More markov cleanup

main
Eric Ihli 3 years ago
parent ae9d5bae9b
commit 1f7a013dca

@ -4,7 +4,6 @@
[clojure.java.io :as io]
[taoensso.nippy :as nippy]
[taoensso.timbre :as timbre]
[com.owoga.prhyme.frp :as frp]
[com.owoga.prhyme.util :as util]
[com.owoga.prhyme.core :as prhyme]
[com.owoga.prhyme.data.bigrams :as bigrams]
@ -20,16 +19,6 @@
[clojure.zip :as zip]
[clojure.walk :as walk]))
(defn weight-fn [word target result]
(let [rimes (frp/consecutive-matching word target :rimes)
nuclei (frp/consecutive-matching word target :nuclei)
onsets (frp/consecutive-matching word target :onsets)
total (apply + (map count [rimes nuclei onsets]))]
total))
(defn pred-fn [word target result]
(< 0 (weight-fn word target result)))
(defn weight-popular [word target result]
(if (dict/popular (:normalized-word word))
10

@ -6,19 +6,14 @@
[com.owoga.prhyme.util :as util]
[com.owoga.prhyme.core :as prhyme]
[com.owoga.prhyme.util.nlp :as nlp]
[com.owoga.prhyme.gen :as gen]
[com.owoga.prhyme.generation.weighted-selection :as weighted-selection]
[taoensso.tufte :as tufte :refer [defnp p profiled profile]]
[com.owoga.prhyme.frp :as frp]
[clojure.java.io :as io]))
(tufte/add-basic-println-handler! {})
(def ^:dynamic *base-url* "https://www.hplovecraft.com/writings/texts/")
(def words-map
(into {} (map #(vector (string/lower-case (:word %)) %) frp/words)))
(defn fetch-url [url]
(html/html-resource (java.net.URL. url)))
@ -63,277 +58,3 @@
(#(str % "\n"))
(append-to-file "lovecraft.txt")))
(take 10 (links))))
(defn tokens-from-file [file]
(with-open [r (io/reader file)]
(tokens (slurp r))))
(defn window [n]
(fn [coll]
(cond
(empty? coll) []
(< (count coll) n) []
:else (cons (take n coll)
(lazy-seq ((window n) (drop n coll)))))))
(defn markov [tokens]
(->> tokens
(map
(fn [token]
(let [k (butlast token)
v (last token)]
[k v])))
(reduce
(fn [a [k v]]
(update-in a [k v] (fnil inc 0)))
{})))
(defn running-total
([coll]
(running-total coll 0))
([coll last-val]
(cond
(empty? coll) nil
:else (cons (+ last-val (first coll))
(lazy-seq
(running-total
(rest coll)
(+ last-val (first coll))))))))
(defn weighted-rand [weights]
(let [running-weights (running-total weights)
rand-val (rand (last running-weights))]
(loop [i 0]
(if (> (nth running-weights i) rand-val)
i
(recur (inc i))))))
(def word-set (into #{} (->> prhyme/words
(map first)
(map string/lower-case)
(map #(string/replace % #"\(\d+\)" "")))))
(defn normalize-tokens [tokens]
(->> tokens
(map string/lower-case)
(filter word-set)))
(defn main []
(->> (tokens-from-file "lovecraft.txt")
(reverse)
(normalize-tokens)
((window 2))
(markov)
(into {})))
(defn synonym?
"Given a possibility, like [\"foo\" 3]
which says that foo follows a particular key with
a weight of 3, a word is a synonym of that possibility
if the word is a synonym ."
[p synonyms]
(synonyms p))
(defn adjust-for-synonyms
"If a word is in a set of synonyms, adjust its weight upwards."
[synonyms]
(fn [possibilities]
(reduce
(fn [p s]
(if (s p)
(update p s #(* 5 %))
p))
possibilities
synonyms)))
(defn adjust-for-rimes
[target-rime dictionary]
(fn [possibilities]
(into
{}
(map
(fn [[p v]]
(let [possibility (get dictionary p)
factor (count
(frp/consecutive-matching
target-rime
possibility
:rimes))]
[p (* v (max 1 (* factor 4)))]))
possibilities))))
(defonce lovecraft-markov (read-string (slurp "lovecraft.edn")))
(defn markov-key [key-fn]
(fn [text]
(key-fn text)))
(defn gen-from [m p initial]
(loop [r (list initial)]
(cond
(p r) (recur (cons (m (list (first r))) r))
:else r)))
(defn rhyming-words
"List of rhyming words sorted by quality of rhyme."
