“FIN Status Report 2017-3-6”版本间的差异

training with different stock set, which consist of 10 stocks belonging to 10 industries.
digitize the asset and set it as reward
clip gradients. stabilizez the training
random goal.

Try to add convolutional layers to the ddpg model.
Use more suitable reward function.

Use Pyalgotrading to test the algorithm.
Start to develop algorithm for in-day market timing.