This is an automated email from the ASF dual-hosted git repository.
git-site-role pushed a commit to branch asf-site
in repository https://gitbox.apache.org/repos/asf/groovy-dev-site.git
The following commit(s) were added to refs/heads/asf-site by this push:
new 580e791 2025/02/18 03:35:56: Generated dev website from
groovy-website@f6cf0c1
580e791 is described below
commit 580e7919f293d2fdf7920ae3a99e240b154cacfe
Author: jenkins <[email protected]>
AuthorDate: Tue Feb 18 03:35:56 2025 +0000
2025/02/18 03:35:56: Generated dev website from groovy-website@f6cf0c1
---
blog/groovy-text-similarity.html | 280 +++++++++++++++++++++++++++------------
blog/img/gameBubble.png | Bin 0 -> 157410 bytes
blog/img/semantle.png | Bin 0 -> 74004 bytes
blog/img/wordle.png | Bin 0 -> 95851 bytes
4 files changed, 192 insertions(+), 88 deletions(-)
diff --git a/blog/groovy-text-similarity.html b/blog/groovy-text-similarity.html
index 29fe869..c87b116 100644
--- a/blog/groovy-text-similarity.html
+++ b/blog/groovy-text-similarity.html
@@ -831,7 +831,8 @@ hippo|hippopotamus 50% 40% 40%
<div class="sectionbody">
<div class="paragraph">
<p>Rather than finding similarity based on a word’s individual letters,
or phonetic mappings,
-<em>machine learning</em>/<em>deep learning</em> tries to relate words with
similar semantic meaning. The approach maps each word (or phrase) in
n-dimensional space (called a <em>word vector</em> or <em>word embedding</em>).
+<em>machine learning</em> and <em>deep learning</em> try to relate words with
similar semantic meaning.
+The approach maps each word (or phrase) in n-dimensional space (called a
<em>word vector</em> or <em>word embedding</em>).
Related words tend to cluster in similar positions within that space.
Typically rule-based, statistical, or neural-based approaches are used to
perform the embedding
and distance measures like <a
href="https://en.wikipedia.org/wiki/Cosine_similarity";>cosine similarity</a>
@@ -880,16 +881,18 @@ and can then call methods like <code>similarity</code>
and <code>wordsNearest</c
</div>
<div class="listingblock">
<div class="content">
-<pre class="prettyprint highlight"><code data-lang="groovy">var path =
Paths.get(ConceptNet.classLoader.getResource('glove-wiki-gigaword-300.bin').toURI()).toFile()
+<pre class="prettyprint highlight"><code data-lang="groovy">var modelName =
'glove-wiki-gigaword-300.bin'
+var path =
Paths.get(ConceptNet.classLoader.getResource(modelName).toURI()).toFile()
Word2Vec model = WordVectorSerializer.readWord2VecModel(path)
String[] words = ['bull', 'calf', 'bovine', 'cattle', 'livestock', 'horse']
println """GloVe similarity to cow: ${
words
.collectEntries { [it, model.similarity('cow', it)] }
.sort { -it.value }
- .collectValues{ sprintf '%4.2f', it }
-}"""
-println "Nearest words in vocab: " + model.wordsNearest('cow', 4)</code></pre>
+ .collectValues('%4.2f'::formatted)
+}
+Nearest words in vocab: ${model.wordsNearest('cow', 4)}
+"""</code></pre>
</div>
</div>
<div class="paragraph">
@@ -905,11 +908,19 @@ Nearest words in vocab: [cows, mad, bovine, cattle]</pre>
<div class="sect2">
<h3 id="_fasttext">FastText</h3>
<div class="paragraph">
-<p>We can swap to a <a href="https://fasttext.cc/";>FastText</a> model. We used
[this model] which has
+<p>We can swap to a <a href="https://fasttext.cc/";>FastText</a> model, simply
by switching to that model:</p>
+</div>
+<div class="listingblock">
+<div class="content">
+<pre class="prettyprint highlight"><code data-lang="groovy">var modelName =
'fasttext-wiki-news-subwords-300.bin'</code></pre>
+</div>
+</div>
+<div class="paragraph">
+<p>We used <a
href="https://huggingface.co/fse/fasttext-wiki-news-subwords-300";>this
model</a> which has
1 million word vectors trained on Wikipedia 2017, UMBC webbase corpus and
statmt.org news dataset (16B tokens).</p>
</div>
<div class="paragraph">
-<p>It has this output:</p>
+<p>When run with the FastText model, the script has this output:</p>
</div>
<div class="listingblock">
<div class="content">
@@ -917,8 +928,119 @@ Nearest words in vocab: [cows, mad, bovine, cattle]</pre>
Nearest words in vocab: [cows, goat, pig, bovine]</pre>
</div>
</div>
+</div>
+<div class="sect2">
+<h3 id="_conceptnet">ConceptNet</h3>
+<div class="paragraph">
+<p>Similarly, we can switch to a ConceptNet model through a change of the
model name.
