search

wassname / rl-portfolio-management

Attempting to replicate "A Deep Reinforcement Learning Framework for the Financial Portfolio Management Problem" https://arxiv.org/abs/1706.10059 (and an openai gym environment)
544 stars 179 forks source link

There's no attribute 'constraints' at object 'Sequential' #1

Closed jd0713 closed 7 years ago

jd0713 commented 7 years ago

Actor network used Sequential() using keras and it goes into DDPG agent. And at agent.compile(Adam(lr=3e-5), metrics=['mse']) there says error that 'Sequential' object has no attribute 'constraints' It seems that I need to set constraints when use Conv2D, but not sure how to do it.

I also went to keras-rl and run ddpg_pendulum.py, but the same error occured

I think there are three possible way to solve it.

Change python version (3.5.0 --> 2.7) Input constraint into convolutional layer Change keras version (2.0.8 --> 1.. whatever)

Can you help me to solve this problem?

Thanks.

wassname commented 7 years ago

I was using python 3.5 so that's OK. These are the package version these so try using the same version of keras-rl. If that doesn't work try "'keras>=1.0.7'" which is required by the keras-rl version I was using. If that still doesn't work let me know and I'll try running on my machine again

I found all the RL libraries I tried to be very picky and buggy, so it's not just you.

jd0713 commented 7 years ago

After I change my keras version 2.0.8 to 2.0.3, it worked.

I should better doing some hyperparameter fitting and apply this code to another data set.

Thanks a lot!

wassname commented 7 years ago

Good to know which version works. Good luck!