🏠 House Price Predictor – An MLOps Learning Project

Welcome to the House Price Predictor project! This is a real-world, end-to-end MLOps use case designed to help you master the art of building and operationalizing machine learning pipelines.

You'll start from raw data and move through data preprocessing, feature engineering, experimentation, model tracking with MLflow, and optionally using Jupyter for exploration – all while applying industry-grade tooling.

🚀 Want to master MLOps from scratch?
Check out the MLOps Bootcamp at School of DevOps to level up your skills.

📦 Project Structure

house-price-predictor/
├── configs/                # YAML-based configuration for models
├── data/                   # Raw and processed datasets
├── deployment/
│   └── mlflow/             # Docker Compose setup for MLflow
├── models/                 # Trained models and preprocessors
├── notebooks/              # Optional Jupyter notebooks for experimentation
├── src/
│   ├── data/               # Data cleaning and preprocessing scripts
│   ├── features/           # Feature engineering pipeline
│   ├── models/             # Model training and evaluation
├── requirements.txt        # Python dependencies
└── README.md               # You’re here!

🛠️ Setting up a Learning / Development Environment

To begin, ensure the following tools are installed on your system:

Python 3.11
Git
Visual Studio Code or your preferred editor
UV – Python package and environment manager
Docker Desktop or Podman Desktop

🚀 Preparing Your Environment

Fork this repo on GitHub.

Clone your forked copy:

# Replace xxxxxx with your GitHub username or org
git clone https://github.com/xxxxxx/house-price-predictor.git
cd house-price-predictor

Setup Python Virtual Environment using UV:

uv venv --python python3.11
source .venv/bin/activate

Install dependencies:
```
uv pip install -r requirements.txt
```

📊 Set up MLflow for Experiment Tracking

To track experiments and model runs:

cd deployment/mlflow
docker compose -f mlflow-docker-compose.yml up -d
docker compose ps

🐧 Using Podman? Use this instead:

podman compose -f mlflow-docker-compose.yml up -d
podman compose ps

Access the MLflow UI at http://localhost:5555

📒 Using JupyterLab (Optional)

If you prefer an interactive experience, launch JupyterLab with:

uv python -m jupyterlab
# or
python -m jupyterlab

🔁 Model Workflow

🧹 Step 1: Data Processing

Clean and preprocess the raw housing dataset:

python src/data/run_processing.py   --input data/raw/house_data.csv   --output data/processed/cleaned_house_data.csv

🧠 Step 2: Feature Engineering

Apply transformations and generate features:

python src/features/engineer.py   --input data/processed/cleaned_house_data.csv   --output data/processed/featured_house_data.csv   --preprocessor models/trained/preprocessor.pkl

📈 Step 3: Modeling & Experimentation

Train your model and log everything to MLflow:

python src/models/train_model.py   --config configs/model_config.yaml   --data data/processed/featured_house_data.csv   --models-dir models   --mlflow-tracking-uri http://localhost:5555

🐳 Docker Image Naming Convention

To avoid issues with Docker reusing incorrect layers when building images for different services, this project uses unique image names for each service rather than relying solely on tags within a single repository.

FastAPI Backend Service: The Docker image for the FastAPI backend is named: [your-docker-id]/house-price-predictor-service
Streamlit Frontend Service: The Docker image for the Streamlit frontend is named: [your-docker-id]/house-price-predictor-ui

This convention ensures that Docker treats each service's image as distinct, preventing build and caching conflicts that can arise when multiple services share the same base image name with only different tags.

🛠️ Recovery Steps for Docker Build Issues

If you encounter persistent issues with Docker reusing old layers or unexpected behavior after builds, you can try the following recovery steps:

Clear Docker Build Cache:
```
docker builder prune --all
```
Comprehensive Docker System Cleanup: (Use with caution, this removes all unused Docker data)
```
docker system prune --all --volumes --force
```
Delete Remote Repositories (if applicable): If you have pushed images with problematic layers to Docker Hub or another registry, you might need to manually delete the affected repositories from the registry's web interface. To do this on Docker Hub, navigate to each service's repository, then go to Settings, and you will find the delete option there. This ensures that docker compose up or docker pull commands do not fetch the old, problematic images.

