在 Keras 的分类预测中返回概率? [英] Returning probabilities in a classification prediction in Keras?
问题描述
我正在尝试进行一个简单的概念验证，我可以在其中查看给定预测的不同类别的概率.
I am trying to make a simple proof-of-concept where I can see the probabilities of different classes for a given prediction.
然而，我尝试的一切似乎只输出预测的类，即使我使用的是 softmax 激活.我是机器学习的新手，所以我不确定我是否犯了一个简单的错误，或者这是否是 Keras 中没有的功能.
However, everything I try seems to only output the predicted class, even though I am using a softmax activation. I am new to machine learning, so I'm not sure if I am making a simple mistake or if this is a feature not available in Keras.
我使用的是 Keras + TensorFlow.我已经改编了 Keras 给出的用于分类的基本示例之一MNIST 数据集.
I'm using Keras + TensorFlow. I have adapted one of the basic examples given by Keras for classifying the MNIST dataset.
我下面的代码与示例完全相同，除了一些(已注释)将模型导出到本地文件的额外行.
My code below is exactly the same as the example, except for a few (commented) extra lines that exports the model to a local file.
'''Trains a simple deep NN on the MNIST dataset.
Gets to 98.40% test accuracy after 20 epochs
(there is *a lot* of margin for parameter tuning).
2 seconds per epoch on a K520 GPU.
'''
from __future__ import print_function
import keras
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense, Dropout
from keras.optimizers import RMSprop
import h5py # added import because it is required for model.save
model_filepath = 'test_model.h5' # added filepath config
batch_size = 128
num_classes = 10
epochs = 20
# the data, shuffled and split between train and test sets
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = x_train.reshape(60000, 784)
x_test = x_test.reshape(10000, 784)
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255
x_test /= 255
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')
# convert class vectors to binary class matrices
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)
model = Sequential()
model.add(Dense(512, activation='relu', input_shape=(784,)))
model.add(Dropout(0.2))
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.2))
model.add(Dense(num_classes, activation='softmax'))
model.summary()
model.compile(loss='categorical_crossentropy',
optimizer=RMSprop(),
metrics=['accuracy'])
history = model.fit(x_train, y_train,
batch_size=batch_size,
epochs=epochs,
verbose=1,
validation_data=(x_test, y_test))
score = model.evaluate(x_test, y_test, verbose=0)
print('Test loss:', score[0])
print('Test accuracy:', score[1])
model.save(model_filepath) # added saving model
print('Model saved') # added log
然后这个的第二部分是一个简单的脚本，它应该导入模型，预测某些给定数据的类别，并打印出每个类别的概率.(我使用 Keras 代码库中包含的相同 mnist 类来使示例尽可能简单).
Then the second part of this is a simple script that should import the model, predict the class for some given data, and print out the probabilities for each class. (I am using the same mnist class included with the Keras codebase to make an example as simple as possible).
import keras
from keras.datasets import mnist
from keras.models import Sequential
import keras.backend as K
import numpy
# loading model saved locally in test_model.h5
model_filepath = 'test_model.h5'
prev_model = keras.models.load_model(model_filepath)
# these lines are copied from the example for loading MNIST data
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = x_train.reshape(60000, 784)
# for this example, I am only taking the first 10 images
x_slice = x_train[slice(1, 11, 1)]
# making the prediction
prediction = prev_model.predict(x_slice)
# logging each on a separate line
for single_prediction in prediction:
print(single_prediction)
如果我运行第一个脚本来导出模型，然后运行第二个脚本来对一些示例进行分类，我会得到以下输出:
If I run the first script to export the model, then the second script to classify some examples, I get the following output:
[ 1. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 1. 0. 0. 0. 0. 0.]
[ 0. 1. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 1.]
[ 0. 0. 1. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 1. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 1. 0. 0. 0. 0. 0. 0.]
[ 0. 1. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 1. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 1. 0. 0. 0. 0. 0. 0.]
这对于查看每个类别的预测类别非常有用，但是如果我想查看每个示例的每个类别的相对概率怎么办?我正在寻找更像这样的东西:
This is great for seeing which class each is predicted to be, but what if I want to see the relative probabilities of each class for each example? I am looking for something more like this:
[ 0.94 0.01 0.02 0. 0. 0.01 0. 0.01 0.01 0.]
[ 0. 0. 0. 0. 0.51 0. 0. 0. 0.49 0.]
...
换句话说，我需要知道每个预测的确定性，而不仅仅是预测本身.我认为看到相对概率是在模型中使用 softmax 激活的一部分，但我似乎无法在 Keras 文档中找到任何会给我概率而不是预测答案的东西.是我犯了一些愚蠢的错误，还是这个功能不可用?
In other words, I need to know how sure each prediction is, not just the prediction itself. I thought seeing the relative probabilities was a part of using a softmax activation in the model, but I can't seem to find anything in the Keras documentation that would give me probabilities instead of the predicted answer. Am I making some kind of silly mistake, or is this feature not available?
推荐答案
结果是我没有完全规范化预测脚本中的数据.
So it turns out that the problem was I was not fully normalizing the data in the prediction script.
我的预测脚本应该有以下几行:
My prediction script should have had the following lines:
# these lines are copied from the example for loading MNIST data
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = x_train.reshape(60000, 784)
x_train = x_train.astype('float32') # this line was missing
x_train /= 255 # this line was missing too
因为数据没有被转换为浮点数，而是除以 255(所以它会在 0 和 1 之间)，所以它只是显示为 1 和 0.
Because the data was not cast to float, and divided by 255 (so it would be between 0 and 1), it was just showing up as 1s and 0s.
这篇关于在 Keras 的分类预测中返回概率?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！