+This model also supports multiple languages and incorporates the language used
into terms, e.g. for English,
+we use "/c/en/cow" instead of "cow":</p>
+</div>
+<div class="listingblock">
+<div class="content">
+<pre class="prettyprint highlight"><code data-lang="groovy">var modelName =
'conceptnet-numberbatch-17-06-300.bin'
+...
+println """ConceptNet similarity to /c/en/cow: ${
+ words
+ .collectEntries { ["/c/en/$it", model.similarity('/c/en/cow',
"/c/en/$it")] }
+ .sort { -it.value }
+ .collectValues('%4.2f'::formatted)
+}
+Nearest words in vocab: ${model.wordsNearest('/c/en/cow', 4)}
+"""</code></pre>
+</div>
+</div>
+<div class="listingblock">
+<div class="content">
+<pre>ConceptNet similarity to /c/en/cow: [/c/en/bovine:0.77,
/c/en/cattle:0.77, /c/en/livestock:0.63, /c/en/bull:0.54, /c/en/calf:0.53,
/c/en/horse:0.50]
+Nearest words in vocab: [/c/ast/vaca, /c/be/карова, /c/ur/گای,
/c/gv/booa]</pre>
+</div>
+</div>
+<div class="paragraph">
+<p>There are benefits and costs with using a multilingual model. The model
itself is bigger and takes longer to load.
+It will typically need more memory to use, but it does allow us to consider
multilingual options if we wanted to
+as the following results show:</p>
+</div>
+<div class="listingblock">
+<div class="content">
+<pre>Algorithm conceptnet
+
+ /c/fr/vache █████████▏
+ /c/de/kuh █████████▏
+/c/en/cow /c/en/bovine ███████▏
+ /c/fr/bovin ███████▏
+ /c/en/bull █████▏
+
+ /c/fr/taureau █████████▏
+ /c/en/cow █████▏
+/c/en/bull /c/fr/vache █████▏
+ /c/de/kuh █████▏
+ /c/fr/bovin █████▏
+
+ /c/de/kuh █████▏
+ /c/en/cow █████▏
+/c/en/calf /c/fr/vache █████▏
+ /c/en/bovine █████▏
+ /c/fr/bovin █████▏
+
+ /c/fr/bovin █████████▏
+ /c/en/cow ███████▏
+/c/en/bovine /c/de/kuh ███████▏
+ /c/fr/vache ███████▏
+ /c/en/calf █████▏
+
+ /c/en/bovine █████████▏
+ /c/fr/vache ███████▏
+/c/fr/bovin /c/de/kuh ███████▏
+ /c/en/cow ███████▏
+ /c/fr/taureau █████▏
+
+ /c/en/cow █████████▏
+ /c/de/kuh █████████▏
+/c/fr/vache /c/fr/bovin ███████▏
+ /c/en/bovine ███████▏
+ /c/fr/taureau █████▏
+
+ /c/en/bull █████████▏
+ /c/fr/bovin █████▏
+/c/fr/taureau /c/fr/vache █████▏
+ /c/en/cow █████▏
+ /c/de/kuh █████▏
+
+ /c/en/cow █████████▏
+ /c/fr/vache █████████▏
+/c/de/kuh /c/fr/bovin ███████▏
+ /c/en/bovine ███████▏
+ /c/en/calf █████▏
+
+ /c/en/cat ████████▏
+ /c/de/katze ████████▏
+/c/en/kitten /c/en/bull ██▏
+ /c/en/cow █▏
+ /c/de/kuh █▏
+
+ /c/de/katze █████████▏
+ /c/en/kitten ████████▏
+/c/en/cat /c/en/bull ██▏
+ /c/en/cow ██▏
+ /c/fr/taureau █▏
+
+ /c/en/cat █████████▏
+ /c/en/kitten ████████▏
+/c/de/katze /c/en/bull ██▏
+ /c/de/kuh ██▏
+ /c/fr/taureau ██▏</pre>
+</div>
+</div>
+<div class="paragraph">
+<p>We won’t use this feature for our game, but it would be a great thing
to add
+if you speak multiple languages or if you were learning a new language.</p>
+</div>
+</div>
+<div class="sect2">
+<h3 id="_angle">AnglE</h3>
<div class="paragraph">
-<p>Using DJL with PyTorch and the Angle model:</p>
