在使用神经网络获得高状态空间的泛化时，输入单位是什么？
例如，如果状态向量是1维，比如说实轴上的位置......那么只有一个输入单位？ （为每个行动提供单独的网络）
答案 0 :(得分：0)
是的，至少如果您使用类似于Q-learning或Sarsa的算法，其中函数逼近器应该学习Q函数Sub StripRow2Node()
'Read the Strip Design table
With Sheets("Design-Moment")
Sheets("Design-Moment").Activate
LastR1 = .Range("B" & Cells.Rows.Count).End(xlUp).Row
DM_arr = .Range(Cells(1, 1), Cells(LastR1, 7)) 'Col 1 to Col 7
DM_count = UBound(DM_arr, 1)
End With
'Read the x and y coordinations and thickness of a node in node design
With Sheets("Design-Shear")
Sheets("Design-Shear").Activate
LastR2 = .Range("B" & Cells.Rows.Count).End(xlUp).Row
DS_arr = .Range(Cells(1, 4), Cells(LastR2, 5)) 'Col 4 to Col 5
SX_arr = .Range(Cells(1, 26), Cells(LastR2, 27))
SY_arr = .Range(Cells(1, 30), Cells(LastR2, 31))
DS_count = UBound(DS_arr, 1)
End With
'** Find correponding reference row in Design-Moment for nodes**
'Match node to striip station and output row index
For i = 5 To DS_count
XStrip = SX_arr(i, 1)
XStation = DS_arr(i, 1)
YStrip = SY_arr(i, 1)
YStation = DS_arr(i, 2)
For j = 5 To DM_count
If DM_arr(j, 1) = XStrip Then 'X-Strip Name is matched
If DM_arr(j, 4) >= XStation And DM_arr(j - 1, 4) < XStation Then
SX_arr(i, 2) = j 'matched row reference for X-strip
End If
End If
If DM_arr(j, 1) = YStrip Then
If DM_arr(j, 5) <= YStation And DM_arr(j - 1, 5) > YStation Then
SY_arr(i, 2) = j
End If
End If
Next j
Next i
'Write the matched strip information to node
For i = 5 To LastR2
With Sheets("Design-Shear")
.Cells(i, 27) = SX_arr(i, 2)
.Cells(i, 31) = SY_arr(i, 2)
End With
Next i
。在您的情况下，如果每个操作使用一个神经网络，则网络必须近似函数Q(s,a)。此外，如果状态具有维度1，那么网络将只需要一个输入神经元。