def do_sum(self):
if self.input is not None and self.weight is not None:
self.set_property(
"value",
self.get_property("input") * self.get_property("weight"))
self.emit("fire")
if globals.runInDebug:
print colored ("Synapse from Neuron %d at Layer %d fires to Neuron %d at Layer %d." % \
(self.source.Id, self.source.LayerId, self.target.Id, self.target.LayerId), "green")
def do_sum (self):
if self.input is not None and self.weight is not None:
self.set_property (
"value",
self.get_property ("input") * self.get_property ("weight")
)
self.emit ("fire")
if globals.runInDebug:
print colored ("Synapse from Neuron %d at Layer %d fires to Neuron %d at Layer %d." % \
(self.source.Id, self.source.LayerId, self.target.Id, self.target.LayerId), "green")
def _simple_dump_init (Object, Name=None, FullExposure=False):
"""
"""
header = ""
what_type = `type (Object)`[8:]
what_type = what_type[:len(what_type)-2]
introspect= dir (Object)
properties = dict ()
inherited_properties = dict ()
methods = dict ()
for m in introspect:
if not FullExposure:
if m.startswith ("__") and m.endswith ("__"):
continue
target = getattr (Object, m)
if callable (target):
methods [m] = str (target).__doc__
else:
t = `type (target)`[7:].strip("'>< ")
#t = t[:len(t)-2]
properties[m] = {
"type" : t,
"value": target
}
# Define the header
if Name is not None:
header = "Dumping instance %s of class %s" % (colored(Name, fg="bold"), colored(what_type, fg="underline"))
else:
header = "Dumping instance of class %s" % (colored(what_type, fg="underline"))
print header
data = {
"method": methods,
"properties": properties
}
return data
def do_sum (self):
if self.input is not None and self.weight is not None:
self.set_property (
"value",
self.get_property ("input") * self.get_property ("weight")
)
if globals.runInDebug:
source = self.lockOwner
target = None
if source is self.startPoint:
target = self.endPoint
else:
target = self.startPoint
print colored ("BidiSynapse sums and fires\n\tFROM %s\n\tTO %s", fg="green") % (source, target)
self.emit ("bifire")
def do_sum(self):
if self.input is not None and self.weight is not None:
self.set_property(
"value",
self.get_property("input") * self.get_property("weight"))
if globals.runInDebug:
source = self.lockOwner
target = None
if source is self.startPoint:
target = self.endPoint
else:
target = self.startPoint
print colored(
"BidiSynapse sums and fires\n\tFROM %s\n\tTO %s",
fg="green") % (source, target)
self.emit("bifire")
def _simple_dump_init(Object, Name=None, FullExposure=False):
"""
"""
header = ""
what_type = ` type(Object) ` [8:]
what_type = what_type[:len(what_type) - 2]
introspect = dir(Object)
properties = dict()
inherited_properties = dict()
methods = dict()
for m in introspect:
if not FullExposure:
if m.startswith("__") and m.endswith("__"):
continue
target = getattr(Object, m)
if callable(target):
methods[m] = str(target).__doc__
else:
t = ` type(target) ` [7:].strip("'>< ")
#t = t[:len(t)-2]
properties[m] = {"type": t, "value": target}
# Define the header
if Name is not None:
header = "Dumping instance %s of class %s" % (colored(
Name, fg="bold"), colored(what_type, fg="underline"))
else:
header = "Dumping instance of class %s" % (colored(what_type,
fg="underline"))
print header
data = {"method": methods, "properties": properties}
return data
nn.createNet ()
inputs = [ 1.0 ]
weights= [-0.80, 1.0, 0.25, \
0.13, -0.47, 0.64, 0.99, 0.20, \
-1.0, 0.72 ]
print "Neural Network Settings:"
print " Input: 1.0"
print " Weights: -0.80, 1.0, 0.25, 0.13, -0.47, 0.64, 0.99, 0.20, -1.0, 0.72\n"
raw_input("\nPress any key for the first run...")
