Distributed Artificial Intelligence by Unknown

Author:Unknown
Language: eng
Format: epub
ISBN: 9783030946623
Publisher: Springer International Publishing

PPO

SDPO

PD-CVaR

PD-VAR

Discount factor

0.99

0.99

0.99

0.99

Batch size

1000

1000

1000

1000

Learning rate (actor)

1e–4

1e–4

1e–4

1e–4

Learning rate (critic)

1e–3

1e–3

/

/

Hidden sizes

(64, 64)

(64, 64)

(64, 64)

(64, 64)

GAE factor

0.9

0.9

/

/

Clip range

0.2

0.2

/

/

/

20

/

/

# quantile atoms N

/

128

/

/

Quantile dimension

/

256

/

/

Refer to Equation (4) in [16]

C.2 Stock Transaction

For the experiment on stock market, we use Quandl1 in Python to load all market data. The trading agent is assumed to have zero market impact and zero transaction cost. When conducting this experiment, We choose 9 stocks in SP500 (AAPL, CSCO, DOW, GE, GS, JNJ, JPM, MMM, MSFT). The agents are initialized with a safe policy that always holding cash, and then trained in a rolling bias in year 2019 to evaluate the offline performance, i.e., at time step t, prices from to t are used for training. The shared part of the actor and critic network is implemented as an LSTM network. The hyper-parameters are listed in Table 4.Table 4.Hyperparameters for experiments on stock transaction.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.