Computational-Intelligence-Lab-2021

Repository for the Computational Intelligence Lab course project at ETH Zurich

How to reproduce results

Step 1: Clone the repository:

git clone https://github.com/AlessandroRuzzi/Computational-Intelligence-Lab-2021

Step 2: Copy new .env file and modify it by adding your environment variables:

cp .env.tmp .env 
vim .env 

Example of .env file:

COMET_API_KEY=Your Key
COMET_WORKSPACE=alessandroruzzi
KAGGLE_USERNAME=alessandroruzzi
KAGGLE_KEY=Your kaggle Key
GOOGLE_MAPS_API_KEY= leave this blank

Step 3: Create virtual environment called venv:

python -m venv venv

Step 4: Run the script to install modules, activate virtual environment, and install packages from requirements.txt:

source ./leonhard_init.sh

Step 5: Open the file configs/config.yaml and insert your eth username at line 10.

Step 6: Train the model on experiment 14 without Google API:

 bsub -W 24:00 -R "rusage[mem=64000, ngpus_excl_p=1]" -R "select[gpu_mtotal0>=10240]" python3 ./run.py +experiment=exp__f014

Step 7: Train the model on experiment 15 without Google API:

 bsub -W 24:00 -R "rusage[mem=64000, ngpus_excl_p=1]" -R "select[gpu_mtotal0>=10240]" python3 ./run.py +experiment=exp__f015

Step 8: Add the Google API key in the .env file (you can find it in the CMT3 submission's comment)

COMET_API_KEY=Your Key
COMET_WORKSPACE=alessandroruzzi
KAGGLE_USERNAME=alessandroruzzi
KAGGLE_KEY=Your kaggle Key
GOOGLE_MAPS_API_KEY= PUT THE API KEY HERE

Step 9: Train the model on experiment 14:

 bsub -W 24:00 -R "rusage[mem=64000, ngpus_excl_p=1]" -R "select[gpu_mtotal0>=10240]" python3 ./run.py +experiment=exp__f014

Step 10: Download the predictions from comet, you will find a file called submission.csv in the comet section called Assets & Artifacts, inside the folder others.

Step 11: Combine the predictions with the run_ensemble.py script (replace file1, file2 and file3 with the actual CSV files' paths):

 python3 ./run_ensemble.py file1,file2,file3 

How to run on Leonhard Cluster

0. Clone repository into cluster.

1. Copy new .env file and modify it by adding your environment variables:

cp .env.tmp .env 
vim .env 

2. Make leonhard_init.sh executable:

chmod +x leonhard_init.sh

3. Create virtual environment called venv:

python -m venv venv

4. Run leonhard_init.sh to install modules, activate virtual environment, and install packages from requirements.txt:

./leonhard_init.sh

5. Run model on a single GPU:

bsub -R "rusage[ngpus_excl_p=1]" ./run.py trainer.gpus=1

Download from cluster

The following command will download a zipped prediction into your current local folder.

scp your_username@login.leonhard.ethz.ch:/Computational-Intelligence-Lab-2021/preds/DATE/preds.zip .