PredOps

PredOps is a forecasting tool with the following features:

• Various forecast methods:
  • Time series statistical methods
  • Forest ensemble algorithms
  • Deep Learning approaches
• Ability to call multiple remote compute environments
• Back-testing
• Segmentation capabilities
• Time segmentation for a more precise forecast

The project will implement the following MLOps practices:

• Experiment tracking
• Performance monitoring
• Retraining
• CI/CD/CT

Installation

python3 -m venv venv
source venv/bin/activate
# On windows powershell
# .\venv\Scripts\Activate.ps1
python -m pip install --upgrade pip setuptools wheel
pip install -e .

Installation of the development environment:

python -m pip install -e ".[dev]"

How to run

CLI

Example: M5 Project

1. Download data

predops download-data m5a

2. Generate base file (with sample option)

predops generate-base-file m5a --sample

3. Train models: search, train and evaluate

# default parameters
predops train m5a --target Quantity
# advanced parameters
predops train m5a --target Quantity --number-predictions 28 --column-segment-groupby store_id --n-predictions-groupby 7

MLFlow

MLFlow allows to track the experiments

mlflow ui --backend-store-uri "sqlite:///mlflow.db"

References

• M5 Competition 4th place
• Best practices of developement fron madewithml repository