Skip to content

Instantly share code, notes, and snippets.

@bellbind bellbind/data.csv
Last active Jun 21, 2018

Embed
What would you like to do?
[python3][pandas][numpy][sklearn]PCA(Principal Component Analysis) with python
ブランド名 プロっぽい カッコいい 好き かわいい 一般的 ミーハー 初心者 ダサイ 嫌い わからない 欲しい
ロシニョール 33 15 38 11 71 94 7 4 8 5 37
ヤマハ 10 6 24 18 75 46 29 5 5 12 10
オガサカ 67 5 18 1 32 4 12 25 24 11 14
カザマ 25 5 12 2 33 4 18 40 23 31 8
アトミック 52 32 24 5 32 32 3 11 5 17 35
ニシザワ 23 7 12 1 8 1 3 32 15 28 11
クナイスル 32 13 12 0 37 2 6 23 16 33 9
K2 48 26 24 3 20 32 2 10 21 20 23
フィッシャー 27 14 18 2 20 7 3 9 12 37 11
ブリザード 41 25 20 4 24 8 2 12 12 28 17
ミズノ 5 3 1 0 48 6 27 74 27 30 3
オーリン 19 17 22 13 22 37 8 17 10 21 16
スワロー 5 1 0 2 18 1 43 69 26 27 2
ケスレー 47 19 12 1 17 2 2 14 11 44 14
エラン 36 8 13 2 7 11 0 10 12 36 8
フォルクル 32 7 10 0 1 4 0 11 8 28 8
# PCA(principal component analysis) with numpy
import numpy as np
import matplotlib as mpl
import matplotlib.pyplot as plt
import pandas
# data from book ISBN:978-4-480-09861-0
frame = pandas.read_csv("./data.csv", index_col=0)
ids = frame.index
props = frame.columns
M = frame.values
n = len(frame) # number of rows
X = M - M.mean(axis=0)
# Cxx: variance-covariance matrix
Cxx = (X.T @ X) / n
# Or, simply use with np.cov() as:
#Cxx = np.cov(M, rowvar=False, bias=True)
# NOTE: result eigval/vecs is not orderd
eigvals, eigvecs = np.linalg.eig(Cxx)
eigidx = eigvals.argsort()[::-1] # descending order index
eigvals, eigvecs = eigvals[eigidx], eigvecs[:, eigidx]
rates = eigvals / sum(eigvals)
print(f"2 explained variance rate: {sum(rates[:2])}")
# component vectors (NOTE: nagate for plot layout in the former book)
v1 = -eigvecs[:, 0]
v2 = -eigvecs[:, 1]
# component scores
f1 = X @ v1
f2 = X @ v2
# plot
mpl.rc("font", family="Noto Sans CJK JP")
fig, (s1, s2) = plt.subplots(1, 2, figsize=(16, 8))
def plot(s, xs, ys, labels):
s.axhline(color="gray")
s.axvline(color="gray")
s.scatter(xs, ys)
for label, x, y in zip(labels, xs, ys):
s.annotate(label, (x, y))
pass
pass
plot(s1, v1, v2, props)
plot(s2, f1, f2, ids)
fig.savefig("pca-numpy-result.png")
plt.show()
# PCA(principal component analysis) with sklearn
import sklearn.decomposition # pip install scipy sklearn
import matplotlib as mpl
import matplotlib.pyplot as plt
import pandas
# data from book ISBN:978-4-480-09861-0
frame = pandas.read_csv("./data.csv", index_col=0)
ids = frame.index
props = frame.columns
M = frame.values
n = len(frame) # number of rows
pca = sklearn.decomposition.PCA(n_components=2)
pca.fit(M)
# component vectors
v1, v2 = pca.components_
# component scores
trans = pca.transform(M)
f1 = trans[:, 0]
f2 = trans[:, 1]
# plot
mpl.rc("font", family="Noto Sans CJK JP")
fig, (s1, s2) = plt.subplots(1, 2, figsize=(16, 8))
def plot(s, xs, ys, labels):
s.axhline(color="gray")
s.axvline(color="gray")
s.scatter(xs, ys)
for label, x, y in zip(labels, xs, ys):
s.annotate(label, (x, y))
pass
pass
plot(s1, v1, v2, props)
plot(s2, f1, f2, ids)
fig.savefig("pca-sklearn-result.png")
plt.show()
@bellbind

This comment has been minimized.

Copy link
Owner Author

bellbind commented Jun 13, 2018

The values in data.csv are referred from: https://www.amazon.co.jp/dp/4480098615


Note on Font

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
You can’t perform that action at this time.