# Unable to train Coach for Banana-v0 Gym environment

I have just started playing with Reinforcement learning and starting from the basics I'm trying to figure out how to solve Banana Gym with coach.

Essentially Banana-v0 env represents a Banana shop that buys a banana for \$1 on day 1 and has 3 days to sell it for anywhere between \$0 and \\$2, where lower price means a higher chance to sell. Reward is the sell price less the buy price. If it doesn't sell on day 3 the banana is discarded and reward is -1 (banana buy price, no sale proceeds). That's pretty simple.

Ideally the algorithm should learn to set a high price on day 1 and reducing it every day if it didn't sell.

To start I took the coach-bundled CartPole_ClippedPPO.py and CartPole_DQN.py preset files and modified them to run Banana-v0 gym.

The trouble is that I don't see any learning progress regardless what I try, even after running like 50,000 episodes. In comparison the CartPole gym successfully trains in under 500 episodes.

I would have expected some improvement after 50k episodes for such a simple task like Banana.

Is it because the Banana-v0 rewards are not predictable? I.e. whether the banana sells or not is still determined by a random number (with success chance based on the price).

Where should I take it from here? How to identify which Coach agent algorithm I should start with and try to tune it?

• Which coach algorithm did you use? If it was one that worked well for CartPole, it may need to be changed to work with a new environment. – Neil Slater Nov 27 '19 at 8:09
• @NeilSlater I tried ClippedPPO and DQN. If it showed at least some improvement I could take it from there and try to tune it, but as I didn't observe any noticeable improvement I doubt these two are fit for Banana. Maybe some other algorithm would do better? Not sure which one to try. – KeepLearning Nov 27 '19 at 20:30
• It's probably not the choice of algorithm, but hyperparameters that is working against you. If I get time I will take a deeper look and try to answer. – Neil Slater Nov 28 '19 at 8:04
• @NeilSlater thanks! – KeepLearning Nov 29 '19 at 2:37