def main():
scriptFile = "../../taesooLib/Samples/scripts/RigidBodyWin/GUI_tools/WRLviewer.lua"
option = ''
if len(sys.argv) == 2:
scriptFile = sys.argv[1]
elif len(sys.argv) == 3:
option = sys.argv[1]
scriptFile = sys.argv[2]
uiscale = 1.5
if platform.system() == 'Darwin':
m.createMainWin(int((10 + 220) * uiscale), int((400 + 100) * uiscale),
int(10 * uiscale), int(400 * uiscale), uiscale,
"../Resource/ogreconfig_mac.txt", "plugins_mac.cfg",
"ogre_mac.cfg")
else:
if option == '--sep':
m.createMainWin(int((10 + 220) * uiscale),
int((400 + 100) * uiscale), int(10 * uiscale),
int(400 * uiscale), uiscale,
"../Resource/ogreconfig_linux_sepwin.txt",
"plugins_linux.cfg", "ogre_linux.cfg")
else:
m.createMainWin(int((1024 + 180) * uiscale),
int((600 + 100) * uiscale), int(1024 * uiscale),
int(600 * uiscale), uiscale)
m.showMainWin()
if scriptFile[0:1] != '.':
scriptFile = os.path.relpath(scriptFile, os.getcwd())
if scriptFile[-3:] != 'lua':
scriptFile = scriptFile + '.lua'
print('loading', scriptFile)
m.getPythonWin().loadScript(scriptFile)
# you can use the following three lines instead of the above line.
# m.getPythonWin().loadEmptyScript()
# m.getPythonWin().dofile(scriptFile)
# m.getPythonWin().dostring('ctor()')
print("ctor finished")
# test lua-python interfacing
lua.out(3)
lua.out('asdf')
lua.out(m.vector3(3, 4, 5))
a = m.vectorn()
a.assign([1, 2, 3, 4, 5])
lua.out(a)
b = m.intvectorn()
b.assign([1, 2, 3, 4, 5])
lua.out(b)
c = np.mat('1 0 0; 0 1 0; 0 0 1')
print(c)
m.startMainLoop() # this finishes when program finishes
def frameMove(fElapsedTime):
l = m.getPythonWin()
if not l.isLuaReady(): return
# lua function call
l.getglobal("frameMove")
l.push(fElapsedTime)
l.call(1, 0)
def onFrameChanged(currFrame):
l = m.getPythonWin()
if not l.isLuaReady(): return
# lua function call
l.getglobal("onFrameChanged")
l.push(currFrame)
l.call(1, 0)
def out(*args):
l = m.getPythonWin()
if not l.isLuaReady(): return
# lua function call
l.getglobal("print")
n = len(args)
for i in range(n):
l.push(args[i])
l.call(n, 0)
def F1_vec(*args):
#l.printStack()
_F(1, *args)
#l.printStack()
l = m.getPythonWin()
if not l.isnil(l.gettop()):
vec = l.popvectorn()
return vec.ref().tolist()
return None
def main():
#scriptFile= 'gym_walk/testPendulumOnlineControlWalkObstacle.lua'
scriptFile = 'gym_walk/testPendulumOnlineControlWalkEnv.lua'
#scriptFile= 'gym_walk/testPendulumOnlineControlWalk.lua'
option = ''
if len(sys.argv) == 2:
scriptFile = sys.argv[1]
elif len(sys.argv) == 3:
option = sys.argv[1]
scriptFile = sys.argv[2]
uiscale = 1.5
if platform.system() == 'Darwin':
uiscale = 1.0
m.createMainWin(int((10 + 220) * uiscale), int((400 + 100) * uiscale),
int(10 * uiscale), int(400 * uiscale), uiscale,
"../Resource/ogreconfig_mac.txt", "plugins_mac.cfg",
"ogre_mac.cfg")
else:
if option == '--sep':
if m.getOgreVersionMinor() >= 12:
m.createMainWin(int((10 + 220) * uiscale),
int((400 + 100) * uiscale), int(10 * uiscale),
int(400 * uiscale), uiscale,
"../Resource/ogreconfig_linux_sepwin.txt",
"plugins_linux12.cfg", "ogre_linux12.cfg")
else:
m.createMainWin(int((10 + 220) * uiscale),
int((400 + 100) * uiscale), int(10 * uiscale),