Dagger CI Pipeline Setup

The Dagger pipeline runs within a carefully configured, self-contained Docker environment to ensure consistent and secure execution. This is necessary because Dagger is not directly compatible with the Nix-based devenv shell. The devenv setup is used only as a convenient wrapper for scripts that manage this containerized workflow.

The `dagger-runner` Environment

All pipeline logic executes inside the dagger-runner container, built from Dockerfile.dagger-runner. This Dockerfile is specifically crafted to create a secure Docker-in-Docker environment for a non-root user. Here is how it works:

Docker CLI Installation: The docker-ce-cli package is installed by the root user from Docker's official repository. This is only the first step and is not sufficient on its own.
Non-Root User and Path: For security, the container switches to a non-root user named dockeruser. After the switch, the container's PATH is explicitly updated to include /usr/bin, ensuring the docker executable is available to this new user.
Docker Socket Permissions (GID Matching): The most critical piece is granting the dockeruser permission to use the host's Docker socket. This is achieved by matching the group ID (GID):
- The devenv script dagger_build_runner_image command dynamically gets the GID of the docker group from the host machine.
- This GID is passed into the docker build process as a build argument.
- The Dockerfile then creates a docker group inside the container with that specific GID and adds dockeruser to it.

This GID matching gives the non-root dockeruser the exact permissions required to interact with the mounted Docker socket, creating a secure and functional Docker-in-Docker setup.

Dagger Pipeline Internals

When the pipeline runs, Dagger performs several actions to ensure security and efficiency:

Secret Management: When publishing the Docker image, credentials like DOCKERHUB_TOKEN are passed to Dagger as secrets. Dagger ensures these secrets are encrypted before being transmitted to the engine and only makes them available to the specific commands that need them. They are never stored in the final image or logs.
Efficient Image Pushing: Before uploading an image layer, Dagger first sends a HEAD request to the container registry (e.g., Docker Hub). This checks if the layer with the same digest already exists. If it does, Dagger skips the upload for that layer, saving significant time and bandwidth.

Running and Debugging the Pipeline

After entering the development environment with devenv shell, the custom scripts defined in devenv.nix are directly available as commands.

Building and Running

Build the Runner Image: First, build the dagger-runner container image. This only needs to be done once, or whenever you change Dockerfile.dagger-runner.
```
dagger_build_runner_image
```
Execute the Pipeline: Run the entire MLOps pipeline. This command starts the dagger-runner container and executes the dagger_pipeline.py script inside it. The underlying run_dagger_pipeline.sh script is configured to output logs in plain text via the --progress=plain flag.
```
dagger_run_pipeline
```

Viewing Logs

Tailing Runner Logs: The run_dagger_pipeline.sh script runs the container in the foreground, so all output is streamed directly to your terminal. If you were to modify it to run in detached mode (-d), you could tail the logs using the container name specified in the script (dagger-runner):
```
docker logs -f dagger-runner
```
Dagger Engine Tracing: The run_dagger_pipeline.sh script correctly configures the Dagger engine to export OpenTelemetry traces on port 6060. While traces are generated, we have not yet successfully configured a local Jaeger instance to consume and visualize them. For a reliable and out-of-the-box observability solution, the recommended approach is to use Dagger Cloud.

Docker Compose Notes

Here are some important notes regarding the docker-compose.yaml configuration:

Environment Variable Overrides: Docker Compose uses .env files by default for environment variable overrides. Shell variables will take precedence over those defined in .env files.
- To specify a different environment file, use: docker compose --env-file .env_dev up
Viewing Substituted Configuration: To see the Docker Compose file after all variable substitutions have been applied, you can use:
```
docker compose config
```
Build Section in Production: The build section in docker-compose.yaml is primarily for local development. It can cause issues with image naming if not handled carefully, as it might build images with invalid names locally (e.g., ${DOCKER_USER_ID}:${SERVICE_VERSION}). For production deployments, it's generally recommended to remove the build section from your compose file and instead use pre-built images from a registry. For this reason, it's a good practice to construct the full image URL (including the registry and repository) directly within your .env_xxxx files for different environments.

