我已经构建了一个小程序，可以使用scikit-learn为给定数据集创建分类器。现在我想尝试this example，看看分类器在工作。例如，clf必须检测“猫”。
这是我继续下去的方式：
我有50张猫的照片和50张“无猫”的照片。
data_set的描述符training_set training_set和test_set的直方图数据从scikit-learn：
尝试此代码tuned_parameters = [{'kernel': ['rbf'], 'gamma': [1e-3, 1e-4],
'C': [1, 10, 100, 1000]},
{'kernel': ['linear'], 'C': [1, 10, 100, 1000]}]
scores = ['precision', 'recall']
for score in scores:
print("# Tuning hyper-parameters for %s" % score)
print()
clf = GridSearchCV(SVC(C=1), tuned_parameters, cv=5, scoring=score)
clf.fit(X_train, y_train)
print("Best parameters set found on development set:")
print()
print(clf.best_estimator_)
print()
print("Grid scores on development set:")
print()
for params, mean_score, scores in clf.grid_scores_:
print("%0.3f (+/-%0.03f) for %r"
% (mean_score, scores.std() / 2, params))
print()
print("Detailed classification report:")
print()
print("The model is trained on the full development set.")
print("The scores are computed on the full evaluation set.")
print()
y_true, y_pred = y_test, clf.predict(X_test)
print y_true
print y_pred
print(classification_report(y_true, y_pred))
print()
print clf.score(X_train, y_train)
print "score"
print clf.best_params_
print "best_params"
pred = clf.predict(X_test)
print accuracy_score(y_test, pred)
print "accuracy_score"
我得到了那个结果：
# Tuning hyper-parameters for recall
()
/usr/local/lib/python2.7/dist-packages/sklearn/metrics/metrics.py:1760: UserWarning: The sum of true positives and false positives are equal to zero for some labels. Precision is ill defined for those labels [ 0.]. The precision and recall are equal to zero for some labels. fbeta_score is ill defined for those labels [ 0.].
average=average)
/usr/local/lib/python2.7/dist-packages/sklearn/metrics/metrics.py:1760: UserWarning: The sum of true positives and false positives are equal to zero for some labels. Precision is ill defined for those labels [ 1.]. The precision and recall are equal to zero for some labels. fbeta_score is ill defined for those labels [ 1.].
average=average)
Best parameters set found on development set:
()
SVC(C=0.001, cache_size=200, class_weight=None, coef0=0.0, degree=3,
gamma=0.001, kernel=rbf, max_iter=-1, probability=False,
random_state=None, shrinking=True, tol=0.001, verbose=False)
()
Grid scores on development set:
()
0.800 (+/-0.200) for {'kernel': 'rbf', 'C': 0.001, 'gamma': 0.001}
0.800 (+/-0.200) for {'kernel': 'rbf', 'C': 0.001, 'gamma': 0.0001}
0.800 (+/-0.200) for {'kernel': 'rbf', 'C': 0.01, 'gamma': 0.001}
0.800 (+/-0.200) for {'kernel': 'rbf', 'C': 0.01, 'gamma': 0.0001}
0.800 (+/-0.200) for {'kernel': 'rbf', 'C': 0.10000000000000001, 'gamma': 0.001}
0.800 (+/-0.200) for {'kernel': 'rbf', 'C': 0.10000000000000001, 'gamma': 0.0001}
0.800 (+/-0.200) for {'kernel': 'rbf', 'C': 1.0, 'gamma': 0.001}
0.800 (+/-0.200) for {'kernel': 'rbf', 'C': 1.0, 'gamma': 0.0001}
0.800 (+/-0.200) for {'kernel': 'rbf', 'C': 10.0, 'gamma': 0.001}
0.800 (+/-0.200) for {'kernel': 'rbf', 'C': 10.0, 'gamma': 0.0001}
0.800 (+/-0.200) for {'kernel': 'rbf', 'C': 100.0, 'gamma': 0.001}
0.800 (+/-0.200) for {'kernel': 'rbf', 'C': 100.0, 'gamma': 0.0001}
0.800 (+/-0.200) for {'kernel': 'rbf', 'C': 1000.0, 'gamma': 0.001}
0.800 (+/-0.200) for {'kernel': 'rbf', 'C': 1000.0, 'gamma': 0.0001}
()
Detailed classification report:
()
The model is trained on the full development set.
