Ml-agents: Executable is running, but Not Training

[OneFiveChen]

When I use mlagents-learn to train , the executable can run and Agents in the scene can take actions ,but the terminal shows me that Not Training

INFO:mlagents.trainers: test_VisSoccer_1-0: Brain: Step: 0. No episode was completed since last summary. Not Training.
INFO:mlagents.trainers: test_VisSoccer_1-0: Brain: Step: 0. No episode was completed since last summary. Not Training.
INFO:mlagents.trainers: test_VisSoccer_1-0: Brain: Step: 0. No episode was completed since last summary. Not Training.

I use visual observation as input. Memory increases and quickly runs out.

Everything ok last week. v0.6 released 6 days ago, is that the reason?
My ml-agents version is 0.5.

[awjuliani]

Hi @OneFiveChen

You need to ensure that there is an episode ending condition for your agents. That can either be something that calls Done() on the agent, or setting a max steps for the agent.

[OneFiveChen]

@awjuliani I build the sample scence: 3dball and train it. It also have the same problem

INFO:mlagents.trainers: test_3dball_1-0: Ball3DBrain: Step: 0. Mean Reward: 0.250. Std of Reward: 0.431. Not Training.
INFO:mlagents.trainers: test_3dball_1-0: Ball3DBrain: Step: 0. Mean Reward: 0.268. Std of Reward: 0.440. Not Training.
INFO:mlagents.trainers: test_3dball_1-0: Ball3DBrain: Step: 0. Mean Reward: 0.264. Std of Reward: 0.413. Not Training.
INFO:mlagents.trainers: test_3dball_1-0: Ball3DBrain: Step: 0. Mean Reward: 0.244. Std of Reward: 0.407. Not Training.

And I check my Done() function and max steps, it's not the origin of the problem

[awjuliani]

@OneFiveChen

I just realized that you likely forgot the --train argument when launching mlagents-learn. That is why Not Training is appearing.

[OneFiveChen]

@awjuliani I got that. A few days no training , I forget the command... Thank u

[prgmrjeong]

@awjuliani thank you for ur answer :)

[SaiAvinashDamera]

@OneFiveChen

I just realized that you likely forgot the --train argument when launching mlagents-learn. That is why Not Training is appearing.

this solved my problem