def testMG(self, plot=False):
c = config.copy()
c["source"] = {"class": OpMackeyGlass}
c["train"]["layer_sizes"] = (5,)
c["train"]["layer_classes"] = (mlp.Sigmoid,)
init_1 = {"class": PCAWeightInitializer}
init_2 = {"class": LeastSquaresWeightInitializer}
c["train"]["weight_initializer"] = (init_1, init_2)
c["train"]["learning_rate"] = .2
c["features"] = {"class": OpRecent, "window_size": 16}
c["target"] = {"class": OpExponentiallySegmentedPattern,
"baseline_size": 8,
"num_segments": 1}
w = Workflow.build(c, workingdir=self.wd)
with warnings.catch_warnings():
warnings.simplefilter("ignore")
w.run()
if plot:
plt.figure()
pred = w._predictionCache.Output[...].wait().squeeze()
target = w._target.Output[...].wait().squeeze()
orig = w._source.Output[...].wait().squeeze()
plt.plot(target, 'b')
plt.plot(pred, 'r+')
plt.plot(orig, 'k.')
plt.legend(("ground truth", "prediction", "original data"))
def testRun(self, plot=False):
c = config.copy()
c["train"]["layer_sizes"] = (20, 10)
c["train"]["layer_classes"] = (mlp.Sigmoid, mlp.Sigmoid)
w = Workflow.build(c, workingdir=self.wd)
with warnings.catch_warnings():
warnings.simplefilter("ignore")
w.run()
return w
def testRF(self):
d = tempfile.mkdtemp()
try:
c = config.copy()
c["train"] = {"class": OpRFTrain}
c["predict"] = {"class": OpRFPredict}
w = Workflow.build(c, workingdir=d)
w.run()
except:
raise
finally:
shutil.rmtree(d)
def testIncompatible(self):
c = config.copy()
def foo():
with self.assertRaises(IncompatibleTargets):
w = Workflow.build(c, workingdir=self.wd)
w.run()
c["target"] = {"class": OpFeatures}
foo()
c["target"] = {"class": OpRegTarget}
c["train"] = {"class": OpRFTrain}
foo()
c["train"] = {"class": OpSVMTrain}
c["predict"] = {"class": OpRFPredict}
foo()
c["predict"] = {"class": OpSVMPredict}
c["report"] = {"class": OpClassificationReport}
foo()
c["report"] = {"class": OpRegressionReport}
Workflow.build(c, workingdir=self.wd)
def testNormTarget(self, plot=False):
c = config.copy()
c["source"] = {"class": OpRandomUnitSquare,
"shape": (10000, 2)}
c["train"]["layer_sizes"] = (1,)
c["train"]["layer_classes"] = (mlp.Sigmoid,)
c["features"] = {"class": OpFeatures}
c["target"] = {"class": OpNormTarget}
w = Workflow.build(c, workingdir=self.wd)
with warnings.catch_warnings():
warnings.simplefilter("ignore")
w.run()
def testDNN(self):
d = tempfile.mkdtemp()
try:
c = config.copy()
c["train"] = {"class": OpDeepTrain,
"num_hidden_layers": 2,
"size_hidden_layers": (2, 2)}
c["predict"] = {"class": OpMLPPredict}
w = Workflow.build(c, workingdir=d)
with warnings.catch_warnings():
warnings.simplefilter("ignore")
w.run()
except:
raise
finally:
shutil.rmtree(d)
def testMLP(self):
d = tempfile.mkdtemp()
try:
c = config.copy()
c["train"] = {"class": OpMLPTrain,
"layer_classes": (mlp.Sigmoid,),
"layer_sizes": (5,)}
c["predict"] = {"class": OpMLPPredict}
w = Workflow.build(c, workingdir=d)
with warnings.catch_warnings():
warnings.simplefilter("ignore")
w.run()
except:
raise
finally:
shutil.rmtree(d)
def main(args):
conf = _get_conf(args)
tempdir = args.workingdir
if tempdir is None:
prefix = "{}_{:%Y-%m-%d_%H-%M}".format(args.mode, datetime.now())
tempdir = tempfile.mkdtemp(prefix=prefix)
workflow = Workflow.build(conf, workingdir=tempdir)
with warnings.catch_warnings():
warnings.simplefilter("ignore")
workflow.run()
if args.plot:
assert _plot_available, "matplotlib is needed for option --plot"
ground_truth = workflow.Target[...].wait()
prediction = workflow.Prediction[...].wait()
plt.figure()
plt.hold(True)
plt.plot(ground_truth, 'b')
plt.plot(prediction, 'r')
plt.hold(False)
plt.show()
print("output written to dir {}".format(tempdir))
def run(args):
c = getConfig(args)
c["workingdir"] = args.workingdir
w = Workflow.build(c)
w.run()
print("Working directory: {}".format(w._workingdir))
def foo():
with self.assertRaises(IncompatibleTargets):
w = Workflow.build(c, workingdir=self.wd)
w.run()
def testBasic(self):
w = Workflow.build(config, workingdir=self.wd)
w.run()