Custom versioning of ML models.
Mixver is a versioning tool for ML models built enterily with Python. Its main capability is the ease of use and integration with different storage platforms such as AWS.
Install the package using pip:
pip install mixverThis example shows how to version ML models locally. First, we need to initialize the storage in the desired path.
from mixver.storages.local_storage import LocalStorage
storage = LocalStorage(storage_path="local_folder/storage")Once the storage it's initialized, we can push any model we want with its corresponding metadata.
from sklearn.linear_model.LogisticRegression
model = LogisticRegression()
metadata = {"accuracy": 0.8}
name = "test_model"
tags = ["latest", "training"]
storage.push(model, name=name, metadata=metadata, tags=tags)With this code we are versioning a LogisticRegression model that achieved an accuracy of 0.8 during training. Note that we also add two tags to it, the "latest" for representing the latest trained model, and the "training" tag to represent the model which best performed on the training data. Thus, we can retrieve the model which has this any of this tags assigned.
The following code retrieves the model with the tag "latest", which can be the newest model created. Besides, we can also retrieve the model with the "training" tag, which is the model that best performed in the training data.
On the other hand, we can also retrieve a model if we pass its name and version. In the case, we don't pass any specific version, it will return the newest version of the model
# Get the model with "latest" tag
latest_model = storage.pull(tag="latest")
# Get the model with "training" tag
best_model_training = storage.pull(tag="training")
# Get the version 1 of "test_model"
model_version_1 = storage.pull(name="test_model", version="1")
# Get the latest version of "test_model"
model_latest_version = storage.pull(name="test_model")storage = LocalStorage(...)
storage.visualize()An output example would be the following:
- Show the models and their latest versions in the CLI
- Show the tags and their corresponding models in the CLI
- Add AWS S3 storage
- Add Google Drive storage
Distributed under the MIT License. See LICENSE.txt for more information.
Hector Lopez Almazan - @hectorLop_- lopez.almazan.hector@gmail.com