[target]
(let [target-phrase (->> target
(prhyme/phrase->word frp/words)
(#(assoc % :rimes? true)))]
(->> target-phrase
(#(assoc % :rimes? true))
(frp/prhyme frp/words)
(sort-by
#(- (count
(frp/consecutive-matching
%
target-phrase
:rimes)))))))
(defn markov-rhymes [markov-data rhyming-words]
(->> (map
(fn [word]
(->> word
:word
string/lower-case
(#(string/replace % #"\(\d+\)" ""))
(#(vector % (get markov-data (list %))))))
rhyming-words)
(into #{})
(remove
(fn [[w p]]
(nil? p)))))
(defn adjust-for-over-syllables
"Adjust weights to prefer not going over the number
of syllables of the target word."
[target]
(fn [words]
(p :adjust-for-syllables
(map
(fn [word]
(if (or (nil? (:syllable-count word))
(nil? (:syllables target)))
(println word target))
(cond
(= (:syllable-count word) (count (:syllables target)))
(as-> word word
(assoc word :weight (* 3 (:weight word)))
(assoc word :adjusted-for-syllables-factor 3))
(< (:syllable-count word) (count (:syllables target)))
(as-> word word
(assoc word :weight (* 2 (:weight word)))
(assoc word :adjusted-for-syllables-factor 2))
:else
(as-> word word
(assoc word :weight (* 1 (:weight word)))
(assoc word :adjusted-for-syllables-factor 1))))
words))))
(defn adjust-for-membership-1
[set_ percent]
(let [ratio (- 1 percent)]
(fn [words]
(let [[members non-members]
((juxt filter remove)
#(set_ (:normalized-word %))
words)
weight-non-members (apply + (map :weight non-members))
target-weight-members (* ratio weight-non-members)
count-members (count members)
adjustment-members (/ target-weight-members count-members)]
(concat
(map
(fn [member]
(as-> member member
(assoc member :weight (* adjustment-members (:weight member)))
(assoc member :adjustment-for-membership adjustment-members)))
members)
non-members)))))
(defn adjust-for-membership [set_]
(fn [words]
(map
(fn [word]
(if (set_ (:normalized-word word))
(as-> word word
(assoc word :weight (* 2 (:weight word)))
(assoc word :adjust-for-membership-factor 2))
(assoc word :adjust-for-membership-factor 1)))
words)))
(defn filter-for-membership [set_]
(fn [words]
(map
(fn [word]
(if-not (set_ (:normalized-word word))
(as-> word word
(assoc word :weight (* 0.01 (:weight word)))
(assoc word :filter-for-membership-factor 0.01))
word))
words)))
(defn adjust-for-markov [markov-options]
(let [markov-set (into #{} (map first (keys markov-options)))]
(fn [words]
(let [result (map
(fn [word]
(if (markov-set (:normalized-word word))
(as-> word word
(assoc word :weight (* 100 (:weight word)))
(assoc word :adjust-for-markov-factor 100))
(assoc word :adjust-for-markov-factor 1)))
words)]
result))))
(comment
(let [markov-adjuster (adjust-for-markov (lovecraft-markov '("help")))]
(take 5 (markov-adjuster frp/words))))
(defn adjust-for-membership-1
[set_ percent]
(let [ratio (- 1 percent)]
(fn [words]
(let [[members non-members]
((juxt filter remove)
#(set_ (:normalized-word %))
words)
weight-non-members (apply + (map :weight non-members))
target-weight-members (* ratio weight-non-members)
