Tensorboard logging does not work in TF2 branch

[lerad]

I got an error, when trying to log tensorboard output in the TF2 branch:

set OPENAI_LOG_FORMAT=stdout,log,csv,tensorboard
python -m baselines.run --alg=ppo2 --env=CartPole-v0 --network=mlp --save_path model --log_path log/ --num_timesteps=30000 --nsteps=128

The following exception occured:

File "[..]\lib\site-packages\baselines\run.py", line 249, in
main(sys.argv)
File "[..]\lib\site-packages\baselines\run.py", line 209, in main
model, env = train(args, extra_args)
File "[..]\lib\site-packages\baselines\run.py", line 79, in train
**alg_kwargs
File "[..]\lib\site-packages\baselines\ppo2\ppo2.py", line 193, in learn
logger.dumpkvs()
File "[..]\lib\site-packages\baselines\logger.py", line 219, in dumpkvs
return get_current().dumpkvs()
File "[..]\lib\site-packages\baselines\logger.py", line 333, in dumpkvs
fmt.writekvs(d)
File "[..]\lib\site-packages\baselines\logger.py", line 163, in writekvs
summary = self.tf.Summary(value=[summary_val(k, v) for k, v in kvs.items()])
AttributeError: module 'tensorflow' has no attribute 'Summary'

I used a virtualenv with only tensorflow==2.0.0, tensorflow-gpu=2.0.0, baselines version
(commit 713f1a0) and python 3.6.7.

[christopherhesse]

@tanzhenyu

[manavchoudhary]

@tanzhenyu , @christopherhesse, @pzhokhov
I too get an error when trying to use tensorboard log format to train PPO2, in the 'tf2' branch:
Command :

OPENAI_LOGDIR=./boxing_log OPENAI_LOG_FORMAT=tensorboard python -m baselines.run --alg=ppo2 --env=BoxingNoFrameskip-v4 --num_timesteps=1e7 --nsteps=128

I got the following error :

Traceback (most recent call last):
File "[..]/anaconda3/envs/extar/lib/python3.7/runpy.py", line 193, in _run_module_as_main
"main", mod_spec)
File "[..]/anaconda3/envs/extar/lib/python3.7/runpy.py", line 85, in _run_code
exec(code, run_globals)
File "[..]/baselines/baselines/run.py", line 250, in
main(sys.argv)
File "[..]/baselines/baselines/run.py", line 211, in main
configure_logger(args.log_path)
File "[..]/baselines/baselines/run.py", line 199, in configure_logger
logger.configure(**kwargs)
File "[..]/baselines/baselines/logger.py", line 395, in configure
output_formats = [make_output_format(f, dir, log_suffix) for f in format_strs]
File "[..]/baselines/baselines/logger.py", line 395, in
output_formats = [make_output_format(f, dir, log_suffix) for f in format_strs]
File "[..]/baselines/baselines/logger.py", line 185, in make_output_format
return TensorBoardOutputFormat(osp.join(ev_dir, 'tb%s' % log_suffix))
File "[..]/baselines/baselines/logger.py", line 156, in init
self.writer = pywrap_tensorflow.EventsWriter(compat.as_bytes(path))
AttributeError: module 'tensorflow.python.pywrap_tensorflow' has no attribute 'EventsWriter'

I am using tensorflow==2.2.0, gym==0.15.7, baselines commit #1032, Python 3.7.6

My understanding of the above error is that, the tensorboard summary writing code in the 'baselines/common/logger.py' of the tf2 branch, has not been properly adapted for TF2.0.
Also, before the above error I got another error regarding 'FlattenDictWrapper' but I fixed that using #1051

Thanks a lot for your help.

[xpz234]

AttributeError: module 'tensorflow.python.pywrap_tensorflow' has no attribute 'EventsWriter'

@manavchoudhary
Do you have any way to solve this problem？

[omsrisagar]

@xpz234

I solved this error with the following:

from tensorflow.python.client import _pywrap_events_writer
self.writer = _pywrap_events_writer.EventsWriter(compat.as_bytes(path))

This is the source I used for obtaining this fix:
https://github.com/tensorflow/tensorflow/blob/87462bfac761435a46641ff2f10ad0b6e5414a4b/tensorflow/python/client/events_writer_test.py