nn.update (inputs, weights)
print "Neural network output:", colored ("%r", fg="yellow") % nn.output ()[0]
print
print
print "We keep the the starting weights, but we input the value: 0.5"
raw_input("\nPress any key for the second run...")
nn.update ( inputs=[0.5] )
print "Neural network output:", colored ("%r", fg="yellow") % nn.output ()[0]
print
print
print "We keep the the starting weights, but we input the value: -0.25"
raw_input("\nPress any key for the second run...")
nn.update ( inputs=[-0.25] )
print "Neural network output:", colored ("%r", fg="yellow") % nn.output ()[0]
print
second_hidden = NeuronLayer(Name="Hidden Layer #2",
Neurons=2,
AddBias=True)
second_hidden.createLayer()
output_layer = NeuronLayer(Name="Output Layer", Neurons=1, AddBias=True)
output_layer.createLayer()
print "*" * 100
input_layer.connectTo(first_hidden)
print "*" * 50
first_hidden.connectTo(second_hidden)
print "*" * 50
second_hidden.connectTo(output_layer)
print "*" * 50
input_layer.putInputs((1.0, 2.0, 3.0))
input_layer.putWeights((0.3, 0.4, -0.11, -0.8, -0.25, 0.25))
first_hidden.putWeights((1.0, 0.8, 0.7, 0.1))
first_hidden.putBiasWeights((0.5, -0.5))
first_hidden.biasLayer.putBiasInputs()
second_hidden.putWeights((0.1, 0.2))
second_hidden.putBiasWeights((0.5, -0.5))
second_hidden.biasLayer.putBiasInputs()
output_layer.putBiasWeights((0.5, ))
output_layer.biasLayer.putBiasInputs()
print colored("Output: %r", fg="yellow") % output_layer[0].getValue()
def dump_properties(Object,
watchProps=None,
Name=None,
FullExposure=False,
showType=False,
showVal=True,
showTitle=False):
"""
"""
data = _simple_dump_init(Object, Name, FullExposure)
if watchProps is not None:
title = "Watching Properties"
props = dict()
for p in watchProps:
if p in data['properties']:
props[p] = data['properties'][p]
else:
title = "Properties"
props = data['properties']
total = len(props)
if total > 0:
if showTitle: print "%s (%d):\n│" % (title, total)
# Make the output pretty :-)
spacing_name = 0
spacing_type = 0
for p in props.iterkeys():
prop_name_len = 0
if FullExposure:
prop_name_len = len(p)
else:
if not p.startswith("__") and not p.endswith("__"):
prop_name_len = len(p)
if prop_name_len > spacing_name: spacing_name = prop_name_len
prop_type_len = len(props[p]['type'])
if prop_type_len > spacing_type: spacing_type = prop_type_len
# Print the properties order by name.
counter = 0
bgcolor = None
s = sorted(props.iterkeys())
for p in s:
prop_name = p.ljust(spacing_name)
prop_type = props[p]['type'].ljust(spacing_type)
prop_val = props[p]['value']
if counter < total - 1:
symbol = "├─"
else:
symbol = "└─"
"""
if p == "__gdoc__" or p == "__doc__":
print "`--", colored(prop_name, fg="bold"), ":", prop_type, ":", prop_val
continue
"""
print symbol, colored(prop_name, fg="bold"),
if showType: print ":", prop_type,
if showVal:
if type (prop_val) is types.DictionaryType or \
type (prop_val) is types.ListType:
print ":",
pprint.pprint(prop_val)
else:
print ":", colored_value(prop_val)
if not showType and not showVal: print
counter += 1
print
This is the same as the "example1.py" in the "neurons" folder.
