理解Keras预测

Question

我得到了以下代码：

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras.preprocessing.text import Tokenizer
from tensorflow.keras.preprocessing.sequence import pad_sequences
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense
from tensorflow.keras.callbacks import EarlyStopping
import numpy as np
from numpy.random import seed
from tensorflow import random

seed(42)
random.set_seed(43)

X = [
    'may it all be fine in the world',
    'this is not for me',
    'pffff ugly bike',
    'dropping by to say leave me alone',
    'getting sarcastic by now'
    'how would one satisfy his or her needs when the earth is boiling'
]

y = [1,2,4,5,3]

tokenizer = Tokenizer(num_words = 13)
tokenizer.fit_on_texts(X)
X_train_seq = tokenizer.texts_to_sequences(X)


X_train_seq_padded = pad_sequences(X_train_seq, maxlen = 15)

model = Sequential()
model.add(Dense(16, input_dim = 15, activation = 'relu', name = 'hidden-1'))
model.add(Dense(16, activation = 'relu', name = 'hidden-2'))
model.add(Dense(16, activation = 'relu', name = 'hidden-3'))
model.add(Dense(5, activation='softmax', name = 'output_layer'))

model.compile(optimizer = 'adam', loss = 'categorical_crossentropy', metrics=['accuracy'])

class CustomCallback(keras.callbacks.Callback):
    def on_epoch_end(self, epoch, logs=None):
        print('finished an epoch')
        zin = 'dropping by to say leave her alone'
        zin = tokenizer.texts_to_sequences(zin)
        zin = pad_sequences(zin, maxlen = 15)
        print(model.predict(zin))
        print(np.argmax(model.predict(zin), axis=-1))
callbacks = [EarlyStopping(monitor = 'accuracy', patience = 5, mode = 'max'), CustomCallback()]

from sklearn.preprocessing import LabelBinarizer
encoder = LabelBinarizer()
y = encoder.fit_transform(y)

history = model.fit(X_train_seq_padded, y, epochs = 100, batch_size = 100, callbacks = callbacks)

我预计在回调model.predict()内部会产生类似这样的结果(因为有5个可能的类)：

[0.4534534, 0.5634246, 0.0045623, 0.0004536, 0.0000056]

和单个数字1、2、3、4或5中的np.argmax(model.predict(zin), axis=-1)。

然而，我收到的输出(显示一个时期)是：

​

​

我必须如何解释这一点，以及如何过滤出模型将预测句子所属的实际类？

Answer 1

print(model.predict(zin)[0])
print(np.argmax(model.predict(zin)[0], axis=-1))

这将为您提供正确的值。

tf模型被设计为以批处理方式工作，而不是一次性使用，因此它为您提供了一个输出列表，但因为您的输入是单个项目，所以它只会将该项目通过NN推送n次，从而产生相同的输出列表