count-members (count members)
adjustment-members (/ target-weight-members count-members)]
(concat
(map
(fn [member]
(as-> member member
(assoc member :weight (* adjustment-members (:weight member)))
(assoc member :adjustment-for-membership adjustment-members)))
members)
non-members)))))
(defn adjust-for-markov-1
[markov-options percent]
(let [ratio (- 1 percent)]
(fn [words]
(if (nil? markov-options)
words
(let [[markovs non-markovs]
((juxt filter remove)
#(markov-options (:normalized-word %))
words)
weight-non-markovs (apply + (map :weight non-markovs))
target-weight-markovs (* ratio weight-non-markovs)
count-markovs (count markovs)
adjustment-markovs (if (= 0 count-markovs) 1 (/ target-weight-markovs count-markovs))]
(concat
(map
(fn [markov]
(as-> markov markov
(assoc markov :weight (* adjustment-markovs (:weight markov)))
(assoc markov :adjustment-for-markov adjustment-markovs)))
markovs)
non-markovs))))))

@ -6,6 +6,7 @@
[com.owoga.trie :as trie]
[com.owoga.tightly-packed-trie :as tpt]
[com.owoga.tightly-packed-trie.encoding :as encoding]
[clojure.set :as set]
[clojure.string :as string]
[clojure.java.io :as io]
[com.owoga.phonetics :as phonetics]
@ -244,22 +245,43 @@
;; => [([("G" "AO1" "B") "bog"] [("G" "AO1" "F") "fog"])
;; ([("G" "AO1" "F") "fog"])
;; ([("G" "AA1" "B") "bog"] [("G" "AO1" "B") "bog"])]
(->> (rhyme-choices-walking-target-rhyme
rhyme-trie
["N" "AH1" "F"]
identity)
rand-nth
((fn [[phones words]]
[[phones] (rand-nth (vec words))])))
)
(defn rhyme-choices-walking-target-rhyme
"All target rhymes need to be in phone form.
`target-rhyme`: [N UH1 F]
If we try to turn string form into phone form,
we'd sometimes be forced to deal with multiple pronunciations.
By only handling phone form here, the caller can handle multiple pronunciations.
Makes for a cleaner API."
[trie target-rhyme]
(loop [target-rhyme target-rhyme
result []]
(let [choices (rhyme-choices trie target-rhyme)]
(if (or (empty? target-rhyme) (prhyme/last-primary-stress? (reverse target-rhyme)))
(into result choices)
(recur (butlast target-rhyme)
(into result choices))))))
Makes for a cleaner API.
`words-fn` gets passed the result of `rhyme-choices` which has this structures
([(G AO1 B) bog] [(G AO1 F) fog])
"
([trie target-rhyme]
(rhyme-choices-walking-target-rhyme
trie
target-rhyme
identity))
([trie target-rhyme words-fn]
(loop [target-rhyme target-rhyme
result []]
(let [choices (words-fn (rhyme-choices trie target-rhyme))]
(if (or (empty? target-rhyme)
(prhyme/last-primary-stress? (reverse target-rhyme)))
(into result choices)
(recur (butlast target-rhyme)
(into result choices)))))))
(comment
(let [words ["bloodclot" "woodrot" "moonshot" "dot" "bog" "pat" "pot" "lot"]
@ -416,38 +438,127 @@
)
(defn tightly-generate-n-syllable-sentence-rhyming-with
(defn tightly-generate-n-syllable-sentence
"It's difficult to mix a tight trie with rhymes. You need
to convert ids using the database."
[database
to convert ids using the database.
This is going to generate sentences backwards.