The scores are computed on the full evaluation set.
()
[ 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
1. 1. 1. 1. 1. 1. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
1. 1. 1. 1. 1. 1. 1. 1. 0. 1. 1. 1. 1. 1.]
precision recall f1-score support
0.0 1.00 0.04 0.08 25
1.0 0.51 1.00 0.68 25
avg / total 0.76 0.52 0.38 50
()
0.52
score
{'kernel': 'rbf', 'C': 0.001, 'gamma': 0.001}
best_params
0.52
accuracy_score
似乎是clf说所有人都认为它是猫......但为什么呢？
data_set是否小到可以获得好结果？
编辑：我正在使用VLFeat来检测筛选描述符
功能：
def create_descriptor_data(data, ID):
descriptor_list = []
datas = numpy.genfromtxt(data,dtype='str')
for p in datas:
locs, desc = vlfeat_module.vlf_create_descriptors(p,str(ID)+'.key',ID) # create descriptors and save descs in file
if len(desc) > 500:
desc = desc[::round((len(desc))/400, 1)] # take between 400 - 800 descriptors
descriptor_list.append(desc)
ID += 1 # ID for filename
return descriptor_list
# create k-mean centers from all *.txt files in directory (data)
def create_center_data(data):
#data = numpy.vstack(data)
n_clusters = len(numpy.unique(data))
kmeans = KMeans(init='k-means++', n_clusters=n_clusters, n_init=1)
kmeans.fit(data)
return kmeans, n_clusters
def create_histogram_data(kmeans, descs, n_clusters):
histogram_list = []
# load from each file data
for desc in descs:
length = len(desc)
# create histogram from descriptors
histogram = kmeans.predict(desc)
histogram = numpy.bincount(histogram, minlength=n_clusters) #minlength = k in k-means
histogram = numpy.divide(histogram, length, dtype='float')
histogram_list.append(histogram)
histogram = numpy.vstack(histogram_list)
return histogram
和电话：
X_desc_pos = lib.dataset_module.create_descriptor_data("./static/picture_set/dataset_pos.txt",0) # create desc from dataset_pos, 25 pics
X_desc_neg = lib.dataset_module.create_descriptor_data("./static/picture_set/dataset_neg.txt",51) # create desc from dataset_neg, 25 pics
X_train_pos, X_test_pos = train_test_split(X_desc_pos, test_size=0.5)
X_train_neg, X_test_neg = train_test_split(X_desc_neg, test_size=0.5)
x1 = numpy.vstack(X_train_pos)
x2 = numpy.vstack(X_train_neg)
kmeans, n_clusters = lib.dataset_module.create_center_data(numpy.vstack((x1,x2)))
X_train_pos = lib.dataset_module.create_histogram_data(kmeans, X_train_pos, n_clusters)
X_train_neg = lib.dataset_module.create_histogram_data(kmeans, X_train_neg, n_clusters)
X_train = numpy.vstack([X_train_pos, X_train_neg])
y_train = numpy.hstack([numpy.ones(len(X_train_pos)), numpy.zeros(len(X_train_neg))])
X_test_pos = lib.dataset_module.create_histogram_data(kmeans, X_test_pos, n_clusters)
X_test_neg = lib.dataset_module.create_histogram_data(kmeans, X_test_neg, n_clusters)
X_test = numpy.vstack([X_test_pos, X_test_neg])
y_test = numpy.hstack([numpy.ones(len(X_test_pos)), numpy.zeros(len(X_test_neg))])
tuned_parameters = [{'kernel': ['rbf'], 'gamma': [1e-3, 1e-4],
'C': [1, 10, 100, 1000]},
{'kernel': ['linear'], 'C': [1, 10, 100, 1000]}]
scores = ['precision', 'recall']
for score in scores:
print("# Tuning hyper-parameters for %s" % score)
print()
clf = GridSearchCV(SVC(C=1), tuned_parameters, cv=5, scoring=score)
clf.fit(X_train, y_train)
print("Best parameters set found on development set:")
print()
print(clf.best_estimator_)
print()
print("Grid scores on development set:")
print()
for params, mean_score, scores in clf.grid_scores_:
print("%0.3f (+/-%0.03f) for %r"
% (mean_score, scores.std() / 2, params))
print()
print("Detailed classification report:")
print()
print("The model is trained on the full development set.")