int(400 * uiscale), uiscale,
"../Resource/ogreconfig_linux_sepwin.txt",
"plugins_linux.cfg", "ogre_linux.cfg")
else:
m.createMainWin(int((1024 + 180) * uiscale),
int((600 + 100) * uiscale), int(1024 * uiscale),
int(600 * uiscale), uiscale)
m.showMainWin()
m.getPythonWin().loadEmptyScript()
m.getPythonWin().dofile(scriptFile)
lua.dostring('useNN=true ctor()')
print("ctor finished")
m.startMainLoop() # this finishes when program finishes
def getglobal_mat(*args):
varname = args[0]
l = m.getPythonWin()
if not l.isLuaReady(): return
l.getglobal(varname)
for i in range(1, len(args)):
l.replaceTop(args[i])
mat = l.popmatrixn()
return mat.ref()
def onCallback(mid, userdata):
l = m.getPythonWin()
if not l.isLuaReady(): return
# lua function call
#l.dostring("function test(a,b) print(a,b) end")
#l.getglobal("test")
l.getglobal("onCallback")
l.push(mid)
l.push(userdata)
#l.printStack()
l.call(2, 0)
def _F(numout, *args):
funcname = args[0]
l = m.getPythonWin()
if not l.isLuaReady(): return
if type(funcname).__name__ == 'tuple':
l.getglobal(*funcname)
else:
l.getglobal(funcname)
for i in range(1, len(args)):
if type(args[i]).__name__ == 'ndarray':
l.push(array(args[i]))
else:
l.push(args[i])
l.call(len(args) - 1, numout)
def main():
lua.init_console()
lua.dostring("g_dataset=util.loadTable('feature_data.dat')")
if True:
l = m.getPythonWin()
l.getglobal('g_dataset')
l.replaceTop(1)
mat = l.popmatrixn()
matfeature = lua.getglobal_mat('g_dataset', 1)
matdata = lua.getglobal_mat('g_dataset', 2)
matdotdata = lua.getglobal_mat('g_dataset', 3)
train(matfeature, matdata, 'walkrun.net')
def handleRendererEvent(ev, button, x, y):
#print( ev, button, x, y)
l = m.getPythonWin()
#l.printStack()
if not l.isLuaReady(): return 0
l.getglobal("handleRendererEvent")
if not l.isnil(l.gettop()):
l.push(ev)
l.push(button)
l.push(x)
l.push(y)
l.call(4, 1)
return l.popnumber()
else:
l.pop()
return 0
def Fref(*args):
funcname = args[0]
l = m.getPythonWin()
if not l.isLuaReady(): return
if type(funcname).__name__ == 'tuple':
l.getglobal(*funcname)
else:
l.getglobal(funcname)
out = [None] * len(args)
for i in range(1, len(args)):
if type(args[i]).__name__ == 'ndarray':
if len(args[i].shape) == 1:
temp = m.vectorn()
temp.assign(args[i])
l.push(temp)
out[i] = temp
elif len(args[i].shape) == 2:
temp = m.matrixn(args[i].shape[0], args[i].shape[1])
for j in range(temp.rows()):
temp.row(j).assign(args[i][j])
l.push(temp)
out[i] = temp
else:
assert (False)
else:
l.push(args[i])
l.call(len(args) - 1, 0)
for i in range(1, len(args)):
if type(args[i]).__name__ == 'ndarray':
if len(args[i].shape) == 1:
args[i][:] = out[i].tolist()
elif len(args[i].shape) == 2:
args[i][:] = out[i].tolist()
else:
assert (False)
def train(matfeature, matdata, filename):
x = np.array(matfeature)
y = np.array(matdata)
x = torch.tensor(x.astype(np.float32))
y = torch.tensor(y.astype(np.float32))
b_eval = True
if b_eval:
lua.dostring("g_dataset_eval=util.loadTable('feature_data_eval.dat')")
l = m.getPythonWin()
l.getglobal('g_dataset_eval')
l.replaceTop(1)
mat2 = l.popmatrixn()
matfeature_eval = lua.getglobal_mat('g_dataset_eval', 1)