Generates the following structure:
[[[[S K AY1]] sky]
[[[DH AH0] [DH AH1] [DH IY0]] the]
[[[K R AE1 K S]] cracks]
[[[G R AW1 N D]] ground]
[[[DH AH0] [DH AH1] [DH IY0]] the]
[[[T UW1] [T IH0] [T AH0]] to]
[[[K IH1 NG D AH0 M]] kingdom]
[[[DH AH0] [DH AH1] [DH IY0]] the]
[[[D IH0 S T R OY1]] destroy]]
"
([database
markov-trie
n-gram-rank
target-sentence-syllable-count]
(tightly-generate-n-syllable-sentence
database
markov-trie
n-gram-rank
target-sentence-syllable-count
(constantly false)))
([database
markov-trie
n-gram-rank
target-sentence-syllable-count
markov-remove-fn]
(let [eos (database prhyme/EOS)
bos (database prhyme/BOS)]
(loop [phrase []]
(if (<= target-sentence-syllable-count
(prhyme/count-syllables-of-phrase
(string/join " " (map second phrase))))
phrase
(recur
(conj
phrase
(let [word (database
(get-next-markov
markov-trie
; Pad sentence with eos markers since we're working backwards
(into (vec (repeat (dec n-gram-rank) eos))
(mapv (comp database second) phrase))
; Remove eos, bos, and forbidden words
(fn [[lookup [word frequency]]]
(or (markov-remove-fn [lookup [word frequency]])
(#{eos bos} word)))))]
[(phonetics/get-phones word) word]))))))))
(comment
(tightly-generate-n-syllable-sentence
database
markov-trie
rhyme-trie
target-rhyme
n-gram-rank
target-rhyme-syllable-count
target-sentence-syllable-count]
(let [rhyme (->> (rhyme-choices-walking-target-rhyme rhyme-trie target-rhyme)
rand-nth
((fn [[phones words]]
[[phones] (rand-nth (vec words))])))]
(loop [phrase [rhyme]]
(if (or (= prhyme/BOS (second (peek phrase)))
(<= target-sentence-syllable-count
(prhyme/count-syllables-of-phrase
(string/join " " (map second phrase)))))
phrase
(recur
(conj
phrase
(let [word (database
(get-next-markov
markov-trie
(into (vec (repeat (dec n-gram-rank) (database prhyme/EOS)))
(mapv (comp database second) phrase))
(fn [[lookup [word frequency]]]
(= (database prhyme/EOS) word))))]
[(phonetics/get-phones word) word])))))))
3
10)
)
(defn tightly-generate-n-syllable-sentence-rhyming-with
"It's difficult to mix a tight trie with rhymes. You need
to convert ids using the database.
`rhyme-wordset-fn` will take something that looks like
([(G AO1 B) bog] [(G AO1 F) fog])
"
([database
markov-trie
rhyme-trie
target-rhyme
n-gram-rank
target-rhyme-syllable-count
target-sentence-syllable-count]
(tightly-generate-n-syllable-sentence-rhyming-with
database
markov-trie
rhyme-trie
target-rhyme
n-gram-rank
target-rhyme-syllable-count
target-sentence-syllable-count
(constantly false)
identity))
([database
markov-trie
rhyme-trie
target-rhyme
n-gram-rank
target-rhyme-syllable-count
target-sentence-syllable-count
markov-remove-fn
rhyme-wordset-fn]
(let [eos (database prhyme/EOS)
bos (database prhyme/BOS)
rhyme (->> (rhyme-choices-walking-target-rhyme
rhyme-trie
target-rhyme
rhyme-wordset-fn)
rand-nth)]
(loop [phrase [rhyme]]
(if (<= target-sentence-syllable-count
(prhyme/count-syllables-of-phrase
(string/join " " (map second phrase))))
phrase
(recur
(conj
phrase
(let [word (database
(get-next-markov
markov-trie
(into (vec (repeat (dec n-gram-rank) eos))
(mapv (comp database second) phrase))
(fn [[lookup [word frequency]]]
(or (markov-remove-fn [lookup [word frequency]])
(#{eos bos} word)))))]
[(phonetics/get-phones word) word]))))))))
;;;; Demo
;;;;
@ -462,10 +573,17 @@
target-rhyme
3
3
7)
7
(fn [[lookup [word freq]]]
(= (database "begun") word))
(fn [rhyming-words]
(->> (map (fn [[phones wordset]]
[phones (set/difference wordset #{"begun"})])
rhyming-words)
(remove (fn [[phones wordset]]
(empty? wordset))))))
(map second)
reverse))
(map (partial remove #{prhyme/BOS}))
(map (partial string/join " "))))
;; => ("funeral has just begun"
;; "dead illusion overdone"
@ -497,29 +615,108 @@
reverse))
(map (partial remove #{prhyme/BOS}))
(map data-transform/untokenize)))
)
(rhyme-choices-walking-target-rhyme
rhyme-trie
(->> (prhyme/phrase->all-flex-rhyme-tailing-consonants-phones
"filling")
first
first
reverse))
(defn sentence->phones
"Sentence is of the format
[[[[F L OW1]] flow]
[[[AH0 N D] [AE1 N D]] and]
[[[S IY1 K]] seek]
[[[F IH1 NG G ER0 Z]] fingers]
[[[Y AO1 R] [Y UH1 R]] your]
[[[TH R UW1]] through]
[[[S T R EH1 NG K TH] [S T R EH1 NG TH]]
strength]
[[[F AY1 N D]] find]
[[[K AE1 N] [K AH0 N]] can]]
Returns the concatenated list of phones so you can pluck some off and find
rhymes.