+<p>Using DJL with PyTorch and the AnglE model:</p>
</div>
<div class="listingblock">
<div class="content">
@@ -972,6 +1094,9 @@ One two three (0.43)
bovine (0.39)</pre>
</div>
</div>
+</div>
+<div class="sect2">
+<h3 id="_uae">UAE</h3>
<div class="paragraph">
<p>Using DJL with Tensorflow and the UAE model:</p>
</div>
@@ -1033,77 +1158,9 @@ One two three (0.17)</pre>
<div class="paragraph">
<p><span class="image"><img src="img/AnimalSemanticMeaningPcaBubblePlot.png"
alt="AnimalSemanticMeaningPcaBubblePlot"></span></p>
</div>
-<div class="listingblock">
-<div class="content">
-<pre>Algorithm conceptnet
-
- /c/fr/vache █████████▏
- /c/de/kuh █████████▏
-/c/en/cow /c/en/bovine ███████▏
- /c/fr/bovin ███████▏
- /c/en/bull █████▏
-
- /c/fr/taureau █████████▏
- /c/en/cow █████▏
-/c/en/bull /c/fr/vache █████▏
- /c/de/kuh █████▏
- /c/fr/bovin █████▏
-
- /c/de/kuh █████▏
- /c/en/cow █████▏
-/c/en/calf /c/fr/vache █████▏
- /c/en/bovine █████▏
- /c/fr/bovin █████▏
-
- /c/fr/bovin █████████▏
- /c/en/cow ███████▏
-/c/en/bovine /c/de/kuh ███████▏
- /c/fr/vache ███████▏
- /c/en/calf █████▏
-
- /c/en/bovine █████████▏
- /c/fr/vache ███████▏
-/c/fr/bovin /c/de/kuh ███████▏
- /c/en/cow ███████▏
- /c/fr/taureau █████▏
-
- /c/en/cow █████████▏
- /c/de/kuh █████████▏
-/c/fr/vache /c/fr/bovin ███████▏
- /c/en/bovine ███████▏
- /c/fr/taureau █████▏
-
- /c/en/bull █████████▏
- /c/fr/bovin █████▏
-/c/fr/taureau /c/fr/vache █████▏
- /c/en/cow █████▏
- /c/de/kuh █████▏
-
- /c/en/cow █████████▏
- /c/fr/vache █████████▏
-/c/de/kuh /c/fr/bovin ███████▏
- /c/en/bovine ███████▏
- /c/en/calf █████▏
-
- /c/en/cat ████████▏
- /c/de/katze ████████▏
-/c/en/kitten /c/en/bull ██▏
- /c/en/cow █▏
- /c/de/kuh █▏
-
- /c/de/katze █████████▏
- /c/en/kitten ████████▏
-/c/en/cat /c/en/bull ██▏
- /c/en/cow ██▏
- /c/fr/taureau █▏
-
- /c/en/cat █████████▏
- /c/en/kitten ████████▏
-/c/de/katze /c/en/bull ██▏
- /c/de/kuh ██▏
- /c/fr/taureau ██▏</pre>
-</div>
</div>
+<div class="sect2">
+<h3 id="_comparing_algorithm_choices">Comparing Algorithm Choices</h3>
<div class="listingblock">
<div class="content">
<pre>Algorithm angle use conceptnet
glove fasttext
@@ -1303,7 +1360,7 @@ a kind of food. It’s a 50/50 guess. Let’s try
the first.</p>
Guess the hidden word (turn 4): budding
LongestCommonSubsequence 6
Levenshtein Distance: 1, Insert: 0, Delete: 0, Substitute: 1
-Jaccard 71% (5/7)
+Jaccard 71%
JaroWinkler PREFIX 90% / SUFFIX 96%
Phonetic Metaphone=BTNK 79% / Soundex=B352 75%
Meaning Angle 52% / Use 35% / ConceptNet 2% / GloVe 4%
/ FastText 25%</pre>
@@ -1321,7 +1378,7 @@ Our other guess of pudding sounds right. Let’s try
it.</p>
Guess the hidden word (turn 5): pudding
LongestCommonSubsequence 7
Levenshtein Distance: 0, Insert: 0, Delete: 0, Substitute: 0
-Jaccard 100% (6/6)
+Jaccard 100%
JaroWinkler PREFIX 100% / SUFFIX 100%
Phonetic Metaphone=PTNK 100% / Soundex=P352 100%
Meaning Angle 100% / Use 100% / ConceptNet 100% / GloVe
100% / FastText 100%
