ann.core
-main
(-main & args)
Artificial Neural Networks with stochastic gradient descent optimization
adjust-weights
(adjust-weights x w y lr)
feeds data into nn and returns adjusted weights
crab2
vector version, reordered dataset
error-check
(error-check input weight threshold)
checks error given `inputs` `weights` `threshold`
error-loop
(error-loop mn mx step data weight)
step between min and max and error-check to examine outliers.
expand
(expand dataset magnitude)
expands the dataset for testing
feed
(feed input weight learnrate)
loops across input and adjustes the weights for all of it.
`input` assumes y values are at the end of the vectors
feed-one
(feed-one x w)
feeds a single `x` into the ANN
find-error
(find-error value theta match)
find the error given a `value` and `theta` and if is what we expect to `match`
gen-matrix
(gen-matrix r c & m)
generates a `r` by `c` matrix with random weights between -1 and 1.
l2v
(l2v matrix)
convert list to vector matrix
mmap
(mmap function matrix)
maps a function on a weight vector matrix
multitiered-forward
(multitiered-forward input w)
takes in weights `w and inputs `x and propagates the inputs though the network
nifty-feeder
(nifty-feeder data magnitude lrs size & {:keys [verbose-flag], :or {verbose-flag false}})
expands and feeds a dataset, useful for finding that special rate
norm-scale
(norm-scale lst)
scales values in matrix to a range between -1 and 1, utilizing max-fld
pluck
(pluck fn matrix)
extract a value from nexted matrix
refeed
(refeed data weight lrs)
refeeds results from training at different learning rates `lrs`
scrub
(scrub data)
scrubs the first and last attribute, species and gender respectivly
sigmoid
(sigmoid z)
takes in `z` and throws it in the sigmoid function
w
adjusted weights for crabv with nifty-feeder
weight-gen
(weight-gen lst)
generates a multitiered matrix