This is my first attempt with a lisp language. I'd like your review to address the following points, already listed in order of relevance:

  1. Is my code making good use of standard library?
  2. Is my code using the right data structures?
  3. Is my code well formated and organized?
  4. Is my code computationally efficient?


(ns clojure-first-try.core)

(use '[clojure.string :only [split]])
(use '[clojure.set :only [intersection]])

(defn sum [arr]
  (reduce + 0.0 arr))

(defn squares [arr]
  (map #(* % %) arr))

(defn words-frequency [text]
  (as-> text $
    (split $ #"\s+")
    (filter (complement empty?) $)
    (frequencies $)))

(defn cos-numerator [map-1 map-2]
  (def common-keys (intersection (set (keys map-1)) (set (keys map-2))))
  (sum (map #(* (map-1 %) (map-2 %)) common-keys)))

(defn cos-denominator [map-1 map-2]
  (def p-1 (sum (squares (vals map-1))))
  (def p-2 (sum (squares (vals map-2))))
  (* (Math/sqrt p-1) (Math/sqrt p-2)))

(defn cos-similarity [freq-1 freq-2]
  (def a (cos-numerator freq-1 freq-2))
  (def b (cos-denominator freq-1 freq-2))
  (if (= b 0.0) 0.0 (/ a b)))

(defn sort-by-similarity [base options]
  (def base-freq (words-frequency base))
  (def sorter (comp (partial cos-similarity base-freq) words-frequency))
  (sort-by sorter > options))


(ns clojure-first-try.core-test
  (:require [clojure.test :refer :all]
            [clojure-first-try.core :refer :all]))

(defn close-enough? [a b]
  (def delta (- a b))
  (def abs-delta (max delta (- delta)))
  (> 0.0001 abs-delta))

(deftest words-frequency-test
  (testing "Counting the frequency of words in a text with actual words"
    (def expected {"a" 3, "bb" 2, "ccc" 4})
    (def input "ccc ccc a bb a bb a ccc ccc")
    (def result (words-frequency input))
    (is (= expected result)))
  (testing "Counting the frequency of words in an empty text"
    (def expected {})
    (def input "")
    (def result (words-frequency input))
    (is (= expected result))))

(deftest cos-similarity-test
  (testing "Getting the similarity of two equal texts"
    (def expected 1.0)
    (def input-1 {:a 1 :b 2})
    (def input-2 {:b 2 :a 1})
    (def result (cos-similarity input-1 input-2))
    (is (close-enough? expected result)))
  (testing "Getting the similarity of two totally different texts"
    (def expected 0.0)
    (def input-1 {:a 1 :b 2})
    (def input-2 {:c 2 :d 1})
    (def result (cos-similarity input-1 input-2))
    (is (close-enough? expected result)))
  (testing "Getting the similarity of two half equal texts"
    (def expected 0.5)
    (def input-1 {:a 2 :b 2})
    (def input-2 {:b 2 :d 2})
    (def result (cos-similarity input-1 input-2))
    (is (close-enough? expected result))))

(deftest sort-by-similarity-test
  (testing "Sorting regular texts"
    (def expected ["a b c" "a b d" "d e f"])
    (def input-1 "a b c")
    (def input-2 ["d e f" "a b c" "a b d"])
    (def result (sort-by-similarity input-1 input-2))
    (is (= expected result)))
  (testing "Sorting empty list of options")
    (def expected [])
    (def input-1 "a b c")
    (def input-2 [])
    (def result (sort-by-similarity input-1 input-2))
    (is (= expected result)))

All tests passing:

lein test clojure-first-try.core-test

Ran 3 tests containing 7 assertions.
0 failures, 0 errors.

First, the most serious problem with your code:

Never, ever, ever use def inside of defn, unless you really know what you're doing. def creates global variables that last even once the function has returned:

(defn func []
  (def x 1))

(println x) ; Prints 1! Ouch!

The only time it's really ever appropriate to use def locally is when you intend to create globals. This is an extremely rare case though.

Instead, you should be using let. For example:

(defn cos-denominator [map-1 map-2]
  (let [p-1 (sum (squares (vals map-1)))
        p-2 (sum (squares (vals map-2)))]
    (* (Math/sqrt p-1) (Math/sqrt p-2))))

let creates a limited scope for the vals that it defines. Once you leave the let, they go out of scope.

That's the only really outright wrong thing I see in your code. The rest of my comments will be best practice related.

use, require and related functions shouldn't really be used directly. It's much cleaner to have them as part of the ns macro:

(ns clojure-first-try.core
  (:use [clojure.string :only [split]]
        [clojure.set :only [intersection]]))

Your use of $ as an identifier is bizarre. I had to double take there. I would recommend using something a little more descriptive. If you can't think of a good identifier to use, I think even a single letter would be better than a odd symbol like $.

(reduce + 0.0 arr))

is more typically written as

(apply + arr)

Your ways not wrong, the latter is just generally considered more idiomatic. This works because + has a var-arg overload that is basically an explicit reduction.

