Closed Vitaliy31 closed 3 years ago
Hi @Vitaliy31,
Thanks for the feedback. Could you please share the train log you receive before the kernel dies and (if possible) the Ipynb file? This info will help us to find the issue faster.
Thanks in advance.
Alex
Sure!
And the train log is below
Lvl_0_Pipe_0_Mod_0_LinearL2 fitting and predicting completed
Time left 576.231377363205
Start fitting Selector_LightGBM ...
===== Start working with fold 0 for Selector_LightGBM =====
Training until validation scores don't improve for 200 rounds
[100] valid's l1: 2114.08
[200] valid's l1: 1960.2
[300] valid's l1: 1901.21
[400] valid's l1: 1868.39
[500] valid's l1: 1839.31
[600] valid's l1: 1817.48
[700] valid's l1: 1796.82
[800] valid's l1: 1782.05
[900] valid's l1: 1771.27
[1000] valid's l1: 1763.77
[1100] valid's l1: 1752
[1200] valid's l1: 1744.96
Did not meet early stopping. Best iteration is:
[1200] valid's l1: 1744.96
Selector_LightGBM fitting and predicting completed
@Vitaliy31,
Thanks! Have tried your code on Ubuntu and MacOS - everything works fine. As I can see you have worked on MacOS - for me it was python 3.7. Could you share your python version and package versions?
Could you also check the jupyter console during the kernel crash - the stack trace can be there...
Alex
@alexmryzhkov ,
I use Python 3.9 and the following packages versions lightautoml 0.2.16 numpy 1.20.1 pandas 1.2.4 torch 1.8.1 lightgbm 3.2.1
the Jupiter console just says
KernelRestarter: restarting kernel (1/5), keep random ports
kernel 36186ccd-6689-4ce8-a7a2-79f6139fbf7a restarted
@Vitaliy31 That's sounds really interesting - I have tried your notebook on python 3.9 inside the kernel created using the commands (and it works fine):
conda create -n new_py39 python=3.9
conda activate new_py39
pip install -U pip
pip install -U jupyter ipython
pip install -U lightautoml
pip install -U ipykernel
python -m ipykernel install --name py39-new
jupyter notebook
I have also appended the requirements.txt file with the versions for all packages requirements.txt.
For the packages you have stated in your message comparing to my env:
Alex
@Vitaliy31 I have tried your notebook with the env above and lightgbm==3.2.1 - still works. Notebook with output: Check_LightAutoML_run.ipynb.zip .
Hi @Vitaliy31,
Have you tried the described above way for env creation? If you still faced the same issue please let me know and reopen this issue.
Alex
Hi!
When linear_l2 ends and lgb starts the kernel dies. Tried updating libraries, nothing worked. Dataset size seems to be ok, about 30k observations and 10 columns Using Jupyter on Mac OS.
Thanks!