Predicting BTC/USD Prices with ARIMA Model using MLOps

Welcome to our project on predicting BTC/USD prices using ARIMA (AutoRegressive Integrated Moving Average) model!📈 This project demonstrates how we leverage machine learning for time series forecasting and deploy models using MLflow and Amazon SageMaker.🚀

Project Overview

In this project, we aimed to predict the future price of Bitcoin (BTC/USD) using historical data and real-time updates from the CryptoCompare API. The ARIMA model was selected for its performance in training and testing phases.

Setup AWS CLI configuration

We followed the setup instructions detailed in this repository, which provided comprehensive guidance on configuring AWS CLI for seamless integration with MLflow and Amazon SageMaker.

Step-by-Step Process

1. Training the Model with MLflow

   Run train.py to initiate training and track experiments with MLflow. This generates the MLflow folder containing experiment logs and artifacts, including the trained model.

2. Building and Pushing Docker Image

   Using the artifacts from MLflow, build a Docker image containing the trained model(go to the artifact directory):

   mlflow sagemaker build-and-push-container 
   

After, check AWS ECR repos list to get the image URI.

3. Deploy image to Sagemaker Run arimadeploy.py to ddeploy image to Sagemaker

4. Store fetched btc data to a s3 bucket Run fetch_bucket.py store a csv file from btc fetched data into a s3 bucket

5. Retrain the model if it's necesary Run test_model.py to predict in base of the data stored in the s3 bucket. If the R2 score is less than 0.80 the model it's retrained and saved as a new version into s3.

Team

This project was developed by: