open-solution-mapping-challenge

Open solution to the CrowdAI Mapping Challenge

Goal

Deliver open source, ready-to-use and extendable solution to this competition. This solution should - by itself - establish solid benchmark, as well as provide good base for your custom ideas and experiments.

Usage: Fast Track

1. clone this repository: git clone https://github.com/neptune-ml/open-solution-mapping-challenge.git
2. install requirements
3. register to Neptune (if you wish to use it) login via:

$ neptune login

4. download/upload competition data and change data-related paths in the configuration file neptune.yaml
5. Prepare the data:

$ neptune experiment run main.py -- prepare_metadata \
--train_data \
--valid_data \
--test_data

6. Put your competition API key in the configuration file
7. run experiment (for example via neptune):

$ neptune login
$ neptune experiment run \
main.py -- train_evaluate_predict --pipeline_name unet --chunk_size 5000

7. check your leaderboard score!

Usage: Detailed

1. clone this repository: git clone https://github.com/minerva-ml/open-solution-talking-data.git
2. install PyTorch and torchvision
3. install requirements: pip3 install -r requirements.txt
4. register to Neptune (if you wish to use it) login via:

$ neptune login

5. open Neptune and create new project called: Mapping Challenge with project key: MC
6. download the data from the competition site
7. upload the data to neptune (if you want to run computation in the cloud) via:

$ neptune data upload YOUR/DATA/FOLDER

8. change paths in the neptune.yaml .

  data_dir:               /path/to/data
  meta_dir:               /path/to/data
  masks_overlayed_dir:    /path/to/masks_overlayed
  experiment_dir:         /path/to/work/dir

9. run experiment:

   • local machine with neptune

   $ neptune login
   $ neptune experiment run \
   main.py -- train_evaluate_predict --pipeline_name unet --chunk_size 5000

   • cloud via neptune

   $ neptune login
   $ neptune experiment send --config neptune.yaml \
   --worker gcp-large \
   --environment pytorch-0.2.0-gpu-py3 \
   main.py -- train_evaluate_predict --pipeline_name solution_1 --chunk_size 5000

   • local pure python

   $ python main.py train_evaluate_predict --pipeline_name unet --chunk_size 5000

User support

There are several ways to seek help:

1. crowdai discussion is our primary way of communication.
2. You can submit an issue directly in this repo.

Contributing

1. Check CONTRIBUTING for more information.
2. Check issues and project to check if there is something you would like to contribute to.