问题的本质:我想用互联网上现成的例子来理解最简单的神经网络。我训练了它,然后我不明白如何在条件用户的输入数据上测试它的有效性。我在互联网上找到了模型的函数:predict((、save((、loaded_model((。如果结果是save((和loaded_model((,并且创建了文件夹"16_model",则会从predict((引发错误。请告诉我如何使用它,或者如何在输入上测试神经网络,而不是在测试数据上。
import keras
from keras.models import Sequential
from keras.layers import Dense
import numpy
numpy.random.seed(2)
dataset = numpy.loadtxt("diabet.csv", delimiter=",")
X, Y = dataset[:,0:8], dataset[:,8]
model = Sequential()
model.add(Dense(12, input_dim=8, activation='relu'))
model.add(Dense(10, activation='relu'))
model.add(Dense(1, activation='sigmoid'))
model.compile(loss="binary_crossentropy", optimizer="adam", metrics=['accuracy'])
model.fit(X, Y, epochs = 100, batch_size=10)
scores = model.evaluate(X, Y)
#model.save('16_model')
#model_loaded = keras.models.load_model('16_model')
print("n%s: %.2f%%" % (model.metrics_names[1], scores[1]*100))
您可以将数据集拆分为training
和testing
部分,如下所示,并使用这些training set
来训练模型,使用test dataset
来评估模型:
import pandas as pd
from sklearn.model_selection import train_test_split
dataset =pd.read_csv("/content/../diabetes.csv")#, delimiter=",")
dataset.head()
X = dataset.drop('Outcome', 1)
y = dataset['Outcome']
Y.shape
train_X, test_X, train_Y, test_Y=train_test_split(X,y, test_size=0.2)
train_X.shape
然后,在用train dataset
定义和训练模型之后,可以用test dataset
评估模型。
model_loss, model_acc = model.evaluate(test_X, test_Y)
print("Model Accuracy", model_acc*100)
model.save('16_model')
model_loaded = keras.models.load_model('16_model')
输出:
5/5 [==============================] - 0s 2ms/step - loss: 0.5637 - accuracy: 0.7208
Model Accuracy 72.07792401313782
INFO:tensorflow:Assets written to: 16_model/assets
现在,在保存并加载回相同的模型后再次评估模型,这显示出与保存模型之前相同的精度。
model_loss, model_acc = model_loaded.evaluate(test_X, test_Y)
输出:
5/5 [==============================] - 0s 3ms/step - loss: 0.5637 - accuracy: 0.7208
这是预测部分:
(由于用于二进制分类的sigmoid
函数,这将显示0到1范围内的所有值(
pred=model_loaded.predict(train_X[:2])
print(pred))
输出:
[[0.925119 ]
[0.45006576]]
<0.5将属于类别0,其余全部将被假定为类别1。
for p in pred:
if p>=.5:
pred_class=1
else:
pred_class=0
print(pred_class)
输出:
1
0
你可以用实际标签验证这个预测:
train_Y[:2]
输出:
369 1
653 0
Name: Outcome, dtype: int64