Submission error

thanish · April 30, 2021, 4:35am

I installed jupyter by sudo apt install jupyter.
If that is wrong can you point me to some document on how to remove all of it and reinstall as recommended?

jyotish · April 30, 2021, 4:54am

@thanish This worked for me.

Exit jupyter
On your terminal, run echo 'export PATH=$PATH:$HOME/.local/bin' >> ~/.bashrc
Restart terminal or run exec bash
Start jupyter

Why do you need to do this?

It looks like pip3 that comes with the OS installs everything to ~/.local and ~/.local/bin is not part of the system path which was leading to aicrowd not found kind of errors.

We recommend using anaconda/miniconda to manage python versions as it takes care of most of such issues. However, as long as things work the other way, I think it should be good.

thanish · April 30, 2021, 5:06am

@jyotish. Thanks that worked. But when I submit I’m getting this error. Can you help me why it is a read only?

jyotish · April 30, 2021, 5:08am

@thanish It looks like the path in AICROWD_ASSETS_DIR is pointing to a read only file system. Can you check if you are able to create files/directories from the directory in which you started jupyter?

thanish · April 30, 2021, 5:22am

Hi @jyotish. You were right, I was trying to open a notebook within the /ds_shared_drive which caused the read only error.
Now I opened a new notebook in the directory /home/desktop0/comp/python.ipynb. I’m getting the directory not found error. Can you please help me which directory should I use to run my notebook?

aman_patkar · April 30, 2021, 5:34am

I am having this issue while submittng/

jyotish · April 30, 2021, 5:36am

Hello @thanish, this looks like a bug on our end . As a temporary workaround, can you start jupyter notebook inside /home/desktop0/comp/ directory and try again?

It should be something like

cd /home/desktop0/comp
jupyter-notebook

We will patch the bug as soon as possible.

thanish · April 30, 2021, 5:43am

That worked. Thanks Made my first Random submission

jyotish · April 30, 2021, 5:47am

Hello @aman_patkar

Do you have multiple instances of jupyter notebook running? Can you stop any additional jupyter notebook servers you started and try again?

This error is displayed when there is no active notebook on jupyter. We consider the jupyter session that was last started and take the active notebook with the most recent changes.

aman_patkar · April 30, 2021, 5:52am

I checked and it is only single notebook in which i am makiing submission.

jyotish · April 30, 2021, 5:57am

Hello @aman_patkar

The screenshot you shared gives list of notebooks/terminals running in the current jupyter server. You can run jupyter-notebook list to get a list of active jupyter servers.

tymur_prorochenko · May 2, 2021, 1:42pm

hello everyone!

I’m using linux VM, and have troubles with submission.
my steps:

launch jupyter lab inside my working folder:

image1293×279 158 KB
specify path variables as follows:

image794×164 46 KB
submit:

image1005×481 139 KB
however submission fails:

image797×178 8.88 KB

long error log in details, and:

5.so what do I do if I have this file:

ashivani · May 2, 2021, 2:12pm

Hi @tymur_prorochenko,

The error is because you have hard coded the AICROWD_DATASET_PATH and other similar variable. Please use environment variable to specify the path of the files. You can use the below code to make your submission work.

AICROWD_DATASET_PATH = os.getenv("DATASET_PATH", "/home/desktop0/competition/validation.csv")
AICROWD_PREDICTIONS_PATH = os.getenv("PREDICTIONS_PATH", "/home/desktop0/competition/prediction.csv")

Ayush

tymur_prorochenko · May 3, 2021, 10:07pm

Thank you, @ashivani ! This did help, and I made several successful submissions.

Now I have another problem and I wonder if you could maybe point where to search for the problem.
I’m getting AttributeError: '_name_of_the_model' object has no attribute 'classes_' during the inference stage. This error may happen if the classifier is not fit, however it is strange since the notebook works fine before submission, and the model is loaded from assets.

jyotish · May 3, 2021, 10:17pm

Hello @tymur_prorochenko

Based on the discussion from this thread this happens when there is a mismatch in the versions of scikit-learn, numpy used to save and load the models.

Can you try adding -U flag in your pip install command (so that the packages used on the evaluation image are upgraded instead of getting skipped)?

For example

!pip install -U numpy scikit-learn

It would be good to pin the versions as well if you can so that the exact same version of the packages are used during evaluation for your submission.