“FIN Status Report 2016-11-21”版本间的差异

Test drrl with 1min data and obtain good results.
Find some problems in training, such as the fact that all the action are +1 or -1.
It seems that it can be trained successful at a probability.

study the problems

Use 1min daily data for dqt and acquire good results.
Design automated network establishing strategy.
Read some papers about local minima in training neural networks.

Try to add factors as part of the input data.
Implement experiments of drrl using 1min data.