I’m running an issue that is similar in this threa...
# dev-metaflow
w
I’m running an issue that is similar in this thread: In my case the error is the following:
metaflow --version
ValueError: Cannot locate step_decorator plugin 'batch' at 'metaflow.plugins.aws.batch.batch_decorator'
Conda version: 23.10.0 It’s weird because when I do
pip install --upgrade metaflow
it returns:
Requirement already satisfied: six>=1.5 in /Users/servandodavidtorresgarcia/miniconda3/envs/relu_ml/lib/python3.8/site-packages (from python-dateutil<3.0.0,>=2.1->botocore<1.35.0,>=1.34.98->boto3->metaflow) (1.16.0)
Runnin Mac https://outerbounds-community.slack.com/archives/C020U025QJK/p1694375271643699?thread_ts=1694159385.792319&cid=C020U025QJK
@happy-wolf-7852 @dry-beach-38304 I believe there was a solution in the thread mentioned 🙂
h
Well I recreated the whole virtual environment …
w
Did you change versions? I’m using conda environment I believe you used pyenv right ?
h
no I use miniconda …
1
w
same
h
yeah I just used that but have to say that I also had another Python 3.9 installed and removed that …
w
Did you remove the entire miniconda installation or just re-create another conda env?
h
IIRC just re-create another conda env
1
d
did you get it working? This may be a separate issue because the previous issue was a bad code (ie: it couldn’t load an extension). Batch is not provided as part of an extension. If you didn’t get it working, you can try to run
METAFLOW_DEBUG_EXT=1 metaflow --version
and let me know what it says - that can give me more info about what is happening on startup.
w
So far so good. Just created another conda env and
pip install metaflow
again. So far :
from metaflow import FlowSpec, step
class HelloFlow(FlowSpec):
@step
def start(self):
print("Hello world!")
self.next(self.end)
@step
def end(self):
pass
if __name__ == "__main__":
HelloFlow()
runs without problem
d
🙂
w
Just a minor thing: After installing libraries like lgbm and sklearn with
conda forge
metaflow magically disappeared I had to pip install again. Should we look at what channels we instal our libraries? conda-forge, pip etc?
d
that is weird. metaflow should also be available in conda.