Note that each word in the sentence can have more than one pronunciation.
This function chooses one randomly.
"
[sentence]
(->> sentence
(map #(update % 0 rand-nth))
(apply map vector)
((fn [[phones words]]
[(string/join " " (reduce into [] phones)) (string/join " " words)]))
(first)))
(defn rhyme-from-scheme
"scheme of format [[A 9] [A 9] [B 5] [B 5] [A 9]]
Will include as many syllables as possible in finding rhymes
and will choose randomly with equal chance from all possible rhymes."
[scheme database markov-trie rhyme-trie]
(loop [scheme scheme
line-phones {}
result []]
(cond
(empty? scheme) result
:else
(let [[pattern syllable-count] (first scheme)
banned-words (into #{} (->> result
(map (comp last last))))
line (if (nil? (get line-phones pattern))
; Here, we need to make a choice about which pronunciation
; we want to use to build line-phones. Choose randomly.
(tightly-generate-n-syllable-sentence
database
markov-trie
3
syllable-count)
(tightly-generate-n-syllable-sentence-rhyming-with
database
markov-trie
rhyme-trie
(take 4 (get line-phones pattern))
3
3
syllable-count
(constantly false)
;; words-fn
;; ([("G" "AO1" "B") "bog"] [("G" "AO1" "F") "fog"])
(fn [rhyming-words]
(->> (map (fn [[phones wordset]]
[phones (set/difference
wordset
banned-words)])
rhyming-words)
(remove (fn [[phones wordset]]
(empty? wordset)))))))
rhyme (reverse (sentence->phones line))]
(recur (rest scheme)
(assoc line-phones pattern rhyme)
(conj result (reverse line)))))))
(comment
(tightly-generate-n-syllable-sentence
database
markov-trie
3
10)
(let [target-rhyme ["IY1" "ER"]]
(->> (repeatedly
10
#(->> (tightly-generate-n-syllable-sentence-rhyming-with
database
markov-trie
rhyme-trie
target-rhyme
3
3
7)
(map second)
reverse))
(map (partial remove #{prhyme/BOS}))
(map (partial string/join " "))))
(rhyme-from-scheme
'[[A 9] [A 9] [B 5] [B 5]]
database
markov-tight-trie
rhyme-trie)
(tightly-generate-n-syllable-sentence-rhyming-with
database
markov-trie
rhyme-trie
(first
(first
(prhyme/phrase->all-flex-rhyme-tailing-consonants-phones
"bother me")))
3
3
9
(constantly false)
(fn [[phones wordset]]
(set/difference wordset (into #{} []))))
)

@ -278,7 +278,7 @@
Returns all possible pronunciations. For hog -> haog, haag.
ROBOT -> '(OW1 AA2 T)"
ROBOT => '([(OW1 AA T) robot] [(OW1 AH T) robot])"
[phrase]
(->> phrase
(#(string/split % #" "))
@ -300,6 +300,7 @@
(comment
(phrase->all-flex-rhyme-tailing-consonants-phones "robot")
;; => ([("OW1" "AA" "T") "robot"] [("OW1" "AH" "T") "robot"])
)

@ -1,312 +0,0 @@
(ns com.owoga.prhyme.gen
(:require [clojure.string :as string]
[com.owoga.prhyme.util.math :as math]
[com.owoga.prhyme.util :as util]
[com.owoga.prhyme.util.weighted-rand :as weighted-rand]
[com.owoga.prhyme.nlp.core :as nlp]
[com.owoga.prhyme.frp :as frp]
[com.owoga.prhyme.core :as prhyme]))
(def words-map
(into {} (map #(vector (string/lower-case (:word %)) %) frp/words)))
(defn merge-phrase-words
"Given multiple `Word`, like the words for 'well off', create a single `Word`
that is syllabified as ('well' 'off') rather than as the combined ('weh'
'loff'). Useful for finding single-word rhymes of multiple-word targets.
