def test_heatmap_coverage():
"""Execute Heatmap code to see if most of the things run without crashing"""
import random
random.seed(0)
import numpy as np
import matplotlib.pyplot as plt
np.set_printoptions(linewidth=200, precision=2)
import models
import robots
import models.forward as forward
import models.testbed as testbed
import models.plots.heat as heat
# Robot
#vm = robots.VowelModel()
vm = robots.KinematicArm2D(dim=3)
# Forward Model
fwd = forward.WeightedNNForwardModel.from_robot(vm)
# Testbed
tb = testbed.Testbed(vm, fwd)
# Training model and creating tests
cases = tb.uniform_motor(10)
for x, y in cases:
fwd.add_xy(x, y)
tb.reset_testcases()
tb.uniform_sensor(10)
# Heatmap
hmf = heat.HotTestbed(tb)
hmf.plot_fwd(res=5)
return True
def test_testbed_coverage():
"""Typical code usage of testbed."""
import robots
import models.learner as learner
import models.testbed as testbed
import random
random.seed(0)
# Robot
arm = robots.KinematicArm2D(dim = 3)
# Instanciating testbed
fwdlearners = [learner.Learner.from_robot(arm, fwd = fwd, inv = 'NN') for fwd in learner.fwdclass]
testbeds = [testbed.Testbed.from_learner(arm, learnr) for learnr in fwdlearners]
# Sharing training
tb0 = testbeds[0]
tb0.train_motor(10)
for tb in testbeds:
tb.fmodel.dataset = tb0.fmodel.dataset
# Sharing testcases
tb0.uniform_motor(10)
for tb in testbeds:
tb.testcases = tb0.testcases
# Testing
for tb in testbeds:
errors = tb.run_forward()
avg, std = tb.avg_std(errors)
fwdname = tb.fmodel.__class__.__name__
return True
def test_kin():
"""Test if KinematicArm2D instanciates properly"""
check = True
arm = robots.KinematicArm2D(dim = 6)
self.assertEqual(arm.dim, 6)
cfg = forest.Tree()
cfg.dim = 1
cfg.lengths = 1.0
cfg.limits = (-1.0, 1.0)
arm = robots.KinematicArm2D(cfg)
self.assertEqual(arm.dim, 1)
cfg.dim = 6
arm = robots.KinematicArm2D(cfg)
self.assertEqual(arm.m_feats, tuple(range(-6, 0)))
self.assertEqual(arm.s_feats, (0, 1))
self.assertEqual(arm.m_bounds, 6*((-1.0, 1.0),))
def test_curve_coverage(): """Instanciate all forward models to being tested.""" # Robots import robots arm = robots.KinematicArm2D(dim=3) # Learners fwdlearners = [learner.Learner.from_robot(vm, fwd=fwd, inv='NN') for fwd in learner.fwdclass.keys()] curve = ErrorCurve(arm, fwdlearners, side='forward', trials=30, tests=2, uniformity='sensor') curve.plot() return True
def test_learner_coverage():
"""Instanciate every combination of fwd and inverse model in a learner"""
# Robots
import robots
#vm = robots.VowelModel()
vm = robots.KinematicArm2D(dim=3)
# Learners
l = learner.Learner.from_robot(vm)
for fwd in learner.fwdclass.keys():
for inv in learner.invclass.keys():
l = learner.Learner.from_robot(vm, fwd=fwd, inv=inv)
return True
import random
import robots
import models.learner
# Creating a 6DOF robotic arm
arm6DOF = robots.KinematicArm2D(dim=6)
# Creating a learner, regrouping a forward and inverse model
learner = models.learner.Learner.from_robot(arm6DOF, fwd = 'WNN', inv = 'L-BFGS-B')
# Training the learner on 1000 examples
for i in range(1000):
order = [random.uniform(-90, 90) for _ in range(6)] # random order of dim 6
effect = arm6DOF.execute_order(order)
learner.add_xy(order, effect)
# Predicting effects.
print learner.predict_effect((14.0, 20.0, -35.0, 61.0, -11.0, 79.0))
# Infering orders.
x = learner.infer_order((1.5, 5.8))
y = arm6DOF.execute_order(x)
print y
"""
Inverse model benchmark.
