def test_knn():
knn = KNN(n_neighbors=1)
chip_data1 = {'ORB_a':1, 'ORB_b':1, 'gnd':1}
chip_data2 = {'ORB_a':-1, 'ORB_b':-1, 'gnd':-1}
chip1 = Chip(chip_data1, LCT_prefix='ORB')
chip2 = Chip(chip_data2, LCT_prefix='ORB')
chips = [chip1, chip2]
knn.fit(chips)
chip3 = Chip({'ORB_a':0.9, 'ORB_b':0.9, 'gnd':1}, LCT_prefix='ORB')
assert knn.predict(chip3) == 1, 'fail: prediction not correct.'
def __init__(self, port = 5000):
self.port = port
self.chips = []
self.num_predicted = 0
self.num_chips_tested = 0
self.test_escapes = 0
self.yield_loss = 0
self.is_trained = False
self.knn = KNN()
self.c = Colors()
# ZeroMQ stuff
self.context = zmq.Context()
self.socket = self.context.socket(zmq.REQ)
# hand off the ZeroMQ context and port number to the ViewServerMixin
# superclass, for logging to view server.
super(BasicTesting, self).__init__('atesim', self.port+2, self.context)
def __init__(self, port=5000):
self.port = port
self.chips = []
self.num_predicted = 0
self.num_chips_tested = 0
self.test_escapes = 0
self.yield_loss = 0
self.is_trained = False
self.knn = KNN()
self.c = Colors()
# ZeroMQ stuff
self.context = zmq.Context()
self.socket = self.context.socket(zmq.REQ)
# hand off the ZeroMQ context and port number to the ViewServerMixin
# superclass, for logging to view server.
super(BasicTesting, self).__init__("atesim", self.port + 2, self.context)
class BasicTesting(ViewServerMixin):
def __init__(self, port=5000):
self.port = port
self.chips = []
self.num_predicted = 0
self.num_chips_tested = 0
self.test_escapes = 0
self.yield_loss = 0
self.is_trained = False
self.knn = KNN()
self.c = Colors()
# ZeroMQ stuff
self.context = zmq.Context()
self.socket = self.context.socket(zmq.REQ)
# hand off the ZeroMQ context and port number to the ViewServerMixin
# superclass, for logging to view server.
super(BasicTesting, self).__init__("atesim", self.port + 2, self.context)
def run(self):
print "Running Basic Machine learning-based testing," + "listening on port %d ..." % self.port
self.socket.connect("tcp://127.0.0.1:%d" % self.port)
try:
while True:
print "[ Train: %s %7d %s ] " % (self.c.GREEN, self.num_train_chips, self.c.ENDC),
print "[ Predicted: %s %7d %s ]" % (self.c.GREEN, self.num_predicted, self.c.ENDC)
self.get_chip()
self.update_model()
self.update_view(self.stats)
except KeyboardInterrupt:
print "\nterminating basic ML testing."
def get_chip(self):
"""Request and retrieve a chip from the ATE simulator.
"""
self.socket.send("REQ:LCT")
chip_lct = json.loads(self.socket.recv())
self.log("REQ:LCT")
self.socket.send("REQ:gnd")
chip_gnd = json.loads(self.socket.recv())
self.log("REQ:gnd")
self.socket.send("REQ:done")
assert self.socket.recv() == "RES:ack", "Error: invalid ack from ATE simulator"
self.log("REQ:done")
self.current_chip = Chip(LCT=chip_lct, gnd=chip_gnd)
self.num_chips_tested += 1
def update_model(self):
"""Add the latest chip to the chip buffer. If we've already collected
1,000 chips, then we go ahead and train the model.
"""
if self.num_chips_tested <= 1000:
self.chips.append(self.current_chip)
if self.num_chips_tested == 1000:
print "Training KNN on 1000 chips"
self.is_trained = True
self.knn.fit(self.chips)
if self.num_chips_tested >= 1000:
time.sleep(0.1)
self.num_predicted += 1
chip_gnd_predicted = self.knn.predict(self.current_chip)
self.log("Actual pf: %d" % self.current_chip.gnd)
self.log("Pred. pf: %d" % chip_gnd_predicted)
# Accounting TE / YL here
if self.current_chip.gnd == -1 and chip_gnd_predicted == 1:
self.test_escapes += 1
if self.current_chip.gnd == 1 and chip_gnd_predicted == -1:
self.yield_loss += 1
def log(self, msg):
# might expand this later
print "\t-> %s %s %s" % (self.c.RED, msg, self.c.ENDC)
@property
def num_train_chips(self):
return len(self.chips)
@property
def stats(self):
"""The self.update_view() method from the ViewServerMixin class
expects a Python object that can be JSONified. This function returns
such an object.