"""
# batteries
import sys
sys.path.append("../../src/")
# spiral framework
from spiral.helper import colored
from spiral.nn import globals
from spiral.nn.impl import Feedforward
if __name__ == "__main__":
globals.runInDebug = True
settings = {"Inputs": 3, "Outputs": 1, "Layers": 2, "NeuronsPerLayer": 2}
nn = Feedforward(**settings)
nn.createNet()
nn.createDefaultSynapses()
inputs = [1.0, 2.0, 3.0]
weights = [0.3, 0.4, -0.1, -0.8, -0.25, 0.25, 1.0, 0.8, 0.7, 0.1, 0.1, 0.2]
nn.update(inputs, weights)
print colored("Neural Network output: %r", fg="yellow") % nn.output(0)
nn = MyNetwork(**settings)
nn.createNet()
inputs = [1.0]
weights= [-0.80, 1.0, 0.25, \
0.13, -0.47, 0.64, 0.99, 0.20, \
-1.0, 0.72 ]
print "Neural Network Settings:"
print " Input: 1.0"
print " Weights: -0.80, 1.0, 0.25, 0.13, -0.47, 0.64, 0.99, 0.20, -1.0, 0.72\n"
raw_input("\nPress any key for the first run...")
nn.update(inputs, weights)
print "Neural network output:", colored("%r", fg="yellow") % nn.output()[0]
print
print
print "We keep the the starting weights, but we input the value: 0.5"
raw_input("\nPress any key for the second run...")
nn.update(inputs=[0.5])
print "Neural network output:", colored("%r", fg="yellow") % nn.output()[0]
print
print
print "We keep the the starting weights, but we input the value: -0.25"
raw_input("\nPress any key for the second run...")
nn.update(inputs=[-0.25])
print "Neural network output:", colored("%r", fg="yellow") % nn.output()[0]
print
def dump_properties (Object, watchProps=None, Name=None, FullExposure=False, showType=False, showVal=True, showTitle=False):
"""
"""
data = _simple_dump_init (Object, Name, FullExposure)
if watchProps is not None:
title = "Watching Properties"
props = dict ()
for p in watchProps:
if p in data['properties']:
props [p] = data['properties'][p]
else:
title = "Properties"
props = data['properties']
total = len (props)
if total > 0:
if showTitle: print "%s (%d):\n│" % (title, total)
# Make the output pretty :-)
spacing_name = 0
spacing_type = 0
for p in props.iterkeys():
prop_name_len = 0
if FullExposure:
prop_name_len = len (p)
else:
if not p.startswith ("__") and not p.endswith ("__"):
prop_name_len = len (p)
if prop_name_len > spacing_name: spacing_name = prop_name_len
prop_type_len = len (props[p]['type'])
if prop_type_len > spacing_type: spacing_type = prop_type_len
# Print the properties order by name.
counter = 0
bgcolor = None
s = sorted(props.iterkeys())
for p in s:
prop_name = p.ljust (spacing_name)
prop_type = props[p]['type'].ljust (spacing_type)
prop_val = props[p]['value']
if counter < total-1:
symbol = "├─"
else:
symbol = "└─"
"""
if p == "__gdoc__" or p == "__doc__":
print "`--", colored(prop_name, fg="bold"), ":", prop_type, ":", prop_val
continue
"""
print symbol, colored(prop_name, fg="bold"),
if showType: print ":", prop_type,
if showVal:
if type (prop_val) is types.DictionaryType or \
type (prop_val) is types.ListType:
print ":",
pprint.pprint (prop_val)
else:
print ":", colored_value (prop_val)
if not showType and not showVal: print
counter += 1
print
if __name__ == "__main__":
globals.runInDebug = True
n1 = Neuron (Id=1, LayerId=0, Name="Input")
n1.setAutoNotify (True)
n2 = Neuron (Id=2, LayerId=1)
n3 = Neuron (Id=3, LayerId=1)
n4 = Neuron (Id=4, LayerId=2, Name="Output")
