rowan.a.tl/propeller
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge branch 'master' into boolean-multipliers

Browse Source
This commit is contained in:
Lee Spector 2023-08-30 14:47:58 -04:00
commit 1733ffd42e
3 changed files with 55 additions and 39 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
README.md
View File

@ -83,6 +83,12 @@ or
lein run -m propeller.problems.simple-regression :variation "{:umad 1.0}"
```
By default, Propeller will conduct many processes concurrently on multiple
cores using threads. If you want to disable this behavior (for example, during
debugging) then provide the argument `:single-thread-mode` with the value `true`.
Threads are not available in Javascript, so no processes are run concurrnetly
when Propeller is run in Clojurescript.
## CLJS Usage

62
src/propeller/gp.cljc
View File

@ -13,7 +13,8 @@
[propeller.push.instructions.polymorphic]
[propeller.push.instructions.string]
[propeller.push.instructions.vector]
[propeller.selection :as selection]))
[propeller.selection :as selection]
[propeller.utils :as utils]))
(defn report
"Reports information for each generation."
@ -32,44 +33,25 @@
(println)))
(defn gp
"Main GP loop.
On each iteration, it creates a population of random plushies using a mapper
function and genome/make-random-plushy function,
then it sorts the population by the total error using the error-function
and sort-by function. It then takes the best individual from the sorted population,
and if the parent selection is set to epsilon-lexicase, it adds the epsilons to the argmap.
The function then checks if the custom-report argument is set,
if so it calls that function passing the evaluated population,
current generation and argmap. If not, it calls the report function
passing the evaluated population, current generation and argmap.
Then, it checks if the total error of the best individual is less than or equal
to the solution-error-threshold or if the current generation is greater than or
equal to the max-generations specified. If either is true, the function
exits with the best individual or nil. If not, it creates new individuals
for the next generation using the variation/new-individual function and the
repeatedly function, and then continues to the next iteration of the loop. "
"Main GP loop."
[{:keys [population-size max-generations error-function instructions
max-initial-plushy-size solution-error-threshold mapper]
:or {solution-error-threshold 0.0
;; The `mapper` will perform a `map`-like operation to apply a function to every individual
;; in the population. The default is `map` but other options include `mapv`, or `pmap`.
mapper #?(:clj pmap :cljs map)}
max-initial-plushy-size solution-error-threshold]
:or {solution-error-threshold 0.0}
:as argmap}]
;;
(prn {:starting-args (update (update argmap :error-function str) :instructions str)})
(println)
;;
(loop [generation 0
population (mapper
population (utils/pmapallv
(fn [_] {:plushy (genome/make-random-plushy instructions max-initial-plushy-size)})
(range population-size))] ;creates population of random plushys
(range population-size)
argmap)] ;creates population of random plushys
(let [evaluated-pop (sort-by :total-error
(mapper
(utils/pmapallv
(partial error-function argmap (:training-data argmap))
population)) ;population sorted by :total-error
population ;population sorted by :total-error
argmap))
best-individual (first evaluated-pop)
argmap (if (= (:parent-selection argmap) :epsilon-lexicase)
(assoc argmap :epsilons (selection/epsilon-list evaluated-pop))
@ -84,17 +66,25 @@ repeatedly function, and then continues to the next iteration of the loop. "
(prn {:total-test-error
(:total-error (error-function argmap (:testing-data argmap) best-individual))})
(when (:simplification? argmap)
(let [simplified-plushy (simplification/auto-simplify-plushy (:plushy best-individual) error-function argmap)]
(prn {:total-test-error-simplified (:total-error (error-function argmap (:testing-data argmap) (hash-map :plushy simplified-plushy)))}))))
(let [simplified-plushy (simplification/auto-simplify-plushy
(:plushy best-individual)
error-function argmap)]
(prn {:total-test-error-simplified
(:total-error (error-function argmap
(:testing-data argmap)
(hash-map :plushy simplified-plushy)))})))
#?(:clj (shutdown-agents)))
;;
(>= generation max-generations)
nil
#?(:clj (shutdown-agents))
;;
:else (recur (inc generation)
(if (:elitism argmap)
(conj (repeatedly (dec population-size)
#(variation/new-individual evaluated-pop argmap))
(conj (utils/pmapallv (fn [_] (variation/new-individual evaluated-pop argmap))
(range (dec population-size))
argmap)
(first evaluated-pop)) ;elitism maintains the most-fit individual
(repeatedly population-size
#(variation/new-individual evaluated-pop argmap))))))))
(utils/pmapallv (fn [_] (variation/new-individual evaluated-pop argmap))
(range population-size)
argmap)))))))

22
src/propeller/utils.cljc
View File

@ -1,6 +1,7 @@
(ns propeller.utils
"Useful functions."
(:require [clojure.zip :as zip]))
(:require [clojure.zip :as zip]
[clojure.repl :as repl]))
(defn first-non-nil
"Returns the first non-nil values from the collection, or returns `nil` if
@ -85,3 +86,22 @@
zip/path
count
(max height))))))
(defn pmapallv
"A utility for concurrent execution of a function on items in a collection.
In single-thread-mode this acts like mapv. Otherwise it acts like pmap but:
1) coll should be finite, 2) the returned sequence will not be lazy, and will
in fact be a vector, 3) calls to f may occur in any order, to maximize
multicore processor utilization, and 4) takes only one coll so far."
[f coll args]
#?(:clj (vec (if (:single-thread-mode args)
(doall (map f coll))
(let [agents (map #(agent % :error-handler
(fn [agnt except]
(repl/pst except 1000)
(System/exit 0)))
coll)]
(dorun (map #(send % f) agents))
(apply await agents)
(doall (map deref agents)))))
:cljs (mapv f coll)))