Building FastAPI and Streamlit

🐉💣 Important Note on Docker Compose and Image Layers 💣🐉

What NOT to do: Initially, this project attempted to use a single base image name (e.g., [your-docker-id]/house-price-predictor) and differentiate services solely by tags (e.g., house-price-predictor:service-latest and house-price-predictor:ui-latest).

What happens if you do: Docker's caching mechanism can get confused when multiple services share the same base image name, even with different tags. This often leads to Docker reusing incorrect layers, resulting in both services running the same application (e.g., both the backend and UI containers might end up running the Streamlit frontend).

Here's an example of what you might see in your Docker Compose logs if this issue occurs, where both house-price-predictor-service-1 and house-price-predictor-ui-1 are running the Streamlit UI:

[+] Running 4/4
 ✔ house-price-predictor-ui                                         Built                                                                                                                                                                                                                                                                      0.0s
 ✔ house-price-predictor-service                                    Built                                                                                                                                                                                                                                                                      0.0s
 ✔ Container house-price-predictor-house-price-predictor-service-1  Recreated                                                                                                                                                                                                                                                                  0.0s
 ✔ Container house-price-predictor-house-price-predictor-ui-1       Recreated                                                                                                                                                                                                                                                                  0.0s
Attaching to house-price-predictor-service-1, house-price-predictor-ui-1
house-price-predictor-service-1  | 
house-price-predictor-service-1  | Collecting usage statistics. To deactivate, set browser.gatherUsageStats to false.
house-price-predictor-service-1  | 
house-price-predictor-service-1  | 
house-price-predictor-service-1  |   You can now view your Streamlit app in your browser.
house-price-predictor-service-1  | 
house-price-predictor-service-1  |   URL: http://0.0.0.0:8501
house-price-predictor-service-1  | 
house-price-predictor-ui-1       | 
house-price-predictor-ui-1       | Collecting usage statistics. To deactivate, set browser.gatherUsageStats to false.
house-price-predictor-ui-1       | 
house-price-predictor-ui-1       | 
house-price-predictor-ui-1       |   You can now view your Streamlit app in your browser.
house-price-predictor-ui-1       | 
house-price-predictor-ui-1       |   URL: http://0.0.0.0:8501
house-price-predictor-ui-1       |

To mitigate this, we've adopted unique image names for each service (house-price-predictor-service and house-price-predictor-ui). If you encounter issues with stale layers, consider using docker compose build --no-cache or docker system prune -a (use with caution as this removes all unused Docker data) to ensure a clean build.

🐉💣 Cache collisions between the UI and the Service docker builds

The last . in the docker build command is the build context. Passing in . changes the context based on where one runs streamlit_app/build.sh. If run from the project root, the build context is the entire project. If run from streamlit_app folder, its restricted to contents of just that folder. Run docker builder prune --all to clear the build cache. cd streamlit_app to build the UI docker image. Run dive celestialseeker/house-price-predictor-ui:v1.0.7 to double check the UI image was built right. 😓

TODO: Change the Dockerfile to use a builder and uv. The in stage 2, create a new image and copy just the uv dependencies without all the build environment baggage ⚓️.

The code for both the apps are available in src/api and streamlit_app already. To build and launch these apps

Add a Dockerfile in the root of the source code for building FastAPI
Add streamlit_app/Dockerfile to package and build the Streamlit app
Add docker-compose.yaml in the root path to launch both these apps. be sure to provide API_URL=http://fastapi:8000 in the streamlit app's environment.

Once you have launched both the apps, you should be able to access streamlit web ui and make predictions.

You could also test predictions with FastAPI directly using ./make.py test.
make.py is used to make it easier to run Docker commands to interact with House Price Predictor FastAPI app.

Makefile uses tabs (not spaces) and I found this made it harder to use. My VSCode setup uses spaces instead of tabs!

curl -X POST "http://localhost:8000/predict" \
-H "Content-Type: application/json" \
-d '{
  "sqft": 1500,
  "bedrooms": 3,
  "bathrooms": 2,
  "location": "suburban",
  "year_built": 2000,
  "condition": fair
}'

Be sure to replace http://localhost:8000/predict with actual endpoint based on where its running.

