Deploying Machine Learning Models

Let’s consider the linear regression model we have previously trained to predict grades given study hours, and saved to a local file:

Code 
from urllib.request import urlopen
import joblib

MODEL_URL = "https://github.com/prof-rossetti/python-for-finance/raw/main/docs/notes/predictive-modeling/model-management/grades-linear-regression/model.joblib"
presaved_model = joblib.load(urlopen(MODEL_URL))

presaved_model
/opt/hostedtoolcache/Python/3.11.10/x64/lib/python3.11/site-packages/sklearn/base.py:348: InconsistentVersionWarning:

Trying to unpickle estimator LinearRegression from version 1.5.2 when using version 1.3.2. 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

LinearRegression()
In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

When we save this model to a local file, it doesn’t take much memory (in this case only 864 Bytes).

However in practice some of the more complicated models can have millions or billions of parameters, and saving them to file would take a lot more memory.

Unfortunately when we want to use these models in a production (user-facing) application, the production server might not have enough room to store the saved model. Or it might have “ephemeral” storage which would make it more difficult to save files to disk.

In these situations where saving it locally may consume significant disk space, we can alternatively upload the trained model to a cloud storage provider such as Amazon S3, or Google Cloud Storage (preferred). Cloud storage scales easily without the need to manage physical storage hardware.

Cloud Storage

Advantages of Cloud Storage

  1. Scalability & Availability:
    • Access Anywhere: Models stored in the cloud can be accessed from anywhere, enabling easy sharing and collaboration between team members across different locations.
    • Cross-Platform Access: The model can be accessed by other cloud services, tools, or applications without being limited to local hardware.
    • Redundancy & Backups: Cloud providers typically offer built-in redundancy and backup features, reducing the risk of data loss.
  2. Integration with Cloud Services:
    • Cloud storage integrates well with cloud-based machine learning pipelines, such as AWS SageMaker, Google AI Platform, or Azure ML, simplifying deployment and management.
    • Cloud storage can be directly linked to other services like automated retraining or model serving platforms, improving workflow efficiency.
  3. Storage and Compute Resources:
    • If the model is large, saving it locally may consume significant disk space. Cloud storage scales easily without the need to manage physical storage hardware.
  4. Security:
    • Cloud providers often offer advanced security features, such as encryption at rest and in transit, access control mechanisms, and detailed logging, providing a secure environment for sensitive models.

Disadvantages of Cloud Storage

  1. Cost:
    • Cloud storage incurs ongoing costs, especially for large models, frequent accesses, or long-term storage.
    • Data transfer costs may also apply if you frequently download the model from the cloud.
  2. Latency:
    • Accessing a model from the cloud can introduce latency compared to retrieving it from a local file, which could impact workflows that require fast or real-time model loading.
  3. Dependency on Internet Connectivity:
    • Cloud storage requires a stable internet connection. If connectivity is slow or unreliable, accessing or saving the model could be problematic.
  4. Security Risks:
    • While cloud providers offer strong security features, there is always a risk of data breaches or misconfigurations (e.g. incorrect access permissions), potentially exposing sensitive models.

When to Use Cloud Storage

  • Distributed teams need access to the model.
  • The model is part of a cloud-based pipeline (training, serving, etc.).
  • The model is large or frequently updated, making local storage inconvenient.
  • There is a need for built-in redundancy and backup.

When to Use Local Storage

  • The model is small or only used locally.
  • You want to avoid the cost of cloud storage.
  • Latency is critical and local access is faster.
  • The infrastructure doesn’t require cloud integration, or security risks are a concern.

In summary, cloud storage is beneficial for scalability, collaboration, and cloud-based workflows, but local storage can be faster and more cost-effective for simpler or smaller-scale operations.