python - ValueError：输入形状的预期轴 -1 的值为 51948，但接收到的输入形状为（无，52）

Question

import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
from sklearn.model_selection import train_test_split
import tensorflow.keras as keras

dataset = pd.read_csv('C:\\Users\\Maxie\\MyStuff\\FinalDatasetEng.csv')
inputs = dataset.iloc[:, 2:54].values
targets = dataset.iloc[:, 55].values

from sklearn.model_selection import train_test_split
inputs_train, inputs_test, targets_train, targets_test = train_test_split(inputs, targets, 
test_size = 0.20, random_state = 0)

import keras
from keras.models import Sequential
from keras.layers import Dense

model = keras.Sequential([

        # input layer
        keras.layers.Flatten(input_shape=(inputs.shape[0], inputs.shape[1])),

        # 1st dense layer
        keras.layers.Dense(520, activation='relu'),

        # 2nd dense layer
        keras.layers.Dense(208, activation='relu'),

        # 3rd dense layer
        keras.layers.Dense(52, activation='relu'),

        # output layer
        keras.layers.Dense(4, activation='softmax')
    ])

optimiser = keras.optimizers.Adam(learning_rate=0.0001)
model.compile(optimizer=optimiser,
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

history = model.fit(inputs_train, targets_train, validation_data=(inputs_test, targets_test), 
batch_size=32, epochs=50)

这是我的代码，我收到此错误：ValueError：dense_20 层的输入 0 与该层不兼容：输入形状的预期轴 -1 具有值 51948，但接收到形状的输入（无，52）。任何人都请帮我解决这个问题。

标签： pythontensorflowkerasneural-network

Answer 1

您有一个一维数组作为您的特征输入，但是您将样本数量和特征数量展平，从而为模型提供 51948 个输入特征（999 个样本input.shape[0]* 52 个特征input.shape[1]= 51948）。因此，您的模型需要一个 51948 输入的数组，但您已经通过inputs_train了 52 列。

推理：

如果您将一维数组作为输入要素，则不应展平您的输入。您的输入特征是 52 列和 999 个样本的数组。代替Flatten层，使用InputLayer.

所以，修改后的代码应该是这样的：

model = keras.Sequential([

        # input layer
        #change this line to input layer and set the input shape to the shape of your input features
        #keras.layers.Flatten(input_shape=(inputs.shape[0], inputs.shape[1])),
        keras.layers.InputLayer(input_shape=(inputs.shape[1],)), 

        # 1st dense layer
        keras.layers.Dense(520, activation='relu'),

        # 2nd dense layer
        keras.layers.Dense(208, activation='relu'),

        # 3rd dense layer
        keras.layers.Dense(52, activation='relu'),

        # output layer
        keras.layers.Dense(4, activation='softmax')
    ])

optimiser = keras.optimizers.Adam(learning_rate=0.0001)
model.compile(optimizer=optimiser,
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

history = model.fit(inputs_train, targets_train, validation_data=(inputs_test, targets_test), 
batch_size=32, epochs=50)