数组维度大小为 3 时的混淆矩阵错误
Confusion matrix error when array dimensions are of size 3
此代码：
from pandas_ml import ConfusionMatrix
y_actu = [1,2]
y_pred = [1,2]
cm = ConfusionMatrix(y_actu, y_pred)
cm.print_stats()
打印：
population: 2
P: 1
N: 1
PositiveTest: 1
NegativeTest: 1
TP: 1
TN: 1
FP: 0
FN: 0
TPR: 1.0
TNR: 1.0
PPV: 1.0
NPV: 1.0
FPR: 0.0
FDR: 0.0
FNR: 0.0
ACC: 1.0
F1_score: 1.0
MCC: 1.0
informedness: 1.0
markedness: 1.0
prevalence: 0.5
LRP: inf
LRN: 0.0
DOR: inf
FOR: 0.0
/opt/conda/lib/python3.5/site-packages/pandas_ml/confusion_matrix/bcm.py:332: RuntimeWarning: divide by zero encountered in double_scalars
return(np.float64(self.TPR) / self.FPR)
这是预期的。
但是，当我修改代码为：
from pandas_ml import ConfusionMatrix
y_actu = [1,2,3]
y_pred = [1,2,3]
cm = ConfusionMatrix(y_actu, y_pred)
cm.print_stats()
使用：
y_actu = [1,2,3]
y_pred = [1,2,3]
它导致此错误：
OrderedDict([('Accuracy', 1.0), ('95% CI', (0.29240177382128668, nan)), ('No Information Rate', 'ToDo'), ('P-Value [Acc > NIR]', 0.29629629629629622), ('Kappa', 1.0), ("Mcnemar's Test P-Value", 'ToDo')])
ValueErrorTraceback (most recent call last)
<ipython-input-30-d8c5dc2bea73> in <module>()
3 y_pred = [1,2,3]
4 cm = ConfusionMatrix(y_actu, y_pred)
----> 5 cm.print_stats()
/opt/conda/lib/python3.5/site-packages/pandas_ml/confusion_matrix/abstract.py in print_stats(self, lst_stats)
446 Prints statistics
447 """
--> 448 print(self._str_stats(lst_stats))
449
450 def get(self, actual=None, predicted=None):
/opt/conda/lib/python3.5/site-packages/pandas_ml/confusion_matrix/abstract.py in _str_stats(self, lst_stats)
427 }
428
--> 429 stats = self.stats(lst_stats)
430
431 d_stats_str = collections.OrderedDict([
/opt/conda/lib/python3.5/site-packages/pandas_ml/confusion_matrix/abstract.py in stats(self, lst_stats)
390 d_stats = collections.OrderedDict()
391 d_stats['cm'] = self
--> 392 d_stats['overall'] = self.stats_overall
393 d_stats['class'] = self.stats_class
394 return(d_stats)
/opt/conda/lib/python3.5/site-packages/pandas_ml/confusion_matrix/cm.py in __getattr__(self, attr)
33 Returns (weighted) average statistics
34 """
---> 35 return(self._avg_stat(attr))
/opt/conda/lib/python3.5/site-packages/pandas_ml/confusion_matrix/abstract.py in _avg_stat(self, stat)
509 v = getattr(binary_cm, stat)
510 print(v)
--> 511 s_values[cls] = v
512 value = (s_values * self.true).sum() / self.population
513 return(value)
/opt/conda/lib/python3.5/site-packages/pandas/core/series.py in __setitem__(self, key, value)
771 # do the setitem
772 cacher_needs_updating = self._check_is_chained_assignment_possible()
--> 773 setitem(key, value)