@@ -1338,7 +1395,7 @@ Congratulations, you guessed correctly!</pre>
Guess the hidden word (turn 1): bail
LongestCommonSubsequence 1
Levenshtein Distance: 7, Insert: 4, Delete: 0, Substitute: 3
-Jaccard 22% (2/9) 2 / 9
+Jaccard 22%
JaroWinkler PREFIX 42% / SUFFIX 46%
Phonetic Metaphone=BL 38% / Soundex=B400 25%
Meaning Angle 46% / Use 40% / ConceptNet 0% / GloVe 0%
/ FastText 31%</pre>
@@ -1363,7 +1420,7 @@ Meaning Angle 46% / Use 40% /
ConceptNet 0% / GloVe 0% /
Guess the hidden word (turn 2): leg
LongestCommonSubsequence 2
Levenshtein Distance: 6, Insert: 5, Delete: 0, Substitute: 1
-Jaccard 25% (2/8) 1 / 4
+Jaccard 25%
JaroWinkler PREFIX 47% / SUFFIX 0%
Phonetic Metaphone=LK 38% / Soundex=L200 0%
Meaning Angle 50% / Use 18% / ConceptNet 11% / GloVe
13% / FastText 37%</pre>
@@ -1395,7 +1452,7 @@ encoded to either an 'L' or 'K'.</p>
Guess the hidden word (turn 3): languish
LongestCommonSubsequence 2
Levenshtein Distance: 8, Insert: 0, Delete: 0, Substitute: 8
-Jaccard 15% (2/13) 2 / 13
+Jaccard 15%
JaroWinkler PREFIX 50% / SUFFIX 50%
Phonetic Metaphone=LNKX 34% / Soundex=L522 0%
Meaning Angle 46% / Use 12% / ConceptNet -11% / GloVe
-4% / FastText 25%</pre>
@@ -1417,7 +1474,7 @@ Meaning Angle 46% / Use 12% /
ConceptNet -11% / GloVe -4%
Guess the hidden word (turn 4): election
LongestCommonSubsequence 5
Levenshtein Distance: 4, Insert: 0, Delete: 0, Substitute: 4
-Jaccard 40% (4/10) 2 / 5
+Jaccard 40%
JaroWinkler PREFIX 83% / SUFFIX 75%
Phonetic Metaphone=ELKXN 50% / Soundex=E423 75%
Meaning Angle 47% / Use 13% / ConceptNet -5% / GloVe
-7% / FastText 26%</pre>
@@ -1448,7 +1505,7 @@ Meaning Angle 47% / Use 13% /
ConceptNet -5% / GloVe -7%
Guess the hidden word (turn 5): elevator
LongestCommonSubsequence 8
Levenshtein Distance: 0, Insert: 0, Delete: 0, Substitute: 0
-Jaccard 100% (7/7) 1
+Jaccard 100%
JaroWinkler PREFIX 100% / SUFFIX 100%
Phonetic Metaphone=ELFTR 100% / Soundex=E413 100%
Meaning Angle 100% / Use 100% / ConceptNet 100% / GloVe
100% / FastText 100%
@@ -1499,7 +1556,7 @@ we aren’t duplicating a letter yet, but we just
want to narrow down the po
Guess the hidden word (turn 2): coarse
LongestCommonSubsequence 3
Levenshtein Distance: 4, Insert: 0, Delete: 0, Substitute: 4
-Jaccard 57% (4/7) 4 / 7
+Jaccard 57%
JaroWinkler PREFIX 67% / SUFFIX 67%
Phonetic Metaphone=KRS 74% / Soundex=C620 75%
Meaning Angle 51% / Use 12% / ConceptNet 5% / GloVe 23%
/ FastText 26%</pre>
@@ -1530,7 +1587,7 @@ and we’ll duplicate one letter, S.</p>
Guess the hidden word (turn 3): roasts
LongestCommonSubsequence 3
Levenshtein Distance: 6, Insert: 0, Delete: 0, Substitute: 6
-Jaccard 67% (4/6) 2 / 3
+Jaccard 67%
JaroWinkler PREFIX 56% / SUFFIX 56%
Phonetic Metaphone=RSTS 61% / Soundex=R232 25%
Meaning Angle 54% / Use 25% / ConceptNet 18% / GloVe
18% / FastText 31%</pre>
@@ -1560,7 +1617,7 @@ Maybe the hidden word is related to roasts.</p>
Guess the hidden word (turn 4): carrot
LongestCommonSubsequence 6
Levenshtein Distance: 0, Insert: 0, Delete: 0, Substitute: 0
