首頁 > 軟體

Python利用Seaborn繪製多標籤的混淆矩陣

2022-07-15 10:00:03

Seaborn - 繪製多標籤的混淆矩陣、召回、精準、F1

匯入seabornmatplotlibscipysklearn等包:

import seaborn as sns
from matplotlib import pyplot as plt
from scipy.special import softmax
from sklearn.metrics import accuracy_score, confusion_matrix, precision_score, recall_score, f1_score

sns.set_theme(color_codes=True)

從dataframe中,獲取y_true(真實標籤)和y_pred(預測標籤):

y_true = df["target"]
y_pred = df['prediction']

計算驗證資料整體的準確率acc、精準率precision、召回率recall、F1,使用加權模式average=‘weighted’:

# 準確率acc,精準precision,召回recall,F1
acc = accuracy_score(df["target"], df['prediction'])
precision = precision_score(y_true, y_pred, average='weighted')
recall = recall_score(y_true, y_pred, average='weighted')
f1 = f1_score(y_true, y_pred, average='weighted')
print(f'[Info] acc: {acc}, precision: {precision}, recall: {recall}, f1: {f1}')

計算混淆矩陣:

# 橫座標是真實類別數,縱座標是預測類別數
cf_matrix = confusion_matrix(y_true, y_pred)

5類矩陣的繪製方案,混淆矩陣、百分比的混淆矩陣、召回矩陣、精準矩陣、F1矩陣:

  • 混淆矩陣是計數,百分比的混淆矩陣是佔比
  • 召回矩陣是,每行的和是1,每行代表真實類別數,佔比就是召回
  • 精準矩陣是,每列的和是1,每列代表預測列表數,佔比就是精準
  • F1矩陣是按照 2PR/(P+R),注意為0的情況,需要補0,使用np.divide(a, b, out=np.zeros_like(a), where=(b != 0))

程式碼如下:

# 橫座標是真實類別數,縱座標是預測類別數
cf_matrix = confusion_matrix(y_true, y_pred)

figure, axes = plt.subplots(2, 2, figsize=(16*1.25, 16))

# 混淆矩陣
ax = sns.heatmap(cf_matrix, annot=True, fmt='g', ax=axes[0][0], cmap='Blues')
ax.title.set_text("Confusion Matrix")
ax.set_xlabel("y_pred")
ax.set_ylabel("y_true")
# plt.savefig(csv_path.replace(".csv", "_cf_matrix.png"))
# plt.show()

# 混淆矩陣 - 百分比
cf_matrix = confusion_matrix(y_true, y_pred)
ax = sns.heatmap(cf_matrix / np.sum(cf_matrix), annot=True, ax=axes[0][1], fmt='.2%', cmap='Blues')
ax.title.set_text("Confusion Matrix (percent)")
ax.set_xlabel("y_pred")
ax.set_ylabel("y_true")
# plt.savefig(csv_path.replace(".csv", "_cf_matrix_p.png"))
# plt.show()

# 召回矩陣,行和為1
sum_true = np.expand_dims(np.sum(cf_matrix, axis=1), axis=1)
precision_matrix = cf_matrix / sum_true
ax = sns.heatmap(precision_matrix, annot=True, fmt='.2%', ax=axes[1][0], cmap='Blues')
ax.title.set_text("Precision Matrix")
ax.set_xlabel("y_pred")
ax.set_ylabel("y_true")
# plt.savefig(csv_path.replace(".csv", "_recall.png"))
# plt.show()

# 精準矩陣,列和為1
sum_pred = np.expand_dims(np.sum(cf_matrix, axis=0), axis=0)
recall_matrix = cf_matrix / sum_pred
ax = sns.heatmap(recall_matrix, annot=True, fmt='.2%', ax=axes[1][1], cmap='Blues')
ax.title.set_text("Recall Matrix")
ax.set_xlabel("y_pred")
ax.set_ylabel("y_true")
# plt.savefig(csv_path.replace(".csv", "_precision.png"))
# plt.show()

# 繪製4張圖
plt.autoscale(enable=False)
plt.savefig(csv_path.replace(".csv", "_all.png"), bbox_inches='tight', pad_inches=0.2)
plt.show()

# F1矩陣
a = 2 * precision_matrix * recall_matrix
b = precision_matrix + recall_matrix
f1_matrix = np.divide(a, b, out=np.zeros_like(a), where=(b != 0))
ax = sns.heatmap(f1_matrix, annot=True, fmt='.2%', cmap='Blues')
ax.title.set_text("F1 Matrix")
ax.set_xlabel("y_pred")
ax.set_ylabel("y_true")
plt.savefig(csv_path.replace(".csv", "_f1.png"))
plt.show()

輸出混淆矩陣、混淆矩陣(百分比)、召回矩陣、精準矩陣:

F1 Score:

到此這篇關於Python利用Seaborn繪製多標籤的混淆矩陣的文章就介紹到這了,更多相關Python Seaborn混淆矩陣內容請搜尋it145.com以前的文章或繼續瀏覽下面的相關文章希望大家以後多多支援it145.com!


IT145.com E-mail:sddin#qq.com