search

terrier-org / ecir2021tutorial

GNU Lesser General Public License v2.1
158 stars 28 forks source link

What versions were used in March? #2

Open tkamm opened 3 years ago

tkamm commented 3 years ago

I'm trying to work through the tutorials in June 2021 and having some issues that I suspect are due to pip installing the latest releases. This might be fixed by updating the notebooks to use the versions as they were in March 2021. I was okay with part 1, but have issues with part 2. I haven't tried parts 3 and 4 yet.

cmacdonald commented 3 years ago

PyTerrier had a release since March, but nothing should have broken. Potentially a dependency has broken, but we're not aware of it. Could you post the error you experience and we can make a recommendation?

tkamm commented 3 years ago

With the latest releases in part 2 notebook, I could not get success on lmart_x_pipe.fit duee to this errorL "Number of rows 32776 exceeds upper limit of 10000 for a query" Once I specified: !pip install --upgrade fastrank==0.7.0 lightgbm==3.1.1 !pip install python-terrier==0.5.0

then lmart_x_pipe.fit worked, but I could not: from pyterrier.measures import * so I reverted back to eval_metrics=['map','ndcg'], and was able to finish the notebook.

cmacdonald commented 3 years ago

For from pyterrier.measures import *, you will need 0.6

This I havent seen before "Number of rows 32776 exceeds upper limit of 10000 for a query"

We'll investigate over the next few days and let you know.

cmacdonald commented 3 years ago

"Number of rows 32776 exceeds upper limit of 10000 for a query"

Was that error created when fitting using xgBoost, FastRank or LightGBM?

tkamm commented 3 years ago

lightgbm==3.2.1 I estimate that 3.1.1 was the version used in March and I confirmed that it works. 3.2.0 is where it starts giving the "Number of rows" error I just confirmed that python-terrier==0.6.0 and fastrank==0.7.0, both the latest versions, seem to work as expected.

cmacdonald commented 3 years ago

Thanks. I have pegged the lighgbm version for notebook2. Once I have verified it, I will close this issue, thanks for the report.

I have promoted your comment about compatibility to a PyTerrier issue.