matdata_eval = lua.getglobal_mat('g_dataset_eval', 2)
matdotdata_eval = lua.getglobal_mat('g_dataset_eval', 3)
xe = np.array(matfeature_eval)
ye = np.array(matdata_eval)
xe = torch.tensor(xe.astype(np.float32))
ye = torch.tensor(ye.astype(np.float32))
# torch can only train on Variable, so convert them to Variable
x, y = Variable(x), Variable(y)
nin = matfeature.shape[1]
nout = y.shape[1]
print(nin, nout) # 105, 1500
nhidden1 = 50
#nhidden2=50 # works ok
#nhidden2=24 # works ok
nhidden2 = 12 # works ok
#nhidden2=15 # works ok
net = torch.nn.Sequential(
torch.nn.Dropout(p=0.01),
torch.nn.Linear(nin, nhidden1),
torch.nn.ELU(),
torch.nn.Linear(nhidden1, nhidden2),
torch.nn.ELU(),
torch.nn.Linear(nhidden2, 200),
torch.nn.ELU(),
#torch.nn.Dropout(p=0.1),
torch.nn.Linear(200, nout),
)
optimizer = torch.optim.Adam(net.parameters(), lr=0.001)
loss_func = torch.nn.MSELoss() # this is for regression mean squared loss
BATCH_SIZE = 32
EPOCH = 2000 # MSE: 0.0007 -> 3705 (15)
# -> 4401 (12)
#EPOCH = 200 # 0.0012 -> MSE: 6710
torch_dataset = Data.TensorDataset(x, y)
loader = Data.DataLoader(
dataset=torch_dataset,
batch_size=BATCH_SIZE,
shuffle=True,
num_workers=2,
)
b_plot = True
# start training
for epoch in range(EPOCH):
print('epoch', epoch)
for step, (batch_x,
batch_y) in enumerate(loader): # for each training step
#print(step)
b_x = Variable(batch_x)
b_y = Variable(batch_y)
prediction = net(b_x) # input x and predict based on x
loss = loss_func(prediction,
b_y) # must be (1. nn output, 2. target)
#regularity = torch.norm(net[0].weight, p=1) + torch.norm(net[2].weight, p=1)
#cost = loss + 0.01*regularity
if step == 1:
print('epoch', epoch, 'loss', loss.data.numpy())
optimizer.zero_grad() # clear gradients for next train
loss.backward() # backpropagation, compute gradients
#cost.backward() # backpropagation, compute gradients
optimizer.step() # apply gradients
if b_eval:
prediction = net(x) # input x and predict based on x
predictione = net(xe) # input x and predict based on x
print(':', epoch, ((y - prediction)**2).mean().detach().numpy(),
((ye - predictione)**2).mean().detach().numpy())
if b_plot:
prediction = net(x) # input x and predict based on x
print(epoch, ((y - prediction)**2).mean())
if b_plot:
torch.save(net, filename + '.plot.dat')
else:
torch.save(net, filename)
prediction = net(x) # input x and predict based on x
print('final MSE=', torch.sum((y - prediction)**2))
print('final MSE col0=', torch.sum((y[:, 0] - prediction[:, 0])**2))
def F1_mat(*args):
_F(1, *args)
l = m.getPythonWin()
mat = l.popmatrixn()
return mat.ref().tolist()
def main():
global g_globals
uiscale = 1.5
m.createMainWin(int((1024 + 180) * uiscale), int((600 + 100) * uiscale),
int(1024 * uiscale), int(600 * uiscale), uiscale)
m.showMainWin()
m.getPythonWin().loadEmptyScript()
parser = argparse.ArgumentParser(description='RL')
parser.add_argument('--seed',
type=int,
default=1,
help='random seed (default: 1)')
parser.add_argument(
'--log-interval',
type=int,
default=10,
help='log interval, one log per n updates (default: 10)')
parser.add_argument(
'--env-name',
default='PongNoFrameskip-v4',
help='environment to train on (default: PongNoFrameskip-v4)')