774 if cacher_needs_updating:
775 self._maybe_update_cacher()
/opt/conda/lib/python3.5/site-packages/pandas/core/series.py in setitem(key, value)
767 pass
768
--> 769 self._set_with(key, value)
770
771 # do the setitem
/opt/conda/lib/python3.5/site-packages/pandas/core/series.py in _set_with(self, key, value)
809 if key_type == 'integer':
810 if self.index.inferred_type == 'integer':
--> 811 self._set_labels(key, value)
812 else:
813 return self._set_values(key, value)
/opt/conda/lib/python3.5/site-packages/pandas/core/series.py in _set_labels(self, key, value)
826 if mask.any():
827 raise ValueError('%s not contained in the index' % str(key[mask]))
--> 828 self._set_values(indexer, value)
829
830 def _set_values(self, key, value):
/opt/conda/lib/python3.5/site-packages/pandas/core/series.py in _set_values(self, key, value)
831 if isinstance(key, Series):
832 key = key._values
--> 833 self._data = self._data.setitem(indexer=key, value=value)
834 self._maybe_update_cacher()
835
/opt/conda/lib/python3.5/site-packages/pandas/core/internals.py in setitem(self, **kwargs)
3166
3167 def setitem(self, **kwargs):
-> 3168 return self.apply('setitem', **kwargs)
3169
3170 def putmask(self, **kwargs):
/opt/conda/lib/python3.5/site-packages/pandas/core/internals.py in apply(self, f, axes, filter, do_integrity_check, consolidate, **kwargs)
3054
3055 kwargs['mgr'] = self
-> 3056 applied = getattr(b, f)(**kwargs)
3057 result_blocks = _extend_blocks(applied, result_blocks)
3058
/opt/conda/lib/python3.5/site-packages/pandas/core/internals.py in setitem(self, indexer, value, mgr)
685 indexer.dtype == np.bool_ and
686 len(indexer[indexer]) == len(value)):
--> 687 raise ValueError("cannot set using a list-like indexer "
688 "with a different length than the value")
689
ValueError: cannot set using a list-like indexer with a different length than the value
我发现了类似的 question，它指出
Using endemic lists is not allowed on assignment and is not recommended to do this at all.
什么是特有列表，我创建了一个吗？
我建议使用 scikit-learn 中的 confusion_matrix。您提到的其他指标，如 Precision、Recall、F1-score 也可从 sklearn.metrics 获得。
>>> from sklearn.metrics import confusion_matrix
>>> y_actu = [1,2,3]
>>> y_pred = [1,2,3]
>>> confusion_matrix(y_actu, y_pred)
array([[1, 0, 0],
[0, 1, 0],
[0, 0, 1]])
有趣的是，当我 运行 你的代码时，我没有得到你收到的错误，并且代码 运行 完美。我建议您通过 运行ning 升级 pandas_ml 库：
pip install --upgrade pandas_ml
此外，您需要通过 运行ning 升级 pandas：
pip install --upgrade pandas
如果这不起作用，您可以使用 pandas 本身创建一个混淆矩阵：
import pandas as pd
y_actu = pd.Series([1, 2, 3], name='Actual')
y_pred = pd.Series([1, 2, 3], name='Predicted')
