使用 TensorFlow 查找 MNIST 数据集的精度和召回率

问题描述

我正在使用本教程来学习如何在 MNIST 数据集上训练模型:

不过，这是我的问题.我尝试将 model.compile 中的指标更改为 metrics=[tf.keras.metrics.Precision()]，但出现错误 ValueError: Shapes (32, 10) 和 (32, 1) 不兼容.

我也尝试通过 scikit-learn 计算 precisiona 和召回率，但我的预测与真实标签不一致.

y_pred = model.predict(x_test)打印(y_pred)精度分数(y_test，y_pred)

输出:

[[ -4.7507367 -7.4252934 -2.8428416 ... 8.855136 -5.937388-2.1762638 ][ -5.0433793 5.554433 12.963128 ... -18.583 -1.6025407-18.721622 ][ -7.623428 6.3951 -1.8510209 ... 0.37932196 -1.2399373-6.59459 ]...---------------------------------------------------------------------------ValueError 回溯(最近一次调用)<ipython-input-44-a82c4d76f544>在 <module>()1 y_pred = model.predict(x_test)2 打印(y_pred)---->3 precision_score(y_test, y_pred)ValueError:分类指标无法处理多类和连续多输出目标的混合

我想我可能需要转换 y_pred，但我不确定如何转换.或者，如果有一种方法可以为指标增加精确度和召回率，那就更好了.我怎样才能得到这个模型的准确率和召回率?

解决方案

假设您使用代码进行预测:

predicted_result=model.predict(x_test)

输出层有数字0到9的概率，即10.所以从预测结果中需要识别类别.

将 numpy 导入为 npclass_preds = np.argmax(predicted_result,axis=-1)

现在，y_test 和 class_preds 在类中，因此可以运行 precision_score.

from sklearn.metrics import precision_scoreprecision_score(y_test，class_preds，average='macro')

或

from sklearn.metrics 导入recall_score召回分数(y_test，class_preds，average='macro')

甚至可以将此自定义函数提供给指标:

from sklearn.metrics import precision_scoredef custom_prec_score(y_true, y_pred):y_true=y_true.numpy()y_pred=y_pred.numpy()y_pred=np.argmax(y_pred,axis=-1)返回 precision_score(y_true, y_pred,average='macro')

model.compile(optimizer='adam',损失=loss_fn，run_eagerly=真，指标=[准确度"，custom_prec_score])模型.fit(x_train, y_train, epochs=5)

I'm using this tutorial to learn how to train a model on the MNIST dataset here: https://www.tensorflow.org/tutorials/quickstart/beginner

Currently, the model only trains on the accuracy, but I want to figure out the F1-score of the model (starting with precision and recall first).

model = tf.keras.models.Sequential([
  tf.keras.layers.Flatten(input_shape=(28, 28)),
  tf.keras.layers.Dense(128, activation='relu'),
  tf.keras.layers.Dropout(0.2),
  tf.keras.layers.Dense(10)
])
loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)
model.compile(optimizer='adam',
              loss=loss_fn,
              metrics=['accuracy'])
model.fit(x_train, y_train, epochs=5)
Epoch 1/5
1875/1875 [==============================] - 4s 2ms/step - loss: 0.2895 - accuracy: 0.9151
Epoch 2/5
1875/1875 [==============================] - 3s 2ms/step - loss: 0.1393 - accuracy: 0.9586
...

Apparently the model also uses log-odd scores which are converted into probabilities by a softmax as well.

This is my problem, though. I tried changing the metrics in model.compile to metrics=[tf.keras.metrics.Precision()], but I got the error ValueError: Shapes (32, 10) and (32, 1) are incompatible.

I also tried calculating the precisiona and recall through scikit-learn, but my predictions aren't lined up with the true labels.

y_pred = model.predict(x_test)
print(y_pred)
precision_score(y_test, y_pred)

Output:

[[ -4.7507367   -7.4252934   -2.8428416  ...   8.855136    -5.937388
   -2.1762638 ]
 [ -5.0433793    5.554433    12.963128   ... -18.583       -1.6025407
  -18.721622  ]
 [ -7.623428     6.3951      -1.8510209  ...   0.37932196  -1.2399373
   -6.59459   ]
 ...
---------------------------------------------------------------------------

ValueError                                Traceback (most recent call last)

<ipython-input-44-a82c4d76f544> in <module>()
      1 y_pred = model.predict(x_test)
      2 print(y_pred)
----> 3 precision_score(y_test, y_pred)

ValueError: Classification metrics can't handle a mix of multiclass and continuous-multioutput targets

I'm thinking I might need to transform y_pred, but I'm not sure how. Or if there is a way to add precision and recall to the metrics that would be even better. How can I get the precision and recall of this model?

解决方案

suppose you predicted using code:

predicted_result=model.predict(x_test)

the output layer has prob for digit 0 to 9, i.e 10. so from the predicted result need to identify the class.

import numpy as np
class_preds = np.argmax(predicted_result, axis=-1)

now, y_test and class_preds are in classes, so can run precision_score.

from sklearn.metrics import precision_score
precision_score(y_test, class_preds,average='macro')

or

from sklearn.metrics import recall_score
recall_score(y_test, class_preds,average='macro')

even can feed this custom function to metrics:

from sklearn.metrics import precision_score
def custom_prec_score(y_true, y_pred):
    y_true=y_true.numpy()
    y_pred=y_pred.numpy()
    y_pred=np.argmax(y_pred, axis=-1)
    return precision_score(y_true, y_pred,average='macro')

model.compile(optimizer='adam',
              loss=loss_fn,run_eagerly=True,
              metrics=["accuracy",custom_prec_score])
model.fit(x_train, y_train, epochs=5)

这篇关于使用 TensorFlow 查找 MNIST 数据集的精度和召回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！