Iris Dataset PCA in Julia

using DataFrames;
using CSV;
using LinearAlgebra;
using Statistics;
using StatsPlots;

# Load iris data
filename = "/home/user/path/to/iris.data";
iris = CSV.read(filename, header=false)
iris = dropmissing(iris, disallowmissing=false);
header = [:sepal_length, :sepal_width, :petal_length, :petal_width, :species];
names!(iris, header);

function pca(X)
    # Standardize the data by subtracting the mean
    # and dividing by the standard deviation
    X_s = (X .- mean(X, dims=1)) ./ std(X, dims=1);
    # Get the number of datapoints in the dataset
    n, _ = size(X_s)
    # Compute the correlation matrix
    gram = transpose(X_s) * X_s / (n - 1);
    # NOTE: eigen-values computed by eigvals
    # are sorted in increasing order by default
    # so are the corresponding eigen-vectors
    lambda = eigvals(gram);
    vecs = eigvecs(gram);
    # Return the standardized data in the new
    # principal components (eigen) basis
    X_s * vecs
end

# Convert iris flower features into matrix form
# and perform Principal Component Analysis with
# the above pca function
features = header[1:4];
X = convert(Matrix, iris[:, features]);
irispca = pca(X)
irispca = DataFrame(irispca)
irispca[:species] = iris[:species]

# Plot datapoints on the plane formed by the first and second principal components
plt1 = @df irispca scatter(:x4, :x3, group=:species,
                title = "Iris Dataset PCA",
                xlabel = "Frist principal component (z-score)", ylabel = "Second principal component (z-score)",
                m=(0.7, [:cross :hex :star7], 5),
                bg=RGB(.2,.2,.2),
                );

# Plot datapoints on the plane formed by sepal length and sepal width
# NOTE: Data on this plot is not standardized
plt2 = @df iris scatter(:sepal_length, :sepal_width, group=:species,
               title = "Iris Dataset",
               xlabel = "Sepal length (cm)", ylabel = "Sepal width (cm)",
               m=(0.7, [:cross :hex :star7], 5),
               bg=RGB(.2,.2,.2),
               );

plot(plt1, plt2)