'''
79. 適合率-再現率グラフの描画
ロジスティック回帰モデルの分類の閾値を変化させることで，適合率-再現率グラフを描画せよ．
'''
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
from sklearn.metrics import precision_recall_curve
from scipy.stats import hmean
from knock72 import load
model = load("model")
labels = load("labels")
features = load("features")
model.classes_ # => array([-1, 1])
probs = model.predict_proba(features)[:, 1]
pre, rec, th = precision_recall_curve(labels, probs) # PR曲線
plt.plot(pre, rec)
plt.xlabel("precision")
plt.ylabel("recall")
plt.savefig("out79_2d.png")
x1, x2, y = zip(*[(p, r, hmean([p, r])) for p, r in zip(pre, rec) if p and r])
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
ax.scatter3D(x1, x2, y)
ax.set_xlabel("precision")
ax.set_ylabel("recall")
ax.set_zlabel("f1")
plt.savefig("out79_3d.png")
'''
74. 予測
73で学習したロジスティック回帰モデルを用い，
与えられた文の極性ラベル（正例なら"+1"，負例なら"-1"）と，
その予測確率を計算するプログラムを実装せよ．
'''
from sklearn.feature_extraction.text import TfidfVectorizer
from knock72 import load, extract_features
model = load("model")
vocab = load("vocabs")
_, docs = extract_features("./test.txt")
vectorizer = TfidfVectorizer(vocabulary=vocab) # 疑惑のコード
features = vectorizer.fit_transform(docs).toarray() # あとで見直す
pp = zip(model.predict(features), model.predict_proba(features))
for predict, proba in pp:
print(f"{predict:+d} : {max(proba):f}")
'''
* sklearn.linear_model.LogisticRegression
https://scikit-learn.org/stable/modules/generated/sklearn.linear_model.LogisticRegression.html
- predict(self, X)
[Methods] Predict class labels for samples in X.
- predict_proba(self, X)
[Methods] Probability estimates.
'''
'''
73. 学習
72で抽出した素性を用いて，ロジスティック回帰モデルを学習せよ．
'''
from sklearn.linear_model import LogisticRegression
from knock72 import load, save
labels = load("labels")
features = load("features")
model = LogisticRegression().fit(features, labels)
save("model", model)