This is an automated email from the ASF dual-hosted git repository.
paulk pushed a commit to branch asf-site
in repository https://gitbox.apache.org/repos/asf/groovy-website.git
The following commit(s) were added to refs/heads/asf-site by this push:
new f6cf0c1 a few more images
f6cf0c1 is described below
commit f6cf0c1e6c10a2ade9c308354635815c5a1506fa
Author: Paul King <[email protected]>
AuthorDate: Tue Feb 18 13:16:37 2025 +1000
a few more images
---
site/src/site/blog/groovy-text-similarity.adoc | 257 ++++++++++++++++---------
site/src/site/blog/img/gameBubble.png | Bin 0 -> 157410 bytes
site/src/site/blog/img/semantle.png | Bin 0 -> 74004 bytes
site/src/site/blog/img/wordle.png | Bin 0 -> 95851 bytes
4 files changed, 171 insertions(+), 86 deletions(-)
diff --git a/site/src/site/blog/groovy-text-similarity.adoc
b/site/src/site/blog/groovy-text-similarity.adoc
index 9f31134..21e8e2c 100644
--- a/site/src/site/blog/groovy-text-similarity.adoc
+++ b/site/src/site/blog/groovy-text-similarity.adoc
@@ -612,7 +612,8 @@ hippo|hippopotamus 50% 40% 40%
== Going Deeper
Rather than finding similarity based on a word's individual letters, or
phonetic mappings,
-_machine learning_/_deep learning_ tries to relate words with similar semantic
meaning. The approach maps each word (or phrase) in n-dimensional space (called
a _word vector_ or _word embedding_).
+_machine learning_ and _deep learning_ try to relate words with similar
semantic meaning.
+The approach maps each word (or phrase) in n-dimensional space (called a _word
vector_ or _word embedding_).
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
https://en.wikipedia.org/wiki/Cosine_similarity[cosine similarity]
@@ -653,16 +654,18 @@ and can then call methods like `similarity` and
`wordsNearest` as shown here:
[source,groovy]
----
-var path =
Paths.get(ConceptNet.classLoader.getResource('glove-wiki-gigaword-300.bin').toURI()).toFile()
+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)
+ .collectValues('%4.2f'::formatted)
+}
+Nearest words in vocab: ${model.wordsNearest('cow', 4)}
+"""
----
Which gives this output:
@@ -674,17 +677,128 @@ Nearest words in vocab: [cows, mad, bovine, cattle]
=== FastText
-We can swap to a https://fasttext.cc/[FastText] model. We used [this model]
which has
+We can swap to a https://fasttext.cc/[FastText] model, simply by switching to
that model:
+
+[source,groovy]
+----
+var modelName = 'fasttext-wiki-news-subwords-300.bin'
+----
+
+We used https://huggingface.co/fse/fasttext-wiki-news-subwords-300[this model]
which has
1 million word vectors trained on Wikipedia 2017, UMBC webbase corpus and
statmt.org news dataset (16B tokens).
-It has this output:
+When run with the FastText model, the script has this output:
----
FastText similarity to cow: [bovine:0.72, cattle:0.70, calf:0.67, bull:0.67,
livestock:0.61, horse:0.60]
Nearest words in vocab: [cows, goat, pig, bovine]
----
-Using DJL with PyTorch and the Angle model:
+=== ConceptNet
+
+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":
+
+[source,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)}
+"""
+----
+
+----
+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]
+----
+
+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:
+
+----
+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 ██▏
+----
+
+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.
+
+=== AnglE
+
+Using DJL with PyTorch and the AnglE model:
----
cow
@@ -737,6 +851,8 @@ One two three (0.43)
bovine (0.39)
----
+=== UAE
+
Using DJL with Tensorflow and the UAE model:
----
@@ -794,75 +910,8 @@ image:img/AnimalSemanticSimilarity.png[]
image:img/AnimalSemanticMeaningPcaBubblePlot.png[]
-----
-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 ██▏
-----
+=== Comparing Algorithm Choices
----
Algorithm angle use conceptnet
glove fasttext
@@ -1032,7 +1081,7 @@ Possible letters: a b c d e f g h i j k l m n o p q r s t
u v w x y z
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%
@@ -1048,7 +1097,7 @@ Possible letters: a b c d e f g h i j k l m n o p q r s t
u v w x y z
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%
@@ -1063,7 +1112,7 @@ Possible letters: a b c d e f g h i j k l m n o p q r s t
u v w x y z
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%
@@ -1076,7 +1125,7 @@ Possible letters: a b c d e f g h i j k l m n o p q r s t
u v w x y z
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%
@@ -1093,7 +1142,7 @@ Possible letters: a b c d e f g h i j k l m n o p q r s t
u v w x y z
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%
@@ -1106,7 +1155,7 @@ Possible letters: a b c d e f g h i j k l m n o p q r s t
u v w x y z
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%
@@ -1122,7 +1171,7 @@ Possible letters: a b c d e f g h i j k l m n o p q r s t
u v w x y z
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%
@@ -1160,7 +1209,7 @@ Possible letters: a b c d e f g h i j k l m n o p q r s t
u v w x y z
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%
@@ -1181,7 +1230,7 @@ Possible letters: a b c d e f g h i j k l m n o p q r s t
u v w x y z
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%
@@ -1201,7 +1250,7 @@ Possible letters: a b c d e f g h i j k l m n o p q r s t
u v w x y z
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%
@@ -1211,6 +1260,42 @@ Congratulations, you guessed correctly!
Success!
+=== Hints
+
+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 `wordsNearest` method
+for the three word2vec models and then selected a subset.
+
+Although not needed in the games we have shown,
+here are what the hints would have been for Round 3.
+
+----
+After round 8: root_vegetable, daucus
+After round 16: diced, cauliflower, cucumber
+After round 24: celery, onion, sticks, zucchini
+----
+
+=== Further Evolution
+
+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
https://semantle.com/[Semantle]:
+
+image:img/semantle.png[semantle game,width=50%]
+
+And we could add a prettier representation of available letters, e.g. greyed
out keys on a keyboard, like in
https://www.nytimes.com/games/wordle/index.html[Wordle]:
+
+image:img/wordle.png[world game,width=30%]
+
+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):
+
+image:img/gameBubble.png[Game BubleChart,width=70%]
+
== Further information [[further_info]]
Source code for this post:
diff --git a/site/src/site/blog/img/gameBubble.png
b/site/src/site/blog/img/gameBubble.png
new file mode 100644
index 0000000..0889b35
Binary files /dev/null and b/site/src/site/blog/img/gameBubble.png differ
diff --git a/site/src/site/blog/img/semantle.png
b/site/src/site/blog/img/semantle.png
new file mode 100644
index 0000000..0e20db8
Binary files /dev/null and b/site/src/site/blog/img/semantle.png differ
diff --git a/site/src/site/blog/img/wordle.png
b/site/src/site/blog/img/wordle.png
new file mode 100644
index 0000000..29ccd04
Binary files /dev/null and b/site/src/site/blog/img/wordle.png differ
