Clustering

Basic Clustering Task

Use the following dataset:


In [24]:
# Pkg.add("RDatasets")
using RDatasets
iris = dataset("datasets", "iris")


INFO: Package RDatasets is already installed
INFO: METADATA is out-of-date — you may not have the latest version of RDatasets
INFO: Use `Pkg.update()` to get the latest versions of your packages
Out[24]:
SepalLengthSepalWidthPetalLengthPetalWidthSpecies
15.13.51.40.2setosa
24.93.01.40.2setosa
34.73.21.30.2setosa
44.63.11.50.2setosa
55.03.61.40.2setosa
65.43.91.70.4setosa
74.63.41.40.3setosa
85.03.41.50.2setosa
94.42.91.40.2setosa
104.93.11.50.1setosa
115.43.71.50.2setosa
124.83.41.60.2setosa
134.83.01.40.1setosa
144.33.01.10.1setosa
155.84.01.20.2setosa
165.74.41.50.4setosa
175.43.91.30.4setosa
185.13.51.40.3setosa
195.73.81.70.3setosa
205.13.81.50.3setosa
215.43.41.70.2setosa
225.13.71.50.4setosa
234.63.61.00.2setosa
245.13.31.70.5setosa
254.83.41.90.2setosa
265.03.01.60.2setosa
275.03.41.60.4setosa
285.23.51.50.2setosa
295.23.41.40.2setosa
304.73.21.60.2setosa
&vellip&vellip&vellip&vellip&vellip&vellip

Use Clustering.jl to cluster using the SepalLength, PetalLength, and PetalWidthfeatures via K-means clustering. Make a scatter plot of the resulting clusters.

Hint: You will need to index the dataframe, convert it to an array, and transpose it. In addition, you will need to use the assignments field of the return to get the cluster assignments.

Advanced Clustering Task

For the the example presented here, we will use a subhset of Word Embedding, trained using Word2Vec.jl. These are 100 dimentional vectors, which encode syntactic and semantic information about words.


In [ ]:
using Embeddings
countries = ["Afghanistan", "Algeria", "Angola", "Arabia", "Argentina", "Australia", "Bangladesh", "Brazil", "Britain", "Canada", "China", "Colombia", "Congo", "Egypt", "England", "Ethiopia", "France", "Germany", "Ghana", "India", "Indonesia", "Iran", "Iraq", "Ireland", "Italy", "Japan", "Kenya", "Korea", "Madagascar", "Malaysia", "Mexico", "Morocco", "Mozambique", "Myanmar", "Nepal", "Nigeria", "Pakistan", "Peru", "Philippines", "Poland", "Russia", "South", "Spain", "Sudan", "Tanzania", "Thailand", "Uganda", "Ukraine", "Usa", "Uzbekistan", "Venezuela", "Vietnam", "Wales", "Yemen"]
usa_cities = ["Albuquerque", "Atlanta", "Austin", "Baltimore", "Boston", "Charlotte", "Chicago", "Columbus", "Dallas", "Denver", "Detroit", "Francisco", "Fresno", "Houston", "Indianapolis", "Jacksonville", "Las", "Louisville", "Memphis", "Mesa", "Milwaukee", "Nashville", "Omaha", "Philadelphia", "Phoenix", "Portland", "Raleigh", "Sacramento", "San", "Seattle", "Tucson", "Vegas", "Washington"]
world_capitals = ["Accra", "Algiers", "Amman", "Ankara", "Antananarivo", "Athens", "Baghdad", "Baku", "Bangkok", "Beijing", "Beirut", "Berlin", "Bogotá", "Brasília", "Bucharest", "Budapest", "Cairo", "Caracas", "Damascus", "Dhaka", "Hanoi", "Havana", "Jakarta", "Kabul", "Kampala", "Khartoum", "Kinshasa", "Kyiv", "Lima", "London", "Luanda", "Madrid", "Manila", "Minsk", "Moscow", "Nairobi", "Paris", "Pretoria", "Pyongyang", "Quito", "Rabat", "Riyadh", "Rome", "Santiago", "Seoul", "Singapore", "Stockholm", "Taipei", "Tashkent", "Tehran", "Tokyo", "Vienna", "Warsaw", "Yaoundé"]
animals = ["alpaca","camel","cattle","dog","dove","duck","ferret","goldfish","goose","rat","llama","mouse","pigeon","yak"]
sports = ["archery","badminton","basketball","boxing","cycling","diving","equestrian","fencing","field","football","golf","gymnastics","handball","hockey","judo","kayak","pentathlon","polo","rowing","rugby","sailing","shooting","soccer","swimming","taekwondo","tennis","triathlon","volleyball","weightlifting","wrestling"]

words_by_class = [countries, usa_cities, world_capitals, animals, sports]
all_words = reduce(vcat, words_by_class)
embedding_table = load_embeddings(Word2Vec; keep_words = all_words) 
@assert Set(all_words) == Set(embedding_table.vocab)

embeddings = embedding_table.embeddings
all_words = embedding_table.vocab
classes = map(all_words) do word
    findfirst(col -> word  col, [countries, usa_cities, world_capitals, animals, sports])
end;

You can download the datased from here, and load it up with JLD as shown below. (or just load it directly if you have cloned the notebooks)

  • Use Affinity Propagraion from Clustering.jl, to cluster word2vec word embeddings, according to meaning.
  • Done right this will seperate locations from sports
  • Done finely and it will seperate ball-sports from other sports, and will seperate locations according to regions, etc

  • Affinity propagraion requires a similarity matrix, which you can set as a negated distance matrix.

  • For this you'll also want Distances.jl for all your distance metric needs.
  • It is traditional with word2vec to use cosine distance.
  • You will as also need to set each item's availability. This is the diagonal of the similarity matrix. Decreasing it roughly corresponds to decreasing the amount each node wants to be in a cluster on its own.