Skip to main content
Financial Innovation

Table 8 The TCN architecture of the models

From: DeepPricing: pricing convertible bonds based on financial time-series generative adversarial networks

Module name

Arguments

Temporal block 1

(\(N_I\),\(N_H\),\(N_H\),1,1)

Temporal block 2

(\(N_H\),\(N_H\),\(N_H\),2,1)

Temporal block 3

(\(N_H\),\(N_H\),\(N_H\),2,2)

Temporal block 4

(\(N_H\),\(N_H\),\(N_H\),2,4)

Temporal block 5

(\(N_H\),\(N_H\),\(N_H\),2,8)

Temporal block 6

(\(N_H\),\(N_H\),\(N_H\),2,16)

Temporal block 7

(\(N_H\),\(N_H\),\(N_H\),2,32)

1 \(\times\) 1 Convolution

(\(N_H\),\(N_O\),1,1)

  1. Table 8 reports the specific arguments of the TCN architecture. A TCN is composed of multiple temporal block modules with arguments \((N_I, N_H, N_O, K, D)\) and an \(1\times 1\) convolution layer, where \(N_I\) stands for the input dimension,\(N_H\) stands for the hidden dimension,\(N_O\) stands for the output dimension data,K stands for the kernel size, and D stands for the dilation. The temporal block modules used in TCN mainly refer to Bai et al. (2018), and the specific module construction details are in http://github.com/locuslab/TCN
Back to article page