-Jaccard 100% (5/5) 1
+Jaccard 100%
JaroWinkler PREFIX 100% / SUFFIX 100%
Phonetic Metaphone=KRT 100% / Soundex=C630 100%
Meaning Angle 100% / Use 100% / ConceptNet 100% / GloVe
100% / FastText 100%
@@ -1572,6 +1629,53 @@ Congratulations, you guessed correctly!</pre>
<p>Success!</p>
</div>
</div>
+<div class="sect2">
+<h3 id="_hints">Hints</h3>
+<div class="paragraph">
+<p>Some word guessing games allow the player to ask for hints.
+For our game, we decided to provide hints at regular intervals,
+giving stronger hints as the game progressed. We used the
+20 nearest similar words as returned by the <code>wordsNearest</code> method
+for the three word2vec models and then selected a subset.</p>
+</div>
+<div class="paragraph">
+<p>Although not needed in the games we have shown,
+here are what the hints would have been for Round 3.</p>
+</div>
+<div class="listingblock">
+<div class="content">
+<pre>After round 8: root_vegetable, daucus
+After round 16: diced, cauliflower, cucumber
+After round 24: celery, onion, sticks, zucchini</pre>
+</div>
+</div>
+</div>
+<div class="sect2">
+<h3 id="_further_evolution">Further Evolution</h3>
+<div class="paragraph">
+<p>Our goal was to introduce you to a number of algorithms that you might use
in a word game,
+rather than create a fully-polished game. If we were going to progress such a
game, one of the
+challenges would be how to represent the large number of parameters to the
user after each round.
+We could work on some pretty bar-charts like in <a
href="https://semantle.com/";>Semantle</a>:</p>
+</div>
+<div class="paragraph">
+<p><span class="image"><img src="img/semantle.png" alt="semantle game"
width="50%"></span></p>
+</div>
+<div class="paragraph">
+<p>And we could add a prettier representation of available letters, e.g.
greyed out keys on a keyboard, like in <a
href="https://www.nytimes.com/games/wordle/index.html";>Wordle</a>:</p>
+</div>
+<div class="paragraph">
+<p><span class="image"><img src="img/wordle.png" alt="world game"
width="30%"></span></p>
+</div>
+<div class="paragraph">
+<p>But we might also just use a bubble-chart, like we showed earlier,
+and let datascience condense the results for us. We might end up with
+a chart something like this (some guesses and hints for Round 3 shown):</p>
+</div>
+<div class="paragraph">
+<p><span class="image"><img src="img/gameBubble.png" alt="Game BubleChart"
width="70%"></span></p>
+</div>
+</div>
</div>
</div>
<div class="sect1">
diff --git a/blog/img/gameBubble.png b/blog/img/gameBubble.png
new file mode 100644
index 0000000..0889b35
Binary files /dev/null and b/blog/img/gameBubble.png differ
diff --git a/blog/img/semantle.png b/blog/img/semantle.png
new file mode 100644
index 0000000..0e20db8
Binary files /dev/null and b/blog/img/semantle.png differ
diff --git a/blog/img/wordle.png b/blog/img/wordle.png
new file mode 100644
index 0000000..29ccd04
Binary files /dev/null and b/blog/img/wordle.png differ
