From a8fc6b59ef1327143e970082ca7b1892febfcb08 Mon Sep 17 00:00:00 2001
From: hagrid67 <jdhwatson@gmail.com>
Date: Fri, 26 Apr 2019 23:30:49 +0100
Subject: [PATCH] added basic agents to editor
---
examples/play_model.py | 76 +++++++++++++++++++++++++++++++++++
flatland/utils/editor.py | 61 +++++++++++++++++++++++++---
flatland/utils/rendertools.py | 9 +----
3 files changed, 132 insertions(+), 14 deletions(-)
diff --git a/examples/play_model.py b/examples/play_model.py
index 6a67397..68530e6 100644
--- a/examples/play_model.py
+++ b/examples/play_model.py
@@ -10,6 +10,82 @@ import matplotlib.pyplot as plt
import time
+
+class Player(object):
+ def __init__(self, env):
+ self.env = env
+ self.handle = env.get_agent_handles()
+
+ self.state_size = 105
+ self.action_size = 4
+ self.n_trials = 9999
+ self.eps = 1.
+ self.eps_end = 0.005
+ self.eps_decay = 0.998
+ self.action_dict = dict()
+ self.scores_window = deque(maxlen=100)
+ self.done_window = deque(maxlen=100)
+ self.scores = []
+ self.dones_list = []
+ self.action_prob = [0]*4
+ self.agent = Agent(self.state_size, self.action_size, "FC", 0)
+ self.agent.qnetwork_local.load_state_dict(torch.load('../flatland/baselines/Nets/avoid_checkpoint9900.pth'))
+
+ self.iFrame = 0
+ self.tStart = time.time()
+
+ # Reset environment
+ self.obs = self.env.reset()
+ for a in range(self.env.number_of_agents):
+ norm = max(1, max_lt(self.obs[a], np.inf))
+ self.obs[a] = np.clip(np.array(self.obs[a]) / norm, -1, 1)
+
+ # env.obs_builder.util_print_obs_subtree(tree=obs[0], num_elements_per_node=5)
+
+ self.score = 0
+ self.env_done = 0
+
+ def step(self):
+ env = self.env
+ for a in range(env.number_of_agents):
+ action = self.agent.act(np.array(self.obs[a]), eps=self.eps)
+ self.action_prob[action] += 1
+ self.action_dict.update({a: action})
+
+ # Environment step
+ next_obs, all_rewards, done, _ = self.env.step(self.action_dict)
+
+ for a in range(env.number_of_agents):
+ norm = max(1, max_lt(next_obs[a], np.inf))
+ next_obs[a] = np.clip(np.array(next_obs[a]) / norm, -1, 1)
+
+ # Update replay buffer and train agent
+ for a in range(self.env.number_of_agents):
+ self.agent.step(self.obs[a], self.action_dict[a], all_rewards[a], next_obs[a], done[a])
+ self.score += all_rewards[a]
+
+ self.iFrame += 1
+
+ self.obs = next_obs.copy()
+ if done['__all__']:
+ self.env_done = 1
+
+
+def max_lt(seq, val):
+ """
+ Return greatest item in seq for which item < val applies.
+ None is returned if seq was empty or all items in seq were >= val.
+ """
+
+ idx = len(seq)-1
+ while idx >= 0:
+ if seq[idx] < val and seq[idx] >= 0:
+ return seq[idx]
+ idx -= 1
+ return None
+
+
+
def main(render=True, delay=0.0):
random.seed(1)
diff --git a/flatland/utils/editor.py b/flatland/utils/editor.py
index 0a0bbc4..935e40d 100644
--- a/flatland/utils/editor.py
+++ b/flatland/utils/editor.py
@@ -15,6 +15,19 @@ from flatland.envs.rail_env import RailEnv, random_rail_generator
# from flatland.core.transitions import RailEnvTransitions
from flatland.core.env_observation_builder import TreeObsForRailEnv
import flatland.utils.rendertools as rt
+from examples.play_model import Player
+
+
+class View(object):
+ def __init__(self, editor):
+ self.editor = editor
+ self.oRT = rt.RenderTool(self.editor.env)
+ plt.figure(figsize=(10,10))
+ self.oRT.renderEnv(spacing=False, arrows=False, sRailColor="gray", show=False)
+ img = self.oRT.getImage()
+ plt.clf()
+ import jpy_canvas
+ self.wid_img = jpy_canvas.Canvas(img)
class JupEditor(object):
@@ -39,6 +52,8 @@ class JupEditor(object):
self.drawMode = "Draw"
self.env_filename = "temp.npy"
self.set_env(env)
+ self.iAgent = None
+ self.player = None
def set_env(self, env):
self.env = env
@@ -56,6 +71,28 @@ class JupEditor(object):
def setDrawMode(self, dEvent):
