CODEWITHSUNDEEP
haskellneural-networkfeed-forward
nat
nat

Reputation: 85

fmap with two functions

Im writing a neural network using haskell. Im basing my code on this http://www-cs-students.stanford.edu/~blynn/haskell/brain.html . I adapted the feedforward method in the following way:

feedForward :: [Float] -> [([Float], [[Float]])] -> [Float]
feedForward = foldl ((fmap tanh . ) . previousWeights)

Where previousWeights is:

previousWeights :: [Float] -> ([Float], [[Float]]) -> [Float]
previousWeights actual_value (bias, weights) = zipWith (+) bias (map (sum.(zipWith (*) actual_value)) weights)

I don't really understand what fmap tanh . From what I read fmap applied to two functions is like a composition. If i change the fmap for map I get the same result.

Upvotes: 2

Views: 139

Answers (1)

MCH
MCH

Reputation: 2214

It is much easier to read if we give the parameters names and remove the consecutive .:

feedForward :: [Float] -> [([Float], [[Float]])] -> [Float]
feedForward actual_value bias_and_weights =
  foldl
  (\accumulator -- the accumulator, it is initialized as actual_value
    bias_and_weight -> -- a single value from bias_and_weights
     map tanh $ previousWeights accumulator bias_and_weight)
  actual_value -- initialization value
  bias_and_weights -- list we are folding over

It might also help to know that type signature of foldl in this case will be ([Float] -> ([Float], [[Float]])-> [Float]) -> [Float] -> [([Float], [[Float]])] -> [Float].

Note: This style of code you have found, while fun to write, can be a challenge for others to read and I generally do not recommend you write this way if for other than fun.

Upvotes: 2

Related Questions