# Create the synapses
n1.createSynapse (1, n2)
n1.createSynapse (1, n3)
n2.createSynapse (2, n4)
n3.createSynapse (2, n4)
print "*" * 40
# Put the weights
n1.putWeights ((0.5, 0.25))
n2.putWeights ((-.75,))
n3.putWeights ((1.0,))
# Insert the inputs
n1.putInput (1.0)
print colored ("Output value: %r", fg="yellow") % n4.getValue ()
"NeuronsPerLayer": 2,
"AddBias": True,
"LearningRate": 0.9
}
bp = Backpropagation (**settings)
bp.createNet ()
bp.createDefaultSynapses ()
random_weights = list ()
for i in range (bp.NumberOfWeights):
random_weights.append (SimpleMath.randomClamped())
bp.putWeights (random_weights)
print colored ("Training network for 1000 iterations ...", fg="yellow")
print colored ("*" * 80, fg="yellow")
for i in range (1):
bp.train ( [0, 0], 0)
bp.train ( [0, 1], 1)
bp.train ( [1, 0], 1)
bp.train ( [1, 1], 0)
print
print colored ("Testing ...", fg="yellow")
print colored ("*" * 80, fg="yellow")
print bp.run ([0, 0])
print bp.run ([0, 1])
print bp.run ([1, 0])
print bp.run ([1, 1])
import sys
sys.path.append("../../src/")
# spiral framework
from spiral.helper import colored
from spiral.nn import globals
from spiral.nn.impl import Feedforward
if __name__ == "__main__":
globals.runInDebug = False
settings = {
"Inputs": 3,
"Outputs": 1,
"Layers": 2,
"NeuronsPerLayer": 2,
"AddBias": True
}
nn = Feedforward(**settings)
nn.createNet()
nn.createDefaultSynapses()
inputs = [1.0, 2.0, 3.0]
weights= [ 0.3, 0.4, -0.1, -0.8, -0.25, 0.25, \
1.0, 0.8, 0.7, 0.1, \
0.1, 0.2, 1.0, 0.3, 0.24, 0.324, 0.12 ]
nn.update(inputs, weights)
print "Neural Network epoch completed in: %r seconds" % nn.timeUpdate()
print colored("Neural Network output: %r", fg="yellow") % nn.output(0)
n2.createSynapse(1, n4)
n2.createSynapse(1, n5)
n3.createSynapse(1, n4)
n3.createSynapse(1, n5)
# Create the synapses for the first hidden layer
n4.createSynapse(2, n6)
n4.createSynapse(2, n7)
n5.createSynapse(2, n6)
n5.createSynapse(2, n7)
# Create the synapses for the second hidden layer
n6.createSynapse(3, n8)
n7.createSynapse(3, n8)
# Put the weights
n1.putWeights((0.3, 0.4))
n2.putWeights((-0.1, -0.8))
n3.putWeights((-0.25, 0.25))
n4.putWeights((1.0, 0.8))
n5.putWeights((0.7, 0.1))
n6.putWeights((0.1, ))
n7.putWeights((0.2, ))
# Insert the inputs
n1.putInput(1.0)
n2.putInput(2.0)
n3.putInput(3.0)
print colored("Output value: %r", fg="yellow") % n8.getValue()
first_hidden.createLayer ()
second_hidden= NeuronLayer (Name="Hidden Layer #2", Neurons=2, AddBias=True)
second_hidden.createLayer ()
output_layer = NeuronLayer (Name="Output Layer" , Neurons=1, AddBias=True)
output_layer.createLayer ()
print "*" * 100
input_layer.connectTo (first_hidden)
print "*" * 50
first_hidden.connectTo (second_hidden)
print "*" * 50
second_hidden.connectTo (output_layer)
print "*" * 50
input_layer.putInputs ( (1.0, 2.0, 3.0) )
input_layer.putWeights ( (0.3, 0.4, -0.11, -0.8, -0.25, 0.25) )
first_hidden.putWeights ( (1.0, 0.8, 0.7, 0.1) )
first_hidden.putBiasWeights ( (0.5, -0.5) )
first_hidden.biasLayer.putBiasInputs ()
second_hidden.putWeights( (0.1, 0.2) )
second_hidden.putBiasWeights ( (0.5, -0.5) )
second_hidden.biasLayer.putBiasInputs ()
output_layer.putBiasWeights ( (0.5,) )
output_layer.biasLayer.putBiasInputs ()
print colored("Output: %r", fg="yellow") % output_layer [0].getValue ()