Connect to our Discord channel
This knowledge challenge was designed for Leaderex and AI Expo, taking place on 3 and 4 September 2019, respectively. This competition is open to anyone and will remain open to allow the Zindi community to learn and test their skills.
Potholes have become a huge problem for most drivers. With the South African government spending over R22 billion over the past 3 years on pothole repair programs and the Automobile Association(AA) acknowledging more than 5% of road deaths to unmaintained road structure (potholes).
The objective of this challenge is to create a machine learning model to accurately predict the likelihood that an image contains a pothole.
The evaluation metric for this challenge is the Area Under the Curve.
The label is the likelihood that the image contains a pothole. Values can be from 0 to 1.
Your submission file should look like:
Models
Annotations
id label
AEJGkTGsvGnwBVQ .543254
AEPaZSgFfneYkLS 0
AEjDuKGztTuzjDC 1
config
│ └── config.yaml
├── Dockerfile
├── dvc.yaml
├── flaskapp.py
├── logs
│ └── logfile.log
├── main.py
├── mlflow.py
├── params.yaml
├── Procfile
├── __pycache__
│ └── mlflow.cpython-38.pyc
├── README.md
├── requirements.txt
├── research
│ ├── 01_data_ingestion.ipynb
│ ├── 02_prepare_base_model.ipynb
│ ├── 03_model_training.ipynb
│ ├── 04_model_evaluation.ipynb
│ └── trials.ipynb
├── scores.json
├── setup.py
├── setup.sh
├── src
│ ├── potholeClassifier
│ │ ├── components
│ │ │ ├── data_ingestion.py
│ │ │ ├── __init__.py
│ │ │ ├── model_evaluation.py
│ │ │ ├── model_training.py
│ │ │ └── prepare_base_model.py
│ │ ├── config
│ │ │ ├── configuration.py
│ │ │ └── __init__.py
│ │ ├── constants
│ │ │ ├── __init__.py
│ │ │ └── __pycache__
│ │ │ └── __init__.cpython-38.pyc
│ │ ├── entity
│ │ │ ├── config_entity.py
│ │ │ └── __init__.py
│ │ ├── __init__.py
│ │ ├── pipeline
│ │ │ ├── __init__.py
│ │ │ ├── stage_01_data_ingestion.py
│ │ │ ├── stage_02_prepare_base_model.py
│ │ │ ├── stage_03_model_training.py
│ │ │ ├── stage_04_model_evaluation.py
│ │ │ └── stage_05_prediction.py
│ │ ├── __pycache__
│ │ │ └── __init__.cpython-38.pyc
│ │ └── utils
│ │ ├── common.py
│ │ ├── _init__.py
│ │ └── __pycache__
│ │ └── common.cpython-38.pyc
│ └── potholeClassifier.egg-info
│ ├── dependency_links.txt
│ ├── PKG-INFO
│ ├── SOURCES.txt
│ └── top_level.txt
├── static
│ ├── CNN workings.png
│ └── mlops.png
├── template.py
└── templates
└── index.html
This project uses the MLOps Level 0: Manual Process.
We developed a custon CNN (1 input layer, 3 conv2D layers, a flatten layer, a dense layer & softmax layer) for this task. We used a keras-tuner to discover the best hyperparameters that gave us the best validation loss. The lowest validation loss, given compute constraints was 0.04 and the accuracy achieved was 97%.
Clone the repository
https://github.com/MartinKalema/MIIA-Pothole-Image-classification.git
Create a conda environment after opening the repository and activate it
conda create -n classifier python=3.8 -y
conda activate classifier
Install the requirements
pip install -r requirements.txt
This Project is connected to Dagshub so all my experiments are sent to dagshub and can be viewed on dagshub itself or on the mlflow platform integrated there.
MLflow is a production grade experiments tracker for managing end-to-end machine learning lifecycle. It helps with experiments tracking, packaging code into reproducible runs and sharing and deploying models.
Do not set the tracking uri using the line of code below. All experiments will be stored inside an auto generated folder called mlruns.
mlflow.set_tracking_uri()
Use the command below to view them in the mlflow web interface
mlflow ui
Connect your github project to your Dagshub account. Visit dagshub here https://dagshub.com
Add the URI, USERNAME && PASSWORD variables to your environment by running the commands below.
export MLFLOW_TRACKING_URI=https://dagshub.com/kalema3502/MIIA-Pothole-Image-classification.mlflow
export MLFLOW_TRACKING_USERNAME=kalema3502
export MLFLOW_TRACKING_PASSWORD=fb3845efcc3b2e46a4157b1d2c977a21e02dd16e
If these are not added to your working environment, the experiment data will be stored in your project root in a folder called mlruns.
Its a tool we used to automate our pipeline. Initialize dvc inside your project using the command below.
dvc init
Add the project pipelines to the dvc.yaml file, then run the command below.
dvc repro
To view the pipeline structure, use the command below
dvc dag
To run the unit tests, use the command below
pytest test/test_file_name.py
Login to the AWS console
Create an IAM user for deployment
The user should have EC2 & ECR access. The deployment steps are mentioned below,
Policy
Create EC2 repo to save docker image
save the URI: 566373416292.dkr.ecr.us-east-1.amazonaws.com/chicken
Create Ubuntu EC2 VM
Install Docker on the EC2 Machine
#optional
sudo apt-get update -y
sudo apt-get upgrade
#required
curl -fsSL https://get.docker.com -o get-docker.sh
sudo sh get-docker.sh
sudo usermod -aG docker ubuntu
newgrp docker
Configure EC2 as self-hosted runner
setting > actions > runner > new self hosted runner > choose os >
Setup github secrets
AWS_ACCESS_KEY_ID=
AWS_SECRET_ACCESS_KEY=
AWS_REGION = us-east-1
AWS_ECR_LOGIN_URI = demo >> 566373416292.dkr.ecr.ap-south-1.amazonaws.com
Press p or to see the previous file or, n or to see the next file
Are you sure you want to delete this access key?
Are you sure you want to delete this access key?
Are you sure you want to delete this access key?
Are you sure you want to delete this access key?