Also, you don't need to specify 0.0 as the starting accumulator of the reduction. + handles that for you.

(reduce + []) ; 0

(filter (complement empty?) arr)

can be written as

(remove empty? arr)

remove adds complement for you! Here's the core definition (basically):

(defn remove [pred coll] ; Abridged defintion
   (filter (complement pred) coll))

It reads much nicer without the explicit call to complement.

(= b 0.0)

can also be written as just:

(zero? b)

Although the gains aren't huge. I find it's better when in an already busy line, like when checking for factors of a number. I find:

(zero? (rem a b))

to be neater than

(= 0 (rem a b))

For cos-numerator:

(intersection (set (keys map-1)) (set (keys map-2)))

is really inefficient. You iterate both maps just to put them in a set to test for intersections. Without doing more timing, it's unclear what the complexity of keys is. If it's O(n), this iterates each map twice, then again as a set. It's much more concise, and significantly faster to just use filter:

(filter map-1 (keys map-2))

Maps return nil on a bad lookup, so they can be used as predicates to check for membership.

I also think this could be cleaned up with ->>. I changed it to:

(defn cos-numerator [map-1 map-2]
  (->> (filter map-1 (keys map-2))
       (map #(* (map-1 %) (map-2 %)))

squares could also be written as

(map * arr arr))

You're going to have a little bit of duplication somewhere. It just depends what you want it to look like. I find this a little neater.

I'm a big fan of local anonymous functions to clean up code. cos-denominator has you doing the exact same thing to both maps, then taking their product.

I'd make the common transformation a local function, then use that instead:

(defn cos-denominator [map-1 map-2]
  (let [proc #(-> % (vals) (squares) (sum) (Math/sqrt))]
    (* (proc map-1) (proc map-2))))

That has much less duplication. You could even go a little overboard and get rid of the duplicate calls to proc:

(defn cos-denominator2 [map-1 map-2]
  (let [proc #(-> % (vals) (squares) (sum) (Math/sqrt))]
    (apply * (map proc [map-1 map-2]))))

Which could also be made a little neater with ->>:

(defn cos-denominator3 [map-1 map-2]
  (let [proc #(-> % (vals) (squares) (sum) (Math/sqrt))]
    (->> [map-1 map-2]
         (map proc)
         (apply *))))

After the above changes, and a few other touchups that I thought made it neater, this is what I ended up with:

(ns clojure-first-try.core
  (:use [clojure.string :only [split]]
        [clojure.set :only [intersection]]
        [clojure.test :refer :all])
  (:require [criterium.core :as c]))

(defn sum [arr]
  (apply + arr))

(defn squares [arr]
  (map * arr arr))

(defn words-frequency [text]
  (as-> text t
        (split t #"\s+")
        (filter (complement empty?) t)
        (frequencies t)))

(defn cos-numerator [map-1 map-2]
  (->> (filter map-1 (keys map-2))
       (map #(* (map-1 %) (map-2 %)))

(defn cos-denominator [map-1 map-2]
  (let [proc #(-> % (vals) (squares) (sum) (Math/sqrt))]
    (->> [map-1 map-2]
         (map proc)
         (apply *))))

(defn cos-similarity [freq-1 freq-2]
  (let [a (cos-numerator freq-1 freq-2)
        b (cos-denominator freq-1 freq-2)]
    (if (zero? b)
      (/ a b))))

(defn sort-by-similarity [base options]
  (let [base-freq (words-frequency base)
        sorter (comp #(cos-similarity base-freq %) words-frequency)]
    (sort-by sorter > options)))
  • \$\begingroup\$ I saw the "bizarre" $ in the docs of as->. clojuredocs.org/clojure.core/as-> \$\endgroup\$ – Gabriel May 3 '18 at 16:57
  • \$\begingroup\$ @Gabriel I'm not sure who that "freed" guy is. The examples in the docs are community supplied. I'm not saying it's necessarily wrong, I just can't say I've ever seen a single, non-letter symbol used before. I wouldn't be surprised if other Clojurists double take on that as well. It's reminiscent of #()'s use of %, which suggested to me originally that it had some special use in a user-defined macro or something. \$\endgroup\$ – Carcigenicate May 3 '18 at 17:01
  • \$\begingroup\$ ACK. Looking forward to your next review with new eyes. \$\endgroup\$ – Gabriel May 3 '18 at 17:20
  • \$\begingroup\$ @Gabriel I'll try to comment on the algorithm. I didn't originally because honestly, I don't fully understand what the code is doing. I'll need to look it over again a few times. \$\endgroup\$ – Carcigenicate May 3 '18 at 17:27
  • \$\begingroup\$ words-frequency accepts a string and returns a map where the key is a word and the value is a number of how many times this word appears in the string. cos-similarity accepts two maps frequencies (generated by words-frequency) and returns a number from 0.0 to 1.0 representing how much the original texts are similar (1.0 when they are fully equal). See this Wikipedia entry: en.wikipedia.org/wiki/Cosine_similarity \$\endgroup\$ – Gabriel May 3 '18 at 17:50

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.