df_confusion = pd.crosstab(y_actu, y_pred)
print df_confusion
这将为您提供您正在寻找的 table。
似乎错误不是因为数组维度：
from pandas_ml import ConfusionMatrix
y_actu = [1,2,2]
y_pred = [1,1,2]
cm = ConfusionMatrix(y_actu, y_pred)
cm.print_stats()
这个（二元分类问题）工作正常。
也许多类分类问题的混淆矩阵刚刚被打破。
更新：
我只是做了这些步骤：
conda update pandas
得到pandas0.20.1
然后
pip install -U pandas_ml
现在多类混淆矩阵一切正常：
from pandas_ml import ConfusionMatrix
y_actu = [1,2,3]
y_pred = [1,2,3]
cm = ConfusionMatrix(y_actu, y_pred)
cm.print_stats()
我得到了输出：
Class Statistics:
Classes 1 2 3
Population 3 3 3
P: Condition positive 1 1 1
N: Condition negative 2 2 2
Test outcome positive 1 1 1
Test outcome negative 2 2 2
TP: True Positive 1 1 1
TN: True Negative 2 2 2
FP: False Positive 0 0 0
FN: False Negative 0 0 0
TPR: (Sensitivity, hit rate, recall) 1 1 1
TNR=SPC: (Specificity) 1 1 1
PPV: Pos Pred Value (Precision) 1 1 1
NPV: Neg Pred Value 1 1 1
FPR: False-out 0 0 0
FDR: False Discovery Rate 0 0 0
FNR: Miss Rate 0 0 0
ACC: Accuracy 1 1 1
F1 score 1 1 1
MCC: Matthews correlation coefficient 1 1 1
Informedness 1 1 1
Markedness 1 1 1
Prevalence 0.333333 0.333333 0.333333
LR+: Positive likelihood ratio inf inf inf
LR-: Negative likelihood ratio 0 0 0
DOR: Diagnostic odds ratio inf inf inf
FOR: False omission rate 0 0 0
我也使用并推荐 sklearn confusion_matrix 函数。就我个人而言，我还保留了一个 "pretty-print confusion matrix" 函数，它有一些额外的便利：
- class 沿混淆矩阵轴打印的标签
- 标准化混淆矩阵统计数据，使所有单元格总和为 1
- 根据单元格值缩放的混淆矩阵单元格颜色
- F 分数等其他指标印在混淆矩阵下方。
像这样：
这是绘图函数，主要基于 this example from the Scikit-Learn documentation
import matplotlib.pyplot as plt
import itertools
from sklearn.metrics import classification_report
def pretty_print_conf_matrix(y_true, y_pred,
classes,
normalize=False,
title='Confusion matrix',
cmap=plt.cm.Blues):
"""
Mostly stolen from: http://scikit-learn.org/stable/auto_examples/model_selection/plot_confusion_matrix.html#sphx-glr-auto-examples-model-selection-plot-confusion-matrix-py
Normalization changed, classification_report stats added below plot
"""
cm = confusion_matrix(y_true, y_pred)
# Configure Confusion Matrix Plot Aesthetics (no text yet)
plt.imshow(cm, interpolation='nearest', cmap=cmap)
plt.title(title, fontsize=14)
tick_marks = np.arange(len(classes))
plt.xticks(tick_marks, classes, rotation=45)
plt.yticks(tick_marks, classes)
plt.ylabel('True label', fontsize=12)
plt.xlabel('Predicted label', fontsize=12)
# Calculate normalized values (so all cells sum to 1) if desired
if normalize:
cm = np.round(cm.astype('float') / cm.sum(),2) #(axis=1)[:, np.newaxis]
# Place Numbers as Text on Confusion Matrix Plot
thresh = cm.max() / 2.
