“FIN Status Report 2016-10-31”版本间的差异

2016年10月31日 (一) 01:39的版本

People	Last Week	This Week
Yang Wang	find some problem in DQN and DDPG : the stock is a Markov process instead of a decision Markov process. implement the recurrent reinforcement learning algorithm.	do some experiments with DRRL.
Yuxin Zhang	Add normalization to the input data and get better experiment results. Try different values of parameters and find more suitable settings. Use multiple stocks for training to improve the performance of the algorithm.	Try to use high frequency stock data as inputs.

@@ 第12行： / 第12行： @@
 |-
 | Yuxin Zhang||
-Read programs created by Yang Wang : DQN DRRL. <br>
+Add normalization to the input data and get better experiment results. <br>
-Discuss the current bottlenecks of the algorithms and possible ways to break them.<br>
+Try different values of parameters and find more suitable settings.<br>
-Read papers related to using rnn for trading.
+Use multiple stocks for training to improve the performance of the algorithm.
 ||
-try to demonstrate if DRRL works well (or at least if it's better than DQN).
+Try to use high frequency stock data as inputs.
 |}