The combiner has sample x=1, y=0, gradient w. The worker has x=1 and x=2,
hence numerator w*1*1 + w*2*2. Its mass is carried as an integer, not a mean.
Divide by the total sample mass, then apply w - (1/4)*g. No gradient or
learning-rate byte pattern is pasted into the program.
137def peerHostContribution =
138 (lambda unrestricted worker : Nat .
139 (lambda unrestricted weight : Nat .
140 (lambda unrestricted numerator : Nat .
141 (lambda unrestricted work : Nat .
142 (lambda unrestricted tail : (family NativePhysicalCommands) .
143 (nat-eliminate
144 (lambda unrestricted role : Nat . (family NativePhysicalCommands))
145 (peerStore numerator (frameNativeLoad weight) tail)
146 (lambda unrestricted previous : Nat .
147 (lambda unrestricted ignored : (family NativePhysicalCommands) .
148 (peerNatural
149 work
150 2
151 (peerBinary
152 peerMultiply
153 work
154 work
155 work
156 (peerBinary
157 peerMultiply
158 numerator
159 weight
160 work
161 (peerBinary peerAdd numerator numerator weight tail))))))
162 worker))))))The compiler supplied declaration spans and resolved links from this source snapshot. This page does not assert that this file belongs to a checked closure.