An example: 'war on crime' -> 'turpentine'.
As opposed to: 'war on crime' -> 'caw fawn lime'."
[phrase phrase-words]
(loop [merged (first phrase-words)
phrase-words (rest phrase-words)]
(cond
(and (empty? phrase-words) (empty? merged)) nil
(empty? phrase-words) (assoc merged :word phrase)
:else (recur (-> merged
(assoc :syllables (concat (:syllables merged)
(:syllables (first phrase-words))))
(assoc :syllable-count (+ (:syllable-count merged)
(:syllable-count (first phrase-words))))
(assoc :rimes (concat (:rimes merged)
(:rimes (first phrase-words))))
(assoc :onsets (concat (:onsets merged)
(:onsets (first phrase-words))))
(assoc :nuclei (concat (:nuclei merged)
(:nuclei (first phrase-words)))))
(rest phrase-words)))))
(defn adjust-for-markov
[markov percent]
(let [target-markov-n (count (first (first markov)))]
(fn [[words target result]]
(if (>= (count result) target-markov-n)
(let [markov-options (markov (->> result
(take target-markov-n)
(map :normalized-word)))
markov-option-avg (/ (apply + (vals markov-options))
(max 1 (count markov-options)))]
(if (nil? markov-options)
[words target result]
(let [[markovs non-markovs]
((juxt filter remove)
#(markov-options (:normalized-word %))
words)
weight-non-markovs (apply + (map :weight non-markovs))
target-weight-markovs (* 100 percent weight-non-markovs)
count-markovs (count markovs)
adjustment-markovs (if (= 0 count-markovs) 1 (/ target-weight-markovs count-markovs))]
[(concat
(map
(fn [m]
(let [option (markov-options (:normalized-word m))]
(as-> m m
(assoc m :weight (* (/ option markov-option-avg) adjustment-markovs (:weight m)))
(assoc m :adjustment-for-markov (* (/ option markov-option-avg) adjustment-markovs)))))
markovs)
non-markovs)
target
result])))
[words target result]))))
(defn adjust-for-markov-with-boundaries
[markov percent]
(let [markov-n (count (first (first markov)))]
(fn [[words target result]]
(let [key (let [k (map :normalized-word (take markov-n result))]
(reverse
(if (> markov-n (count k))
(concat k (repeat (- markov-n (count k)) nil))
k)))
markov-options (markov key)
markov-option-avg (/ (apply + (vals markov-options))
(max 1 (count markov-options)))]
(if (nil? markov-options)
[words target result]
(let [[markovs non-markovs]
((juxt filter remove)
#(markov-options (:normalized-word %))
words)
weight-non-markovs (apply + (map :weight non-markovs))
target-weight-markovs (- (/ weight-non-markovs (- 1 percent))
weight-non-markovs)
count-markovs (count markovs)
adjustment-markovs (if (= 0 count-markovs) 1 (/ target-weight-markovs count-markovs))]
[(concat
(map
(fn [m]
(let [option (markov-options (:normalized-word m))]
(as-> m m
(assoc m :weight (* (/ option markov-option-avg) adjustment-markovs (:weight m)))
(assoc m :adjustment-for-markov (* (/ option markov-option-avg) adjustment-markovs)))))
markovs)
non-markovs)
target
result]))))))
(defn adjust-for-rimes
[dictionary percent]
(fn [[words target result]]
(let [words-with-rime-count
(map
(fn [word]
(assoc word :num-matching (count (frp/consecutive-matching target word :rimes))))
words)
[rhyming non-rhyming]
((juxt filter remove)
#(< 0 (:num-matching %))
words-with-rime-count)
weight-non-rhyming (apply + (map :weight non-rhyming))
target-weight-rhyming (* 100 percent weight-non-rhyming)
count-rhyming (count rhyming)
adjustment-rhyming (if (= 0 count-rhyming) 1 (/ target-weight-rhyming count-rhyming))]
[(concat
(map
(fn [word]
(as-> word word
(assoc word :weight (* adjustment-rhyming (:weight word)))
(assoc word :adjustment-for-rimes adjustment-rhyming)))
rhyming)
non-rhyming)
target
result])))
(defn attempt-gen-target-by-syllable-count [adj syllable-count words]
(loop [result '()]
(cond
(<= syllable-count (apply + (cons 0 (map :syllable-count result))))
result
:else
(let [[weighted-words target result] (adj [words nil result])]
(recur (cons (weighted-rand/weighted-selection :weight weighted-words) result))))))
(defn gen-sentence-with-syllable-count [adj syllable-count words]
(->> (repeatedly
(fn []
(attempt-gen-target-by-syllable-count adj syllable-count words)))
(filter #(= syllable-count (apply + (map :syllable-count %))))
(map #(map :normalized-word %))
(map #(string/join " " %))
(filter nlp/valid-sentence?)