parser.add_argument(
'--load-dir',
default='./trained_models/',
help='directory to save agent logs (default: ./trained_models/)')
parser.add_argument('--non-det',
action='store_true',
default=False,
help='whether to use a non-deterministic policy')
args = parser.parse_args()
args.det = not args.non_det
env = make_vec_envs(args.env_name,
args.seed + 1000,
1,
None,
None,
device='cpu',
allow_early_resets=False)
# Get a render function
render_func = get_render_func(env)
# We need to use the same statistics for normalization as used in training
actor_critic, ob_rms = \
torch.load(os.path.join(args.load_dir, args.env_name + ".pt"))
vec_norm = get_vec_normalize(env)
if vec_norm is not None:
vec_norm.eval()
vec_norm.ob_rms = ob_rms
recurrent_hidden_states = torch.zeros(
1, actor_critic.recurrent_hidden_state_size)
masks = torch.zeros(1, 1)
obs = env.reset()
if render_func is not None:
render_func('human')
g_globals = (actor_critic, obs, recurrent_hidden_states, masks, args, env,
render_func)
m.startMainLoop() # this finishes when program finishes
def dofile(fn):
m.getPythonWin().dofile(fn)
def main():
lua.init_console()
# now you can use lua funtions.
# test lua-python interfacing
lua.out(3)
lua.out('asdf')
lua.out(m.vector3(3, 4, 5))
a = m.vectorn()
a.assign([1, 2, 3, 4, 5])
lua.out(a)
a.assign(np.array([5, 4, 3, 2, 1]).tolist())
lua.out(a)
print(a.ref())
a.ref()[1] = 10
lua.out(a)
b = m.intvectorn()
b.assign([1, 2, 3, 4, 5])
lua.out(b)
c = np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1]])
print(c)
# lua.array converts ndarray to vectorn, matrixn or hypermatrixn
lua.out(lua.array(c))
d = m.matrixn(3, 3)
lua.out(d.row(0))
d.row(0).assign([1, 2, 3])
d.row(1).assign([7, 3, 4])
d.row(2).assign([3, 4, 5])
lua.out(d)
print(d.row(0).ref())
print(d.column(0).ref())
print(d.column(0).ref().tolist())
print(d.ref().tolist())
e = m.matrixn(10, 10)
e.setAllValue(0)
e.row(1).setAllValue(10)
e.column(1).setAllValue(20)
e.ref()[0, 0] = 10
f = e.range(5, 7, 5, 7).ref()
f[0, 0] = 22
f[1, 1] = 23
print(e.ref())
# low-level apis
l = m.getPythonWin()
l.dostring('g_dataset={ 1,2,3}')
# out=g_dataset[2]
l.getglobal('g_dataset')
l.replaceTop(2)
print('the second element in the table is ',
l.popnumber()) # other types are also supported l.popmatrixn()
print('Starting python console. Type "cont" to finish.')
print("""
try the followings:
v=m.vector3()
v.x=3
lua.out(v)
v=m.vectorn()
v.assign([1,2,3,4,5])
lua.out(v)
v.set(0,4)
lua.out(v)
...
""")
pdb.set_trace()
print('...')
print('Starting python interactive console. Type "ctrl+d" to finish.')
print('Try the same examples above:')
code.interact(local=dict(globals(), **locals()))
print('Starting lua console. Type "cont" to finish.')
m.getPythonWin().dostring('dbg.console()')
def init_console():
# absolutely necessary. The mainWin won't be shown without m.showMainWin() though.
uiscale = 1.5
m.createMainWin(int((1024 + 180) * uiscale), int((600 + 100) * uiscale),
int(1024 * uiscale), int(600 * uiscale), uiscale)
m.getPythonWin().loadEmptyScript()
def dostring(arg):
l = m.getPythonWin()
if not l.isLuaReady(): return
l.dostring(arg)