - dones)
-        return rewards
-
-
-
 def swap_and_flatten01(arr):
     """
     swap and then flatten axes 0 and 1

download_networks.py

@@ -2,6 +2,7 @@
 import os
 import zipfile
 
+
 def download_file_from_google_drive(id, destination):
     URL = "https://docs.google.com/uc?export=download"
 
@@ -35,10 +36,11 @@ def save_response_content(response, destination):
     file_id = '1mVGfN0Xoj--26_NIPgXS48ub-Rr89GIw'
     destination = os.path.dirname(os.path.abspath(__file__)) + '/networks.zip'
     extract_destination = os.path.dirname(os.path.abspath(__file__)) + '/nn'
+    print('Starting networks download')
     if os.path.isfile(destination):
         print('Networks were already downloaded')
     else:
         download_file_from_google_drive(file_id, destination)
         print('Networks were succesfully downloaded.')
         with zipfile.ZipFile(destination, 'r') as zip_ref:
-            zip_ref.extractall(extract_destination)
+            zip_ref.extractall(extract_destination)

requirements.txt

@@ -7,6 +7,11 @@ tensorboard==1.14.0
 tensorboardX==1.6
 tensorflow-gpu==1.14.0
 gast==0.2.2
+Pillow>=4.2.1,<=5.4.1
+protobuf>=3.6,<3.7
+grpcio>=1.11.0,<1.12.0
+pyyaml>=5.1
+jsonpickle>=1.2
 
 
 

test_a2c.ipynb

@@ -46,9 +46,7 @@
    "source": [
     "import env_configurations\n",
     "import games_configurations\n",
-    "env_name = 'AnimalAIRay'#'PongNoFrameskip-v4' #'MountainCar-v0' #'SuperMarioBros-v1'# 'PongNoFrameskip-v4' #SuperMarioBros-v1'\n",
-    "#env_name = 'CartPole-v1'#'CartPole-v1' #'RoboschoolAnt-v1' #'CarRacing-v0' #'LunarLander-v2' #'Acrobot-v1' #\n",
-    "\n",
+    "env_name = 'AnimalAIRay'\n",
     "obs_space, action_space = env_configurations.get_obs_and_action_spaces('AnimalAI')\n",
     "config = games_configurations.animal_ai_ray_times3\n",
     "\n",
@@ -56141,7 +56139,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "agent.save('nn/last84_10_6')\n",
+    "agent.save('nn/last84_10_7')\n",
     "ray.shutdown()"
    ]
   },
@@ -56167,9 +56165,7 @@
    "execution_count": null,
    "metadata": {},
    "outputs": [],
-   "source": [
-    "gym.envs.registry.all()"
-   ]
+   "source": []
   },
   {
    "cell_type": "code",
@@ -56207,4 +56203,4 @@
  },
  "nbformat": 4,
  "nbformat_minor": 2
-}
+}