"""
return {
"name": "basic",
"datetime": datetime.datetime.utcnow().isoformat(),
"parms": {
"num_train_chips": {"desc": "Number of chips in training set", "value": str(self.num_train_chips)},
"num_predicted": {"desc": "Number of chips predicted with k-NN", "value": str(self.num_predicted)},
"test_escape": {
"desc": "Test escape rate",
#'value': '%1.3f%%' % (self.test_escapes * 100.0 / self.num_chips_tested)
"value": (self.test_escapes * 100.0 / self.num_chips_tested),
},
"yield_loss": {
"desc": "Yield loss rate",
#'value': '%1.3f%%' % (self.yield_loss * 100.0 / self.num_chips_tested)
"value": (self.yield_loss * 100.0 / self.num_chips_tested),
},
"is_trained": {"desc": "Model trained", "value": (lambda (x): "Yes" if x else "No")(self.is_trained)},
},
}
class BasicTesting(ViewServerMixin):
def __init__(self, port = 5000):
self.port = port
self.chips = []
self.num_predicted = 0
self.num_chips_tested = 0
self.test_escapes = 0
self.yield_loss = 0
self.is_trained = False
self.knn = KNN()
self.c = Colors()
# ZeroMQ stuff
self.context = zmq.Context()
self.socket = self.context.socket(zmq.REQ)
# hand off the ZeroMQ context and port number to the ViewServerMixin
# superclass, for logging to view server.
super(BasicTesting, self).__init__('atesim', self.port+2, self.context)
def run(self):
print 'Running Basic Machine learning-based testing,' + \
'listening on port %d ...' % self.port
self.socket.connect("tcp://127.0.0.1:%d" % self.port)
try:
while True:
print '[ Train: %s %7d %s ] ' \
% (self.c.GREEN, self.num_train_chips, self.c.ENDC),
print '[ Predicted: %s %7d %s ]' \
% (self.c.GREEN, self.num_predicted, self.c.ENDC)
self.get_chip()
self.update_model()
self.update_view(self.stats)
except KeyboardInterrupt:
print '\nterminating basic ML testing.'
def get_chip(self):
"""Request and retrieve a chip from the ATE simulator.
"""
self.socket.send('REQ:LCT')
chip_lct = json.loads(self.socket.recv())
self.log('REQ:LCT')
self.socket.send('REQ:gnd')
chip_gnd = json.loads(self.socket.recv())
self.log('REQ:gnd')
self.socket.send('REQ:done')
assert self.socket.recv() == 'RES:ack', \
'Error: invalid ack from ATE simulator'
self.log('REQ:done')
self.current_chip = Chip(LCT=chip_lct, gnd=chip_gnd)
self.num_chips_tested += 1
def update_model(self):
"""Add the latest chip to the chip buffer. If we've already collected
1,000 chips, then we go ahead and train the model.
"""
if self.num_chips_tested <= 1000:
self.chips.append(self.current_chip)
if self.num_chips_tested == 1000:
print "Training KNN on 1000 chips"
self.is_trained = True
self.knn.fit(self.chips)
if self.num_chips_tested >= 1000:
time.sleep(0.1)
self.num_predicted += 1
chip_gnd_predicted = self.knn.predict(self.current_chip)
self.log('Actual pf: %d' % self.current_chip.gnd)
self.log('Pred. pf: %d' % chip_gnd_predicted)
# Accounting TE / YL here
if self.current_chip.gnd == -1 and chip_gnd_predicted == 1:
self.test_escapes += 1
if self.current_chip.gnd == 1 and chip_gnd_predicted == -1:
self.yield_loss += 1
def log(self, msg):
# might expand this later
print '\t-> %s %s %s' % ( self.c.RED, msg, self.c.ENDC )
@property
def num_train_chips(self):
return len(self.chips)
@property
def stats(self):
"""The self.update_view() method from the ViewServerMixin class
expects a Python object that can be JSONified. This function returns
such an object.
"""
return {
'name' : 'basic',
'datetime' : datetime.datetime.utcnow().isoformat(),
'parms' :
{
'num_train_chips' :
{
'desc' : 'Number of chips in training set',
'value': str(self.num_train_chips)
},
'num_predicted' :
{
'desc' : 'Number of chips predicted with k-NN',
'value': str(self.num_predicted)
},
'test_escape' :
{
'desc' : 'Test escape rate',
#'value': '%1.3f%%' % (self.test_escapes * 100.0 / self.num_chips_tested)
'value': (self.test_escapes * 100.0 / self.num_chips_tested)
},
'yield_loss' :
{
'desc' : 'Yield loss rate',
#'value': '%1.3f%%' % (self.yield_loss * 100.0 / self.num_chips_tested)
'value': (self.yield_loss * 100.0 / self.num_chips_tested)
},
'is_trained' :
{
'desc' : 'Model trained',
'value': (lambda (x): 'Yes' if x else 'No')(self.is_trained)
}
}
}