☸️ Deploying to Kubernetes (Kind)

This section guides you through deploying the house-price-predictor-service (FastAPI backend) and house-price-predictor-ui (Streamlit frontend) to a local Kubernetes cluster using Kind.

🚀 Kind Cluster Setup

First, ensure you have a Kind cluster running. If not, you can set one up using the following commands:

git clone https://github.com/initcron/k8s-code.git
cd k8s-code/helper/kind/
kind create cluster --config kind-three-node-cluster.yaml

👁️ Installing Kube-ops-view (Optional)

To visualize your cluster's state, you can install kube-ops-view:

cd ~
git clone https://github.com/schoolofdevops/kube-ops-view
kubectl apply -f kube-ops-view/deploy/.

🚢 Deploying the Services

The Streamlit UI is configured to communicate with the FastAPI service using the internal Kubernetes service name. The API_URL for the UI was modified to http://home-price-predictor-service:8000.

Here are the Kubernetes manifests to deploy both services. You can save these as deployment.yaml and service.yaml in your deployment/kubernetes/ directory, or apply them directly.

FastAPI Backend (house-price-predictor-service)

# deployment/kubernetes/house-price-predictor-service-deployment.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: house-price-predictor-service
  labels:
    app: house-price-predictor-service
spec:
  replicas: 1
  selector:
    matchLabels:
      app: house-price-predictor-service
  template:
    metadata:
      labels:
        app: house-price-predictor-service
    spec:
      containers:
      - name: house-price-predictor-service
        image: [your-docker-id]/house-price-predictor-service:latest # Replace with your actual image
        ports:
        - containerPort: 8000
---
# deployment/kubernetes/house-price-predictor-service-service.yaml
apiVersion: v1
kind: Service
metadata:
  name: house-price-predictor-service
spec:
  selector:
    app: house-price-predictor-service
  ports:
    - protocol: TCP
      port: 8000
      targetPort: 8000
  type: ClusterIP

Streamlit Frontend (house-price-predictor-ui)

# deployment/kubernetes/house-price-predictor-ui-deployment.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: house-price-predictor-ui
  labels:
    app: house-price-predictor-ui
spec:
  replicas: 1
  selector:
    matchLabels:
      app: house-price-predictor-ui
  template:
    metadata:
      labels:
        app: house-price-predictor-ui
    spec:
      containers:
      - name: house-price-predictor-ui
        image: [your-docker-id]/house-price-predictor-ui:latest # Replace with your actual image
        ports:
        - containerPort: 8501
        env:
        - name: API_URL
          value: http://house-price-predictor-service:8000 # Internal Kubernetes service name
---
# deployment/kubernetes/house-price-predictor-ui-service.yaml
apiVersion: v1
kind: Service
metadata:
  name: house-price-predictor-ui
spec:
  selector:
    app: house-price-predictor-ui
  ports:
    - protocol: TCP
      port: 80
      targetPort: 8501
  type: LoadBalancer # Use LoadBalancer for external access in Kind

Apply the Manifests

To deploy both services to your Kind cluster, apply these manifests:

kubectl apply -f deployment/kubernetes/house-price-predictor-service-deployment.yaml
kubectl apply -f deployment/kubernetes/house-price-predictor-service-service.yaml
kubectl apply -f deployment/kubernetes/house-price-predictor-ui-deployment.yaml
kubectl apply -f deployment/kubernetes/house-price-predictor-ui-service.yaml

After applying, you can check the status of your deployments and services:

kubectl get deployments
kubectl get services

To access the Streamlit UI, find the external IP or port of the house-price-predictor-ui service (if using LoadBalancer in Kind, it might be localhost on a specific port).

Errors with Joblib

The version of the joblib library in the Docker files should match the version used to create the model pkl files.

