浅野直樹の学習日記

import matplotlib.pyplot as plt
import numpy as np
from sklearn import datasets
from sklearn.decomposition import PCA

#データの読み込み
digits = datasets.load_digits()

#PCAによって削減される特徴量数と失われるデータの分散との関係を表示
pca = PCA().fit(digits.data)
plt.plot(np.cumsum(pca.explained_variance_ratio_))
plt.xlabel('number of components')
plt.ylabel('cumulative explained variance')

#PCAで次元削減
pca = PCA(0.95).fit(digits.data)
reduced_data = pca.transform(digits.data)
restored_data = pca.inverse_transform(reduced_data)
print('pca.n_components_:', pca.n_components_)
print('digits.data.shape:', digits.data.shape)
print('reduced_data.shape:', reduced_data.shape)
print('restored_data.shape:', restored_data.shape)

####次元削減をしてから元に戻した手書き数字データの画像表示####
#描画領域の確保
fig, axes = plt.subplots(10, 10, figsize=(15, 15), subplot_kw={'xticks':[], 'yticks':[]}, gridspec_kw=dict(hspace=0.5, wspace=0.5))

#最初の100枚の確保した描画領域に表示
for i, ax in enumerate(axes.flat):
    ax.imshow(restored_data[i].reshape(8, 8), cmap=plt.cm.gray_r, interpolation='nearest')
    ax.set_title(digits.target[i])

pca.n_components_: 29
digits.data.shape: (1797, 64)
reduced_data.shape: (1797, 29)
restored_data.shape: (1797, 64)

from sklearn.model_selection import train_test_split
from sklearn import svm
from sklearn.pipeline import Pipeline
from sklearn.metrics import accuracy_score

#訓練データとテストデータに分割
X_train, X_test, y_train, y_test = train_test_split(digits.data, digits.target, shuffle=False)

#パイプラインの構築
steps = [('pca', PCA()), ('model', svm.SVC())]
pipe = Pipeline(steps)

#構築したパイプラインで学習
pipe.fit(X_train, y_train)

#学習結果のテスト
y_model = pipe.predict(X_test)
accuracy_score(y_test, y_model)

0.9622222222222222

from sklearn.model_selection import cross_val_score

#交差検証
scores = cross_val_score(pipe, digits.data, digits.target, cv=3)
print('scores:', scores)
print('average_score:', scores.mean())

scores: [0.96994992 0.98330551 0.96994992]
average_score: 0.9744017807456872

import warnings
import pandas as pd
from sklearn import datasets
from sklearn.utils import all_estimators
from sklearn.decomposition import PCA
from sklearn.pipeline import Pipeline
from sklearn.model_selection import cross_val_score

#警告の非表示
warnings.filterwarnings('ignore')

#結果を格納するデータフレームの作成
df = pd.DataFrame(columns=['model_name', 'score1', 'score2', 'score3', 'average_score'])

#データの読み込み
digits = datasets.load_digits()

#全てのモデルで学習
for (name, algorithm) in all_estimators(type_filter="classifier"):
    try:
        steps = [('pca', PCA()), ('model', algorithm())]
        pipe = Pipeline(steps)
        scores = cross_val_score(pipe, digits.data, digits.target, cv=3)
        df.loc[len(df)+1] = [name, *scores, scores.mean()]
    except:
        pass

#スコアのいい順に並べ替えて表示
sorted_df = df.sort_values('average_score', ascending=False)
sorted_df.set_axis(range(1, (len(df)+1) ), axis='index')

from sklearn.model_selection import train_test_split
from sklearn import svm
from sklearn.pipeline import Pipeline
from sklearn.model_selection import GridSearchCV
from sklearn.metrics import accuracy_score
import seaborn as sns

#訓練データとテストデータに分割
X_train, X_test, y_train, y_test = train_test_split(digits.data, digits.target, shuffle=False)

#パイプラインの構築
steps = [('pca', PCA()), ('model', svm.SVC())]
pipe = Pipeline(steps)

#グリッドサーチのパラメータの設定
param_grid = {
    'model__C': [0.1, 1, 10, 100, 1000],
    'model__gamma': [0.00001, 0.0001, 0.001, 0.01, 0.1]
}

#グリッドサーチの実行
grid_search = GridSearchCV(pipe, param_grid, cv=3)
grid_search.fit(X_train, y_train)

#結果の表示
print('cv_score:', grid_search.best_score_)
print('best parameters:', grid_search.best_params_)

#テストデータへの適用と結果の表示
y_best_model = grid_search.predict(X_test)
print('test_score:', accuracy_score(y_test, y_best_model))

#結果の可視化
cv_result = pd.DataFrame(grid_search.cv_results_)
cv_result = cv_result[['param_model__gamma', 'param_model__C', 'mean_test_score']]
cv_result_pivot = cv_result.pivot_table('mean_test_score', 'param_model__gamma', 'param_model__C')
heat_map = sns.heatmap(cv_result_pivot, cmap='viridis', annot=True)

cv_score: 0.96362286562732
best parameters: {'model__C': 10, 'model__gamma': 0.001}
test_score: 0.9688888888888889

import os
from skimage import io, color
from skimage.transform import resize
import numpy as np
import matplotlib.pyplot as plt

#オリジナルデータを格納するリストの作成
X_original = []
y_original = []

#ディレクトリから画像の読み込み
image_files = os.listdir('./original_images/')
for filename in image_files:
    image = io.imread('./original_images/' + filename)
    inverted_image = np.invert(image)
    gray_image = color.rgb2gray(inverted_image)
    resized_image = resize(gray_image, (8, 8))
    scaled_image = resized_image * 16
    X_original.append(scaled_image)
    y_original.append(int(filename[0]))

#画像データを機械学習に適した形にする
X_original_data = np.array(X_original).reshape(10, -1)
y_original_target = np.array(y_original)

#以前に作成したモデルでオリジナルデータの予測
y_original_model = grid_search.predict(X_original_data)

#描画領域の確保
fig, axes = plt.subplots(1, 10, figsize=(15, 15), subplot_kw={'xticks':[], 'yticks':[]}, gridspec_kw=dict(hspace=0.5, wspace=0.5))

#オリジナルデータを「予測値→真の値」というラベルとともに表示
for i, ax in enumerate(axes.flat):
    ax.imshow(X_original_data[i].reshape(8, 8), cmap=plt.cm.gray_r, interpolation='nearest')
    label = str(y_original_model[i]) + '→' + str(y_original_target[i])
    title_color = "black" if y_original_model[i] == y_original_target[i] else "red"
    ax.set_title(label, color=title_color)