Managing Dependencies
JFrog ML supports a variety of Python frameworks to manage model dependencies.
Supported Python Versions
When building and managing your Python projects, different tools have varying levels of support for Python versions. Below is a summary of the supported Python versions for each tool:
- Poetry supports Python versions: 3.8 - 3.11
- Conda supports Python versions: 3.8 - 3.11
- requirements.txt (pip) supports only Python 3.9
Using Poetry with JFrogML
JFrogML uses Poetry version 1.8.3.
JFrog ML supports poetry.lock
files as long as they're under the same scope as the pyproject.toml
file.
Model Directory Structure:
qwak_based_model/
βββ main/
βββββ pyproject.toml
βββββ poetry.lock
βββ tests/
Both files pyproject.toml
and poetry.lock
will be used by Poetry while executing poetry install cmd.
Example Project Setup
[tool.poetry]
name = "example-project"
version = "0.1.0"
description = "Example project for production and development"
authors = ["Your Name <[email protected]>"]
[tool.poetry.dependencies]
python = ">=3.11,<3.12"
scipy = "^1.7"
scikit-learn = "^0.24"
catboost = "^1.0"
[tool.poetry.dev-dependencies]
qwak-sdk = "*"
[build-system]
requires = ["poetry-core>=1.0.0"]
build-backend = "poetry.core.masonry.api"
When specifying dependencies in Poetry, using * as the version for qwak-sdk instructs Poetry to install the latest available version of the qwak-sdk package. This approach ensures that your project always utilizes the most recent features and fixes. However, it's important to consider the implications of automatically adopting new versions, as they may introduce breaking changes or compatibility issues. For more controlled dependency management, consider the following alternatives:
qwak-sdk = "^0.5.61"
: This specifies that Poetry should install a version of qwak-sdk that is at least as new as 0.5.61 but less than the next major version (1.0.0). It allows for updates that include backwards-compatible features and fixes. This approach balances the benefits of receiving updates with the safety of avoiding major changes that could break your project.qwak-sdk = "0.5.61"
: This pins qwak-sdk to a specific version, ensuring that your project will always use version 0.5.61 of the SDK. This is the safest option if your project depends on the specific behavior of this version, as it eliminates the risk of unexpected changes due to updates. However, it also means that you will not automatically benefit from new features or fixes introduced in later versions.
The
qwak-sdk
dependency is included only in thedev
section, as it's needed for local development but not for remote builds. When you run theqwak models build
command, the SDK version you used locally will be automatically included in the remote environment.
Using Conda with JFrogML
JFrogML uses Conda version 24.7.1
Model Directory Structure
qwak_based_model/
βββ main/
βββββ conda.yml
βββββ ...
βββ tests/
The conda.yml
file can be placed at the root level alongside main
or within itβboth structures work equally well.
Example Project Setup
To get started, hereβs a basic conda.yml
setup:
name: example_conda_model
channels:
- defaults
- conda-forge
dependencies:
- python=3.11
- scipy
- scikit-learn
- catboost
- pip:
- # additional pip dependencies
Thereβs no need to manually add
qwak-sdk
to the environment. JFrogMLβs build process includes it automatically based on your local version.
.qwakignore file
Occasionally, we may want to exclude a file from the JFrog ML build but keep it in the repository with the model code. In such cases, we should add the .qwakignore
file to the root directory of our project.
In the file, we define the patterns to match files to exclude from the model build.
For example, suppose we have the following file structure:
.qwakignore
main/
__init__.py
model.py
README.md
tests/
test_model.py
research/
paper_a.pdf
paper_b.pdf
if we want to exclude the entire research
directory and the README.md
file from the build, our .qwakignore
file may contain:
research
README.md
Hidden files
By default, JFrog ML disregards hidden files. Hidden files are files or directories whose names start with a dot (.) in Unix-like operating systems, or they may have the "Hidden" attribute set in Windows. These files are typically used to store configuration data or hold temporary information.
Suppose you have a directory with files and subdirectories, including a hidden file named
.config_file
. JFrog ML, following its default behavior, will exclude this file from processing when triggering a remote build.
Incorporating Python Dependencies from .whl
Files
.whl
FilesQwak facilitates the use of Python dependencies packaged as .whl
files through requirements.txt
and conda.yaml
for managing dependencies. It's important to note that Poetry does not support dependencies from .whl
files.
- Preparing Your
.whl
Files:
First, ensure your .whl
file(s) are either uploaded with your model code or fetched from external storage. For instructions on uploading additional dependencies, refer to the Qwak CLI documentation (qwak models build --help
). Below is an example directory structure for your model, where main
is uploaded by default and the dep
directory, containing the pandas dependency in a .whl
file, is included via the --dependency-required-folders dep
option in the Qwak command.
The wheel file has to be uploaded as part of additional dependencies folder, and not as part of the main model folder.
/qwak/model_dir/
.
βββ main # Main directory containing core code
β βββ __init__.py # An empty file that indicates this directory is a Python package
β βββ model.py # Defines the Credit Risk Model
β βββ conda.yaml # Conda environment configurationdata
β
βββ dep # Additional dependency directory added with --dependency-required-folders
β βββ pandas-2.2.1-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
β
βββ tests # Empty directory reserved for future test
β βββ ... # Future tests
|
βββ
- Configuring Dependency Management Files:
Conda: Include the .whl
file in your conda.yaml as follows:
name: test_model
channels:
- defaults
- conda-forge
dependencies:
- python=3.9
- pip:
- "/qwak/model_dir/dep/pandas-2.2.1-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl"
Poetry
[tool.poetry]
name = "example-project"
version = "0.1.0"
description = "Example project for production and development"
authors = ["Your Name <[email protected]>"]
[tool.poetry.dependencies]
python = "^3.9"
scipy = "^1.7"
scikit-learn = "^0.24"
catboost = "^1.0"
pandas = { path = 'dep/pandas-2.2.1-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl' }
[tool.poetry.dev-dependencies]
qwak-sdk = "*"
[build-system]
requires = ["poetry-core>=1.0.0"]
build-backend = "poetry.core.masonry.api"
Requirements.txt: Directly reference the .whl
file path relative to the requirements file location:
# requirements file located in main model folder
./../deps/wheel_test-0.1-py3-none-any.whl
# requirements file located in model dir
./deps/wheel_test-0.1-py3-none-any.whl
- Using the Dependency in Your Code:
Once the dependency is properly configured, you can import and use it in your Python code as usual:
import pandas as pd
Updated about 2 months ago