时间戳

首页 > 解决方案 > 在 Excel 中编写 ANN：VBA 类型不匹配错误

excel - 在 Excel 中编写 ANN：VBA 类型不匹配错误

问题描述

所以我正在尝试使用 Excel VBA 编写一个基本的人工神经网络。我一直在关注一个特别的例子：

https://www.analyticsvidhya.com/blog/2017/05/neural-network-from-scratch-in-python-and-r/

我一直基于文章中 Python 示例的代码（代码示例在文章的底部）。不幸的是，我无法访问 Python，所以我试图用 VBA 来做到这一点。我已尽力将代码转换为 Excel 的可用格式，但是我遇到了一个问题，我不知道如何解决它：

这是我的 VBA 代码：

Sub ANN()
 Dim X(1010, 1011, 101) As Integer 'Input
 Dim Y(1, 1, 0) As Double 'output for comparison
 Dim E(0, 0, 0) As Double 'Error

'Variable Initialization
 Dim Epoch As Long
 Dim LearnRate As Double
 Dim InputLayer_Neurons() As Integer 'Number of Features in data set
 ReDim InputLayer_Neurons(ArrayLen(X))
 Dim HiddenLayer_Neurons As Integer
 Dim Output_Neurons As Integer
 Dim hidden_layer_input1 As Variant
 Dim hidden_layer_input As Variant
 Dim hiddenlayer_activations As Variant
 Dim output_layer_input1 As Variant
 Dim output_layer_input As Variant
 Dim slope_output_layer As Variant
 Dim slope_hidden_layer As Variant
 Dim d_output As Variant
 Dim Output As Variant
 Dim Wh As Double 'Weight Hidden Layer
 Dim Bh As Double 'Bias Hidden Layer
 Dim Wout As Double 'Weight output Layer
 Dim Bout As Double 'Bias Ouput layer
 Dim i As Long
 Epoch = 5000 'Training Iterations
 LearnRate = 0.1 'Learning Rate
 HiddeLayer_Neurons = 3 'Number of Neurons in Hidden Layer
 Output_Neurons = 1 'Number of Neurons at output layer

'Weight & Bias Initialization
 Wh = Application.WorksheetFunction.RandBetween(InputLayer_Neurons, HiddenLayer_Neurons)
 Bh = Application.WorksheetFunction.RandBetween(1, HiddenLayer_Neurons)
 Wout = Application.WorksheetFunction.RandBetween(HiddenLayer_Neurons, Output_Neurons)
 Bout = Application.WorksheetFunction.RandBetween(1, Output_Neurons)

For i = 0 To Epoch

'Forward Propagation
 hidden_layer_input1 = WorksheetFunction.MMult(X, Wh)
 hidden_layer_input = hidden_layer_input1 + Bh
 hiddenlayer_activations = Sigmoid_Activation(hidden_layer_input)
 output_layer_input1 = WorksheetFunction.MMult(hiddenlayer_activations, Wout)
 output_layer_input = output_layer_input1 + Bout
 Output = Derivatives_Sigmoid(output_layer_input)

'Backpropagation
 E = Y - Output
 slope_output_layer = Derivatives_Sigmoid(Output)
 slope_hidden_layer = Derivatives_Sigmoid(hiddenlayer_activations)
 d_output = E * slope_output_layer
 Error_at_hidden_layer = WorksheetFunction.MMult(d_output, Transpose(Wout))
 d_hiddenlayer = Error_at_hidden_layer * slope_hidden_layer
 Wout = Wout + WorksheetFunction.MMult(Transpose(hiddenlayer_activations), d_output) * LearnRate
 Bout = Bout + WorksheetFunction.Sum(d_ouput) * LearnRate
 Wh = Wh + WorksheetFunction.MMult(Transpose(X), d_hiddenlayer) * LearnRate
 Bh = Bh + WorksheetFunction.Sum(d_hiddenlayer) * LearnRate

Next

Debug.Print Output

End Sub

Function Sigmoid_Activation(X) As Variant
 Sigmoid_Activation = 1 / (1 + Application.WorksheetFunction.Power(-X))
 End Function

Function Derivatives_Sigmoid(X) As Double
 Derivatives_Sigmoid = X * (1 - X)
 End Function

Public Function ArrayLen(arr As Variant) As Integer
 ArrayLen = UBound(arr) - LBound(arr) + 1
 End Function

我的问题是：

在反向传播下的部分。我在该行收到类型不匹配错误：

E = Y – Output

我想这是因为在 VBA 中没有用于从数组（Y）中减去 Double 类型值（输出）的内置函数。即使 Output 被声明为变体，我相信它将包含的值将包含一个浮点值。我不确定这是否是冲突的原因。但是，您似乎可以在 Python 中执行此操作，这可能就是它用于科学计算的原因。

无论如何，解决此问题的最佳方法是什么？

标签： excelvbamachine-learningneural-networkbackpropagation

解决方案

不幸的是，VBA 不支持对数组的直接操作。相反，您必须循环:(

如果你真的想用 VBA 来做，你可以看看这个CodeReview 帖子。

如果您想使用 Excel 来完成，但无法接受 VBA 的限制，有一些 Python 工具允许使用 Python 编写 Excel。快速搜索会让你得到很多，包括https://www.xlwings.org/，它已经在我的列表中尝试太久了。

推荐阅读