for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):
plt.text(j, i, cm[i, j],
horizontalalignment="center",
color="white" if cm[i, j] > thresh else "black",
fontsize=12)
# Add Precision, Recall, F-1 Score as Captions Below Plot
rpt = classification_report(y_true, y_pred)
rpt = rpt.replace('avg / total', ' avg')
rpt = rpt.replace('support', 'N Obs')
plt.annotate(rpt,
xy = (0,0),
xytext = (-50, -140),
xycoords='axes fraction', textcoords='offset points',
fontsize=12, ha='left')
# Plot
plt.tight_layout()
下面是用于生成绘图图像的虹膜数据示例：
from sklearn import datasets
from sklearn.svm import SVC
#get data, make predictions
(X,y) = datasets.load_iris(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X,y, train_size=0.5)
clf = SVC()
clf.fit(X_train,y_train)
y_test_pred = clf.predict(X_test)
# Plot Confusion Matrix
plt.style.use('classic')
plt.figure(figsize=(3,3))
pretty_print_conf_matrix(y_test, y_test_pred,
classes= ['0', '1', '2'],
normalize=True,
title='Confusion Matrix')
此代码：
from pandas_ml import ConfusionMatrix
y_actu = [1,2]
y_pred = [1,2]
cm = ConfusionMatrix(y_actu, y_pred)
cm.print_stats()
打印：
population: 2
P: 1
N: 1
PositiveTest: 1
NegativeTest: 1
TP: 1
TN: 1
FP: 0
FN: 0
TPR: 1.0
TNR: 1.0
PPV: 1.0
NPV: 1.0
FPR: 0.0
FDR: 0.0
FNR: 0.0
ACC: 1.0
F1_score: 1.0
MCC: 1.0
informedness: 1.0
markedness: 1.0
prevalence: 0.5
LRP: inf
LRN: 0.0
DOR: inf
FOR: 0.0
/opt/conda/lib/python3.5/site-packages/pandas_ml/confusion_matrix/bcm.py:332: RuntimeWarning: divide by zero encountered in double_scalars
return(np.float64(self.TPR) / self.FPR)
这是预期的。
但是，当我修改代码为：
from pandas_ml import ConfusionMatrix
y_actu = [1,2,3]
y_pred = [1,2,3]
cm = ConfusionMatrix(y_actu, y_pred)
cm.print_stats()
使用：
y_actu = [1,2,3]
y_pred = [1,2,3]
它导致此错误：
OrderedDict([('Accuracy', 1.0), ('95% CI', (0.29240177382128668, nan)), ('No Information Rate', 'ToDo'), ('P-Value [Acc > NIR]', 0.29629629629629622), ('Kappa', 1.0), ("Mcnemar's Test P-Value", 'ToDo')])
ValueErrorTraceback (most recent call last)
<ipython-input-30-d8c5dc2bea73> in <module>()
3 y_pred = [1,2,3]
4 cm = ConfusionMatrix(y_actu, y_pred)
----> 5 cm.print_stats()
/opt/conda/lib/python3.5/site-packages/pandas_ml/confusion_matrix/abstract.py in print_stats(self, lst_stats)
446 Prints statistics
447 """
--> 448 print(self._str_stats(lst_stats))
449
450 def get(self, actual=None, predicted=None):
/opt/conda/lib/python3.5/site-packages/pandas_ml/confusion_matrix/abstract.py in _str_stats(self, lst_stats)
427 }
428
--> 429 stats = self.stats(lst_stats)
430
431 d_stats_str = collections.OrderedDict([
/opt/conda/lib/python3.5/site-packages/pandas_ml/confusion_matrix/abstract.py in stats(self, lst_stats)
390 d_stats = collections.OrderedDict()
391 d_stats['cm'] = self
--> 392 d_stats['overall'] = self.stats_overall
393 d_stats['class'] = self.stats_class
394 return(d_stats)
/opt/conda/lib/python3.5/site-packages/pandas_ml/confusion_matrix/cm.py in __getattr__(self, attr)
33 Returns (weighted) average statistics
34 """
---> 35 return(self._avg_stat(attr))
/opt/conda/lib/python3.5/site-packages/pandas_ml/confusion_matrix/abstract.py in _avg_stat(self, stat)
509 v = getattr(binary_cm, stat)
510 print(v)
--> 511 s_values[cls] = v
