Writing Tensorboard log files causes Google Colab runtime to crash.

Describe the bug When running in a Google Colab, if a model is set up to write tensorboard, the runtime crashes after the first write out (typically in the call to model.learn()).

Code example Running the example for ppo2 (or any other algorithm) with the addition of the tensorbaord_log argument. E.g.

model = PPO2(MlpPolicy, env, verbose=1, tensorboard_log='./log')
model.learn(25000)  # Causes runtime to crash.

System Info Google Colab with dependencies installed as per the documented example Colab here.

[Charelvanhoof]

Hi Araffin,

You are right, in fact I already had it figured out in the past and your feedback triggered me to find it back. Leaving some notes for maybe helping other people

First run

!wget https://bin.equinox.io/c/4VmDzA7iaHb/ngrok-stable-linux-amd64.zip
!unzip ngrok-stable-linux-amd64.zip 

Next:

LOG_DIR = './PPO2_breakout_tensorboard/'
get_ipython().system_raw(
    'tensorboard --logdir {} --host 0.0.0.0 --port 6006 &'
    .format(LOG_DIR)
)

get_ipython().system_raw('./ngrok http 6006 &')

! curl -s http://localhost:4040/api/tunnels | python3 -c \
    "import sys, json; print(json.load(sys.stdin)['tunnels'][0]['public_url'])"

The link will bring you to the tensorboard.

[Charelvanhoof]

Found it: when you put it on accelerator = GPU (iso none). Then the last model.learn(2000) line crashes.