self.drawMode = dEvent["new"]
+ def on_click(self, wid, event):
+ x = event['canvasX']
+ y = event['canvasY']
+ rcCell = ((array([y, x]) - self.yxBase) / self.nPixCell).astype(int)
+
+ if self.drawMode == "Origin":
+ self.iAgent = len(self.env.agents_position)
+ self.env.agents_position.append(rcCell)
+ self.env.agents_handles.append(max(self.env.agents_handles + [-1]) + 1)
+ self.env.agents_direction.append(0)
+ self.env.agents_target.append(rcCell) # set the target to the origin initially
+ self.env.number_of_agents = self.iAgent + 1
+ self.drawMode = "Destination"
+
+ elif self.drawMode == "Destination" and self.iAgent is not None:
+ self.env.agents_target[self.iAgent] = rcCell
+ self.drawMode = "Origin"
+
+ self.log("agent", self.drawMode, self.iAgent, rcCell)
+
+ self.redraw()
+
def event_handler(self, wid, event):
"""Mouse motion event handler
"""
@@ -150,9 +187,6 @@ class JupEditor(object):
# This updates the image in the browser to be the new edited version
self.wid_img.data = writableData
- def on_click(self, event):
- pass
-
def redraw(self, hide_stdout=True, update=True):
if hide_stdout:
@@ -161,7 +195,8 @@ class JupEditor(object):
stdout_dest = sys.stdout
# TODO: bit of a hack - can we suppress the console messages from MPL at source?
- with redirect_stdout(stdout_dest):
+ #with redirect_stdout(stdout_dest):
+ with self.wid_output:
plt.figure(figsize=(10, 10))
self.oRT.renderEnv(spacing=False, arrows=False, sRailColor="gray", show=False)
img = self.oRT.getImage()
@@ -178,6 +213,13 @@ class JupEditor(object):
def clear(self, event):
self.env.rail.grid[:, :] = 0
+ self.env.number_of_agents = 0
+ self.env.agents_position = []
+ self.env.agents_direction = []
+ self.env.agents_handles = []
+ self.env.agents_target = []
+ self.player = None
+
self.redraw_event(event)
def setFilename(self, filename):
@@ -201,15 +243,22 @@ class JupEditor(object):
self.env = RailEnv(width=self.regen_size,
height=self.regen_size,
rail_generator=random_rail_generator(cell_type_relative_proportion=[1, 1] + [0.5] * 6),
- number_of_agents=0,
+ number_of_agents=self.env.number_of_agents,
obs_builder_object=TreeObsForRailEnv(max_depth=2))
self.env.reset()
self.set_env(self.env)
+ self.player = Player(self.env)
self.redraw()
def setRegenSize_event(self, event):
self.regen_size = event["new"]
-
+
+ def step_event(self, event=None):
+ if self.player is None:
+ self.player = Player(self.env)
+ self.player.step()
+ self.redraw()
+
def fix_env(self):
self.env.width = self.env.rail.width
self.env.height = self.env.rail.height
diff --git a/flatland/utils/rendertools.py b/flatland/utils/rendertools.py
index 1444645..6c5a175 100644
--- a/flatland/utils/rendertools.py
+++ b/flatland/utils/rendertools.py
@@ -21,7 +21,6 @@ class MPLGL(GraphicsLayer):
plt.plot(*args, **kwargs)
def scatter(self, *args, **kwargs):
- print(args, kwargs)
plt.scatter(*args, **kwargs)
def text(self, *args, **kwargs):
@@ -209,7 +208,7 @@ class RenderTool(object):
xyDir = np.matmul(rcDir, rt.grc2xy) # agent direction in xy
xyPos = np.matmul(rcPos - rcDir / 2, rt.grc2xy) + rt.xyHalf
- self.gl.scatter(*xyPos, color=color, s=40) # agent location
+ self.gl.scatter(*xyPos, color=color, marker="o", s=100) # agent location
xyDirLine = array([xyPos, xyPos + xyDir/2]).T # line for agent orient.
self.gl.plot(*xyDirLine, color=color, lw=5, ms=0, alpha=0.6)
@@ -219,12 +218,6 @@ class RenderTool(object):
xyTarget = np.matmul(rcTarget, rt.grc2xy) + rt.xyHalf
self._draw_square(xyTarget, 1/3, color)
- if False:
- # mark the next cell we're heading into
- rcNext = rcPos + rcDir
- xyNext = np.matmul(rcNext, rt.grc2xy) + rt.xyHalf
- self.gl.scatter(*xyNext, color=color)
-
def plotTrans(self, rcPos, gTransRCAg, color="r", depth=None):
"""
plot the transitions in gTransRCAg at position rcPos.
--
GitLab