def processCollisionLayer(bumps : list):
# COLLISIONS_INPUT, PROXIMITY -> COLLISION -------------------------- LAYER 1
i = {i:bumps[i] for i in nns.active_ts}
layer = 1
w = nns.connectivities[layer]
# Proximity Layer Output
o = nns.outputs[layer - 1]
hf = nns.compositionFunction[layer]
g = nns.activationFunction[layer]
f = nns.outputFunction[layer]
h_collision = annutils.sparseLayerInputComposition(i, o, w, hf)
a_collision = annutils.layerActivationLevel(h_collision, g)
nns.outputs[layer] = annutils.sparseArray(
list(nns.outputs[layer].keys()),
annutils.layerOutput(a_collision, f)
)
logger.info(f"Collision Layer Composed Input: {h_collision}")
logger.info(f"Weights Proximity to Collision:")
for n, conn in w.items(): logger.info(f"{n}->{conn}")
logger.info(f"Collision Layer Output: {nns.outputs[layer]}")
def processProximityLayer(distances: list):
# DISTANCES_INPUT -> PROXIMITY ---------------------------- LAYER 0
i = [distances[i] for i in nns.active_ps]
layer = 0
hf = nns.compositionFunction[layer]
g = nns.activationFunction[layer]
f = nns.outputFunction[layer]
# summed activations of each neuron in the Proximity Layer 0
h_proximity = annutils.layerInputComposition(i=i, o=[], w=[], h=hf)
# activation level of each neuron in the Proximity Layer 0
a_proximity = annutils.layerActivationLevel(h_proximity, g)
# output level of each neuron that will be passed to the next layer
nns.outputs[layer] = annutils.sparseArray(
list(nns.outputs[layer].keys()), annutils.layerOutput(a_proximity, f))
logger.info(f"Proximity Layer Output {nns.outputs[layer]}")
def processMotorLayer():
# COLLISION -> MOTORS ----------------------------- LAYER 3
i = []
layer = 3
w = nns.connectivities[layer]
o = nns.outputs[layer - 2] # Collision Layer Output
hf = nns.compositionFunction[layer]
g = nns.activationFunction[layer]
f = nns.outputFunction[layer]
h_motor = annutils.sparseLayerInputComposition(i, o, w, hf)
a_motor = annutils.layerActivationLevel(h_motor, g)
nns.outputs[layer] = annutils.sparseArray(list(nns.outputs[layer].keys()),
annutils.layerOutput(a_motor, f))
logger.info(f"Motor Layer Composed Input: {h_motor}")
logger.info(f"Motor Layer Output: {nns.outputs[layer]}")
def processReverseLayer():
# COLLISION -> REVERSE ----------------------------- LAYER 2
i = []
layer = 2
w = nns.connectivities[layer]
# Collision Layer Output
o = nns.outputs[layer - 1]
hf = nns.compositionFunction[layer]
g = nns.activationFunction[layer]
f = nns.outputFunction[layer]
h_reverse = annutils.sparseLayerInputComposition(i, o, w, hf)
a_reverse = annutils.layerActivationLevel(h_reverse, g)
nns.outputs[layer] = annutils.sparseArray(
list(nns.outputs[layer].keys()), annutils.layerOutput(a_reverse, f))
logger.info(f"Reverse Layer Composed Input: {h_reverse}")
logger.info(f"Reverse Layer Output: {nns.outputs[layer]}")
def processCollisionLayer(bumps):
# COLLISIONS_INPUT, PROXIMITY -> COLLISION -------------------------- LAYER 1
layer = 1
w = nns.connectivities[layer]
o = nns.outputs[layer - 1] # Proximity Layer Output
hf = nns.compositionFunction[layer]
g = nns.activationFunction[layer]
f = nns.outputFunction[layer]
#h_collision = [annutils.inputComposition(bumps[n], o, w[n], hf) for n in range(0, len(bumps))]
bumperOutputByConnection = annutils.mapToSensorsOutput(bumps, nns.bumpersConnections)
h_collision = [annutils.inputComposition(bumperOutputByConnection[n], o, w[n], hf) for n in range(0, nns.nCollisionNodes)]
a_collision = annutils.layerActivationLevel(h_collision, g)
nns.outputs[layer] = annutils.sparseArray(
list(nns.outputs[layer].keys()),
annutils.layerOutput(a_collision, f)
)
logger.info(f"Collision Layer Composed Input: {h_collision}")
logger.info(f"Weights Proximity to Collision:")
for n, conn in w.items(): logger.info(f"{n}->{conn}")
logger.info(f"Collision Layer Output: {nns.outputs[layer]}")
# for each neuron of layer[j] the matrix holds the weights to each neuron of level[i = j - 1]
# in the form of neuron[n] of layer[j] -> [ w[n][0], ..., w[n][m] ] of neuron[0...m] of layer[i]
connectivities = {
1:
_proximityToCollisionConnections, # Collision <- Proximity ==> FULLY CONNECTED
2:
_collisionToReverseConnections, # Reverse Command <- Collision, not fully connected
3:
_collisionToMotorConnections # Motor Command <- Collision, not fully connected
}
# layer -> output
# results of f(activation[i]) where f is the output function
outputs = {
0:
annutils.sparseArray(
active_ps, [0.0 for _ in active_ps]), # output: Proximity -> Collision
1:
annutils.sparseArray(
active_ts,
[0.0 for _ in active_ts]), # output: Collison -> Motor, Reverse
2:
annutils.sparseArray(
list(connectivities[2].keys()),
[0.0 for _ in connectivities[2].keys()]), # output: Reverse -> ...
3:
annutils.sparseArray(
list(connectivities[3].keys()),
[0.0 for _ in connectivities[3].keys()]) # output: Motor -> ...
}
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