512 value = (s_values * self.true).sum() / self.population
513 return(value)
/opt/conda/lib/python3.5/site-packages/pandas/core/series.py in __setitem__(self, key, value)
771 # do the setitem
772 cacher_needs_updating = self._check_is_chained_assignment_possible()
--> 773 setitem(key, value)
774 if cacher_needs_updating:
775 self._maybe_update_cacher()
/opt/conda/lib/python3.5/site-packages/pandas/core/series.py in setitem(key, value)
767 pass
768
--> 769 self._set_with(key, value)
770
771 # do the setitem
/opt/conda/lib/python3.5/site-packages/pandas/core/series.py in _set_with(self, key, value)
809 if key_type == 'integer':
810 if self.index.inferred_type == 'integer':
--> 811 self._set_labels(key, value)
812 else:
813 return self._set_values(key, value)
/opt/conda/lib/python3.5/site-packages/pandas/core/series.py in _set_labels(self, key, value)
826 if mask.any():
827 raise ValueError('%s not contained in the index' % str(key[mask]))
--> 828 self._set_values(indexer, value)
829
830 def _set_values(self, key, value):
/opt/conda/lib/python3.5/site-packages/pandas/core/series.py in _set_values(self, key, value)
831 if isinstance(key, Series):
832 key = key._values
--> 833 self._data = self._data.setitem(indexer=key, value=value)
834 self._maybe_update_cacher()
835
/opt/conda/lib/python3.5/site-packages/pandas/core/internals.py in setitem(self, **kwargs)
3166
3167 def setitem(self, **kwargs):
-> 3168 return self.apply('setitem', **kwargs)
3169
3170 def putmask(self, **kwargs):
/opt/conda/lib/python3.5/site-packages/pandas/core/internals.py in apply(self, f, axes, filter, do_integrity_check, consolidate, **kwargs)
3054
3055 kwargs['mgr'] = self
-> 3056 applied = getattr(b, f)(**kwargs)
3057 result_blocks = _extend_blocks(applied, result_blocks)
3058
/opt/conda/lib/python3.5/site-packages/pandas/core/internals.py in setitem(self, indexer, value, mgr)
685 indexer.dtype == np.bool_ and
686 len(indexer[indexer]) == len(value)):
--> 687 raise ValueError("cannot set using a list-like indexer "
688 "with a different length than the value")
689
ValueError: cannot set using a list-like indexer with a different length than the value
我发现了类似的 question，它指出
Using endemic lists is not allowed on assignment and is not recommended to do this at all.
什么是特有列表，我创建了一个吗？
我建议使用 scikit-learn 中的 confusion_matrix。您提到的其他指标，如 Precision、Recall、F1-score 也可从 sklearn.metrics 获得。
>>> from sklearn.metrics import confusion_matrix
>>> y_actu = [1,2,3]
>>> y_pred = [1,2,3]
>>> confusion_matrix(y_actu, y_pred)
array([[1, 0, 0],
[0, 1, 0],
[0, 0, 1]])
有趣的是，当我 运行 你的代码时，我没有得到你收到的错误，并且代码 运行 完美。我建议您通过 运行ning 升级 pandas_ml 库：
pip install --upgrade pandas_ml
此外，您需要通过 运行ning 升级 pandas：
pip install --upgrade pandas
如果这不起作用，您可以使用 pandas 本身创建一个混淆矩阵：
import pandas as pd
y_actu = pd.Series([1, 2, 3], name='Actual')
y_pred = pd.Series([1, 2, 3], name='Predicted')
df_confusion = pd.crosstab(y_actu, y_pred)
print df_confusion
这将为您提供您正在寻找的 table。
似乎错误不是因为数组维度：
from pandas_ml import ConfusionMatrix
y_actu = [1,2,2]
y_pred = [1,1,2]
cm = ConfusionMatrix(y_actu, y_pred)
cm.print_stats()
这个（二元分类问题）工作正常。
也许多类分类问题的混淆矩阵刚刚被打破。
更新： 我只是做了这些步骤：
conda update pandas
得到pandas0.20.1 然后
pip install -U pandas_ml
现在多类混淆矩阵一切正常：
