'''
75. 素性の重み
73で学習したロジスティック回帰モデルの中で，
重みの高い素性トップ10と，重みの低い素性トップ10を確認せよ．
'''
from sklearn.linear_model import LogisticRegression
from no73 import load_feats
import sys
feats, labels, sents = load_feats(sys.argv[1])
model = LogisticRegression(C=1e5)
model.fit(feats, labels)
with open('data/feats.csv') as f:
header = f.readline()
header = header.split(",")[2:]
feats_dict = {}
for word, weight in zip(header, model.coef_[0]):
feats_dict[word] = weight
for i, elem in enumerate(sorted(feats_dict.items(), key=lambda x: x[1])):
if i > 9:
break
print("{}: {}".format(elem[0], elem[1]))
for i, elem in enumerate(sorted(feats_dict.items(), key=lambda x: -x[1])):
if i > 9:
break
print("{}: {}".format(elem[0], elem[1]))
predicts.append(1)
else:
predicts.append(-1)
return predicts
if __name__ == '__main__':
from no73 import load_feats
from no77 import get_metrics
from no78 import split_seq, get_train
from sklearn.linear_model import LogisticRegression
import matplotlib.pyplot as plt
import sys
fname = sys.argv[1]
K = 5
feats, labels, _ = load_feats(fname)
split_feats = split_seq(feats, K)
split_labels = split_seq(labels, K)
probs_list = []
for i in range(K):
model = LogisticRegression(C=1e5)
model.fit(get_train(split_feats, i), get_train(split_labels, i))
probs = model.predict_proba(split_feats[i])
probs_list.append(probs)
threshold_list = []
precision_list = []
recall_list = []
for threshold in range(101):
threshold /= 100
metrics_list = []