first))
(defn remove-selection-from-target [target selection]
(->> target
(#(assoc % :syllables (drop-last
(:syllable-count
selection)
(:syllables
target))))
(#(assoc % :rimes (prhyme/rimes (:syllables %))))
(#(assoc % :onsets (prhyme/onset+nucleus (:syllables %))))
(#(assoc % :nuclei (prhyme/nucleus (:syllables %))))))
(defn selection-seq
([words adjust target]
(selection-seq words adjust target '()))
([words adjust target result]
(let [[weighted-words _ _] (adjust [words target result])
selection (math/weighted-selection :weight weighted-words)
new-target (remove-selection-from-target target selection)
new-result (cons selection result)]
(cons selection
(lazy-seq (selection-seq words adjust new-target new-result))))))
(defn selection-stream
"Continuously make the first selection."
([words adjust target]
(selection-stream words adjust target '()))
([words adjust target result]
(let [[weighted-words _ _] (adjust [words target result])]
(repeatedly #(math/weighted-selection :weight weighted-words)))))
(defn generate-prhyme [words adjust target stop?]
(loop [result '()]))
(defn prhyme
"2020-10-21 iteration"
[words weights-adjuster target stop?]
(let [target (assoc target :original-syllables (:syllables target))
words (map #(assoc % :weight 1) words)]
(loop [target target
result '()
sentinel 0]
(if (or (stop? target result)
(> sentinel 5))
result
(let [[weighted-words _ _] (weights-adjuster [words target result])
rng (weighted-rand/from-weights (map :weight weighted-words))
index (weighted-rand/nextr rng nil)
selection (nth weighted-words index)
new-target (->> target
(#(assoc % :syllables (drop-last
(:syllable-count
selection)
(:syllables
target))))
(#(assoc % :rimes (prhyme/rimes (:syllables %))))
(#(assoc % :onsets (prhyme/onset+nucleus (:syllables %))))
(#(assoc % :nuclei (prhyme/nucleus (:syllables %)))))
result (cons selection result)]
(recur new-target result (inc sentinel)))))))
(defn attempt-gen-rhyme-with-syllable-count [adj syllable-count words target]
(prhyme
words
adj
target
(fn [target result]
(<= syllable-count (apply + (map :syllable-count result))))))
(defn gen-rhyme-with-syllable-count [adj syllable-count words target]
(->> (repeatedly
(fn []
(attempt-gen-rhyme-with-syllable-count adj syllable-count words target)))
(filter #(= syllable-count (apply + (map :syllable-count %))))
(map #(map :normalized-word %))
(map #(string/join " " %))
(filter nlp/valid-sentence?)