from pandas_ml import ConfusionMatrix
y_actu = [1,2,3]
y_pred = [1,2,3]
cm = ConfusionMatrix(y_actu, y_pred)
cm.print_stats()
我得到了输出：
Class Statistics:
Classes 1 2 3
Population 3 3 3
P: Condition positive 1 1 1
N: Condition negative 2 2 2
Test outcome positive 1 1 1
Test outcome negative 2 2 2
TP: True Positive 1 1 1
TN: True Negative 2 2 2
FP: False Positive 0 0 0
FN: False Negative 0 0 0
TPR: (Sensitivity, hit rate, recall) 1 1 1
TNR=SPC: (Specificity) 1 1 1
PPV: Pos Pred Value (Precision) 1 1 1
NPV: Neg Pred Value 1 1 1
FPR: False-out 0 0 0
FDR: False Discovery Rate 0 0 0
FNR: Miss Rate 0 0 0
ACC: Accuracy 1 1 1
F1 score 1 1 1
MCC: Matthews correlation coefficient 1 1 1
Informedness 1 1 1
Markedness 1 1 1
Prevalence 0.333333 0.333333 0.333333
LR+: Positive likelihood ratio inf inf inf
LR-: Negative likelihood ratio 0 0 0
DOR: Diagnostic odds ratio inf inf inf
FOR: False omission rate 0 0 0
我也使用并推荐 sklearn confusion_matrix 函数。就我个人而言，我还保留了一个 "pretty-print confusion matrix" 函数，它有一些额外的便利：
- class 沿混淆矩阵轴打印的标签
- 标准化混淆矩阵统计数据，使所有单元格总和为 1
- 根据单元格值缩放的混淆矩阵单元格颜色
- F 分数等其他指标印在混淆矩阵下方。
像这样：
这是绘图函数，主要基于 this example from the Scikit-Learn documentation
import matplotlib.pyplot as plt
import itertools
from sklearn.metrics import classification_report
def pretty_print_conf_matrix(y_true, y_pred,
classes,
normalize=False,
title='Confusion matrix',
cmap=plt.cm.Blues):
"""
Mostly stolen from: http://scikit-learn.org/stable/auto_examples/model_selection/plot_confusion_matrix.html#sphx-glr-auto-examples-model-selection-plot-confusion-matrix-py
Normalization changed, classification_report stats added below plot
"""
cm = confusion_matrix(y_true, y_pred)
# Configure Confusion Matrix Plot Aesthetics (no text yet)
plt.imshow(cm, interpolation='nearest', cmap=cmap)
plt.title(title, fontsize=14)
tick_marks = np.arange(len(classes))
plt.xticks(tick_marks, classes, rotation=45)
plt.yticks(tick_marks, classes)
plt.ylabel('True label', fontsize=12)
plt.xlabel('Predicted label', fontsize=12)
# Calculate normalized values (so all cells sum to 1) if desired
if normalize:
cm = np.round(cm.astype('float') / cm.sum(),2) #(axis=1)[:, np.newaxis]
# Place Numbers as Text on Confusion Matrix Plot
thresh = cm.max() / 2.
for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):
plt.text(j, i, cm[i, j],
horizontalalignment="center",
color="white" if cm[i, j] > thresh else "black",
fontsize=12)
# Add Precision, Recall, F-1 Score as Captions Below Plot
rpt = classification_report(y_true, y_pred)
rpt = rpt.replace('avg / total', ' avg')
rpt = rpt.replace('support', 'N Obs')
plt.annotate(rpt,
xy = (0,0),
xytext = (-50, -140),
xycoords='axes fraction', textcoords='offset points',
fontsize=12, ha='left')
# Plot
plt.tight_layout()
下面是用于生成绘图图像的虹膜数据示例：
from sklearn import datasets
from sklearn.svm import SVC
#get data, make predictions
(X,y) = datasets.load_iris(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X,y, train_size=0.5)
clf = SVC()
clf.fit(X_train,y_train)
y_test_pred = clf.predict(X_test)
# Plot Confusion Matrix
plt.style.use('classic')
plt.figure(figsize=(3,3))
pretty_print_conf_matrix(y_test, y_test_pred,
classes= ['0', '1', '2'],
normalize=True,
title='Confusion Matrix')