print("The scores are computed on the full evaluation set.")
print()
y_true, y_pred = y_test, clf.predict(X_test)
print y_true
print y_pred
print(classification_report(y_true, y_pred))
print()
print clf.score(X_train, y_train)
print "score"
print clf.best_params_
print "best_params"
pred = clf.predict(X_test)
print accuracy_score(y_test, pred)
print "accuracy_score"
编辑：通过更新范围和savae进行一些更改“准确度”
# Tuning hyper-parameters for accuracy
()
Best parameters set found on development set:
()
SVC(C=1000.0, cache_size=200, class_weight=None, coef0=0.0, degree=3,
gamma=1.0, kernel=rbf, max_iter=-1, probability=False, random_state=None,
shrinking=True, tol=0.001, verbose=False)
()
Grid scores on development set:
()
...
()
Detailed classification report:
()
The model is trained on the full development set.
The scores are computed on the full evaluation set.
()
[ 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
1. 1. 1. 1. 1. 1. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 0. 1. 0. 1. 1. 1.
1. 1. 1. 0. 1. 1. 1. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0.
0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0.]
precision recall f1-score support
0.0 0.88 0.92 0.90 25
1.0 0.92 0.88 0.90 25
avg / total 0.90 0.90 0.90 50
()
1.0
score
{'kernel': 'rbf', 'C': 1000.0, 'gamma': 1.0}
best_params
0.9
accuracy_score
但是在带有
的图片上进行测试rslt = clf.predict(test_histogram)
答案 0 :(得分：3)
似乎是clf说所有人都认为它是猫......但为什么呢？
从您的粘贴输出中有点难以辨别，但似乎这是scores = ['precision', 'recall']循环的第二次迭代，因此您需要进行优化以进行调用。这与分类报告一致，该报告指出回忆是正面课程的1.00（完美）。
什么时候回忆完美？好吧，当没有假阴性时，没有猫没有被发现。因此，获得完美召回的简便方法就是预测“猫”和“猫”。对于每个输入图片，无论它是否是一只猫，GridSearchCV都找到了一个完全相同的分类器。
当您优化精确度时，可能会发生类似的事情：永远不会预测“猫”和“猫”，可以实现完美的精确度。因为你没有误报。
要避免这种情况，请优化准确度而不是精确度或召回率，如果遇到类别不平衡的情况，则优化Fᵦ。
答案 1 :(得分：2)
这种行为有很多可能性：
C和gamma参数范围可能过于狭窄 - 此变量高度依赖于数据，您的表示形式的值可能需要完全不同的C和{{1}然后是当前使用的[under / over fitting] 我的个人猜测（因为没有数据很难重现问题）这里是第三个选项 - 错误的gamma和C参数来找到一个好的模型。
修改
您应该尝试多更大范围的值，例如
gamma和C 之间10^-5
在10^15和gamma 之间 10^-14
10^2<强> EDIT2
一旦参数'范围得到纠正，现在您应该执行实际的“案例研究”。收集更多图像，分析你的数据表示（直方图真的足以完成这项任务吗？），处理你的数据（它已经规范化了吗？也许尝试一些去相关？），考虑使用更简单的内核 - rbf可能非常具有欺骗性 - 一方面它在培训期间可以获得很高分，但另一方面 - 在测试期间完全失败。这是其过度拟合功能的结果（对于任何一致的数据集，RBF-SVM在训练期间可以达到100％得分），因此在模型的功能和泛化能力之间找到平衡是一个难题。这是实际的“机器学习之旅”开始，玩得开心！