Inverse model performance is tested in function of the forward model,
the inverse and forward parameters and the robots.
"""
import random
random.seed(30)
import robots
from models.learner import Learner
from models.plots.ercurve import ErrorCurve
# Robot
arm = robots.KinematicArm2D(dimM=6)
# Learner
invlearners = (
[Learner.from_robot(arm, fwd='WNN', inv='BFGS', k=5, sigma=5.0)]
#+ [Learner.from_robot(arm, fwd = 'WNN', inv = 'BFGS', sigma = 10.0)]
#+ [Learner.from_robot(arm, fwd = 'AvgNN', inv = 'BFGS', k = 5, sigma = 5.0)]
+ [Learner.from_robot(arm, fwd='ES-LWLR', inv='BFGS', k=20)])
# Plot
curve = ErrorCurve(arm,
invlearners,
side='inverse',
trials=1000,
tests=250,
uniformity='sensor')
curve.plot()
import robots print 'Loading 6 DOFs kinematic arm...' robotArm = robots.KinematicArm2D(dim=6) print ' m_feats : %s' % (robotArm.m_feats, ) print ' m_bounds : %s' % (robotArm.m_bounds, ) print ' s_feats : %s' % (robotArm.s_feats, ) print '' print 'Executing orders...' order = tuple(0.0 for _ in range(6)) print ' order : %s\t result : %s' % (order, robotArm.execute_order(order)) order = tuple(10.0 for _ in range(6)) print ' order : %s\t result : %s' % (order, robotArm.execute_order(order))
"""A straightforward example of model benchmarking"""
import robots
import models.learner as learner
import models.testbed as testbed
import random
random.seed(0)
# Robot
arm = robots.KinematicArm2D(dim=3)
# Instanciating testbed
fwdlearners = [
learner.Learner.from_robot(arm, fwd=fwd, inv='NN')
for fwd in learner.fwdclass
]
testbeds = [
testbed.Testbed.from_learner(arm, learnr) for learnr in fwdlearners
]
# Sharing training
tb0 = testbeds[0]
tb0.train_motor(100)
for tb in testbeds:
tb.fmodel.dataset = tb0.fmodel.dataset
# Sharing testcases
tb0.uniform_motor(100)
for tb in testbeds:
tb.testcases = tb0.testcases
"""Benchmarking a simple 3 DOF, 2D Kinematic Arm in function of timesteps.""" import random random.seed(0) import forest import robots from models.learner import Learner, fwdclass from models.plots.ercurve import ErrorCurve # Robot cfg = forest.Tree() cfg.dim = 6 cfg.lengths = 1.0 cfg.limits = (-360.0, 360.0) arm = robots.KinematicArm2D(cfg) fwdlearners = [] fwdlearners.append(Learner.from_robot(arm, fwd = 'ES-LWLR', inv = 'L-BFGS-B', k = 2*cfg.dim+1)) curve = ErrorCurve(arm, fwdlearners, side = 'forward', trials = 50000, tests = 200, uniformity = 'motor') curve.plot() curve.show()
plt.xlabel('timesteps')
plt.title(robot.__class__.__name__)
plt.ylabel('average error')
for i, tb in enumerate(testbeds):
lines.append(plt.plot(teststeps, errors_avg[i]))
plt.fill_between(teststeps,
lower_bound[i],
upper_bound[i],
facecolor=color,
alpha=0.2)
return lines, tuple(tb.fmodel.__class__.__name__ for tb in testbeds)
# Robot
arm3 = robots.KinematicArm2D(dim=3)
arm6 = robots.KinematicArm2D(dim=6)
vm = robots.VowelModel()
ergo = robots.Ergorobot()
gs = gridspec.GridSpec(2, 2)
gs.update(wspace=0.4, hspace=0.4)
handles, labels = fwdcurves(arm3, gs[0, 0])
handles, labels = fwdcurves(arm6, gs[0, 1])
handles, labels = fwdcurves(vm, gs[1, 0])
handles, labels = fwdcurves(ergo, gs[1, 1])
fig = plt.figure(1)
fig.set_facecolor('white')
fig.suptitle('Benchmarks of Forward Model for Differents Robots')
leg = fig.legend(