first))
(defn prhymer [words weights-adjuster target stop]
(cons (prhyme
words
weights-adjuster
target
stop)
(lazy-seq (prhymer words weights-adjuster target stop))))
(defn sentence-stop [target]
(fn [inner-target result]
(let [result-sentence (string/join " " (map :normalized-word result))]
(when-not (empty? result)
(or (nlp/valid-sentence? result-sentence)
(< (:syllable-count target)
(apply + (map :syllable-count result)))
(< 5 (count result)))))))
(defn gen-prhymes [words adjust poem-lines]
(let [words (map #(assoc % :weight 1) words)
words-map (into {} (map #(vector (:normalized-word %) %) words))]
(map (fn [line]
(let [target (prhyme/phrase->word words line)
stop (sentence-stop target)
r (prhymer words adjust target stop)]
(string/join " " (map #(:normalized-word %) (first r)))))
poem-lines)))
(defn phrase-syllable-count [phrase]
(->> phrase
(#(string/split % #" "))
(map (partial prhyme/phrase->word frp/words))
(map :syllable-count)
(apply +)))
(defn filter-for-syllable-count [syllable-count coll]
(filter #(= syllable-count (phrase-syllable-count %)) coll))
(defn syllable-stop
[target]
(fn [inner-target result]
(<= (:syllable-count target)
(apply + (map :syllable-count result)))))
(defn generate-rhyme-for-phrase
[words adjust phrase]
(let [words (map #(assoc % :weight 1) words)
words-map (into {} (map #(vector (:normalized-word %) %) words))
target (prhyme/phrase->word words phrase)]
(prhymer words adjust target (syllable-stop target))))
#_(defn generate-prhymes [poem]
(let [r (partial generate-rhyme-for-phrase frp/popular adj)]
(fn []
(->> poem
(map (fn [phrase]
(let [target (prhyme/phrase->word frp/popular phrase)]
(first
(filter
#(and
(or (< 0.9 (rand))
(nlp/valid-sentence? (string/join " " (map :normalized-word %))))
(= (:syllable-count target)
(apply + (map :syllable-count %))))
(r phrase))))))
(map (fn [line] (map #(:normalized-word %) line)))
(map #(string/join " " %))))))
(defn generate-prhymes-darkov [words adj phrase]
(let [target (prhyme/phrase->word words phrase)
r (generate-rhyme-for-phrase words adj target)]
(first
(filter
#(and
(or (< 0.9 (rand))
(nlp/valid-sentence? (string/join " " (map :normalized-word %))))
(= (:syllable-count target)
(apply + (map :syllable-count %))))
r))
(map (fn [line] (map #(:normalized-word %) line)))
(map #(string/join " " %))))

@ -1,6 +1,5 @@
(ns com.owoga.prhyme.limerick
(:require [com.owoga.prhyme.gen :as gen]
[com.owoga.prhyme.generation.weighted-selection :as weighted-selection]
(:require [com.owoga.prhyme.generation.weighted-selection :as weighted-selection]
[com.owoga.prhyme.util.math :as math]
[com.owoga.prhyme.nlp.core :as nlp]
[clojure.string :as string]

@ -1,35 +0,0 @@
(ns com.owoga.frp.infrastructure-test
(:require [com.owoga.frp.infrastructure :as frp]
[clojure.test :refer [deftest is testing]]))
(deftest test-project
(testing "projection"
(let [Offer (frp/->BaseRelVar 'Offer (atom #{}) '())
OfferPrices (frp/project Offer [:price])]
(frp/load! Offer #{{:address "123 Fake St." :price 2e5}})
(is (= @OfferPrices #{{:price 2e5}})))))
(deftest test-insert!
(testing "insert!"
(let [Offer (frp/->BaseRelVar 'Offer (atom #{}) '())]
(frp/insert! Offer {:address "123 Fake St." :price 1.5e5})
(is (= @Offer #{{:address "123 Fake St." :price 1.5e5}})))))
(deftest test-defrelvar
(testing "failed constraint raises"
(let [Offer (frp/defrelvar Offer (fn [offers] (map #(> (:price %) 0) offers)))]
(is (thrown-with-msg?
Exception
#"Constraint Exception"
(frp/insert! Offer {:price -1})))))
(testing "passed constraint doesn't raise"
(let [Offer (frp/defrelvar Offer (fn [offers] (map #(> (:price %) 0) offers)))]
(frp/insert! Offer {:price 20})
(is (= @Offer #{{:price 20}})))))
(deftest test-extend
(testing "extend-"
(let [Offer (frp/->BaseRelVar 'Offer (atom #{}) '())]
(frp/load! Offer #{{:price 1e6}})
(frp/extend- Offer [:price-band (fn [e] (if (> (:price e) 1e6) :high :low))])
(is (= :low (-> @Offer first :price-band))))))
Loading…
Cancel
Save