Run pip list to get a list of all the installed packages and match versions against requirements.txt

/usr/local/lib/python3.11/site-packages/sklearn/base.py:318: UserWarning: Trying to unpickle estimator DecisionTreeRegressor from version 1.6.1 when using version 1.2.2. This might lead to breaking code or invalid results. Use at your own risk. For more info please refer to:
https://scikit-learn.org/stable/model_persistence.html#security-maintainability-limitations
  warnings.warn(
Traceback (most recent call last):
  File "/app/inference.py", line 11, in <module>
    model = joblib.load(MODEL_PATH)
            ^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/joblib/numpy_pickle.py", line 658, in load
    obj = _unpickle(fobj, filename, mmap_mode)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/joblib/numpy_pickle.py", line 577, in _unpickle
    obj = unpickler.load()
          ^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/pickle.py", line 1213, in load
    dispatch[key[0]](self)
  File "/usr/local/lib/python3.11/site-packages/joblib/numpy_pickle.py", line 415, in load_build
    self.stack.append(array_wrapper.read(self))
                      ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/joblib/numpy_pickle.py", line 252, in read
    array = self.read_array(unpickler)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/joblib/numpy_pickle.py", line 152, in read_array
    array = pickle.load(unpickler.file_handle)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ModuleNotFoundError: No module named 'numpy._core'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/usr/local/bin/uvicorn", line 8, in <module>
    sys.exit(main())
             ^^^^^^
  File "/usr/local/lib/python3.11/site-packages/click/core.py", line 1442, in __call__
    return self.main(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/click/core.py", line 1363, in main
    rv = self.invoke(ctx)
         ^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/click/core.py", line 1226, in invoke
    return ctx.invoke(self.callback, **ctx.params)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/click/core.py", line 794, in invoke
    return callback(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/uvicorn/main.py", line 410, in main
    run(
  File "/usr/local/lib/python3.11/site-packages/uvicorn/main.py", line 578, in run
    server.run()
  File "/usr/local/lib/python3.11/site-packages/uvicorn/server.py", line 61, in run
    return asyncio.run(self.serve(sockets=sockets))
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/asyncio/runners.py", line 190, in run
    return runner.run(main)
           ^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/asyncio/runners.py", line 118, in run
    return self._loop.run_until_complete(task)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/asyncio/base_events.py", line 654, in run_until_complete
    return future.result()
           ^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/uvicorn/server.py", line 68, in serve
    config.load()
  File "/usr/local/lib/python3.11/site-packages/uvicorn/config.py", line 473, in load
    self.loaded_app = import_from_string(self.app)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/uvicorn/importer.py", line 21, in import_from_string
    module = importlib.import_module(module_str)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/importlib/__init__.py", line 126, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<frozen importlib._bootstrap>", line 1204, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1176, in _find_and_load
  File "<frozen importlib._bootstrap>", line 1147, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 690, in _load_unlocked
  File "<frozen importlib._bootstrap_external>", line 940, in exec_module
  File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
  File "/app/main.py", line 3, in <module>
    from inference import predict_price, batch_predict
  File "/app/inference.py", line 14, in <module>
    raise RuntimeError(f"Error loading model or preprocessor: {str(e)}")
RuntimeError: Error loading model or preprocessor: No module named 'numpy._core'

🧠 Learn More About MLOps

This project is part of the MLOps Bootcamp at School of DevOps, where you'll learn how to:

Build and track ML pipelines
Containerize and deploy models
Automate training workflows using GitHub Actions or Argo Workflows
Apply DevOps principles to Machine Learning systems

🔗 Get Started with MLOps →

🤝 Contributing

We welcome contributions, issues, and suggestions to make this project even better. Feel free to fork, explore, and raise PRs!

Using Devenv

List all installed python packages

pip list

Happy Learning!
— Team School of DevOps

Name		Name	Last commit message	Last commit date
.github/workflows		.github/workflows
configs		configs
data		data
deployment		deployment
models/trained		models/trained
notebooks		notebooks
src		src
streamlit_app		streamlit_app
.env		.env
.gitignore		.gitignore
Dockerfile		Dockerfile
Dockerfile.dagger-runner		Dockerfile.dagger-runner
LICENSE		LICENSE
README.md		README.md
dagger_pipeline.py		dagger_pipeline.py
devenv.lock		devenv.lock
devenv.nix		devenv.nix
devenv.yaml		devenv.yaml
docker-compose.yaml		docker-compose.yaml
make.py		make.py
requirements.txt		requirements.txt
run_dagger_pipeline.sh		run_dagger_pipeline.sh

Latest commit

History

Folders and files

Repository files navigation