pyTigerGraph Release Notes

[1.6] - 2024-04-30

Release of pyTigerGraph version 1.6.

Added:

TigerGraph CoPilot. Added a new submodule for interacting with TigerGraph CoPilot, a framework for integrating Generative AI with TigerGraph.

[1.5] - 2023-09-25

Release of pyTigerGraph version 1.5.

Added:

Object-oriented schema definition and modifcation. Define graph schemas in native Python, without knowing GSQL.
gsql() handles some common error cases and raises an exception if they occur.

Changed:

Dataloaders that experience a parsing error due to missing/dirty data handle the error more gracefully.
Removed the use of pyTigerDriver for GSQL operations.
Various bug fixes.

[1.4] - 2023-05-16

Release of pyTigerGraph version 1.4.

Note: if you are using the Graph Data Science dataloaders, you must upgrade both the ML Workbench and pyTigerGraph to 1.4 at the same time. There is a incompatibility between v1.3 and v1.4 of pyTigerGraph and the corresponding ML Workbench versions.

Added:

Additional Query Management Support
- showQuery() returns the GSQL of a given query.
- getQueryMetadata() returns the metadata details about a query, such as input parameters and what is returned in PRINT statements.
- getRunningQueries() shows the statistics of queries currently running on the graph.
- abortQuery() aborts a selected query by ID or all queries on the graph.
Additional System Management Support
- ping() is a public API to check the health of the TigerGraph server.
- getSystemMetrics() monitors system metrics such as CPU and RAM usage.
- getQueryPerformance() returns real-time query performance statistics over a given period.
- getServiceStatus() returns the status of TigerGraph services specified in the request.
- rebuildGraph() rebuilds the graph immediately.
Built in Graph ML models and Trainer
- Various GraphSAGE models for vertex classification and regression, as well as link prediction.
- NodePiece MLP model for vertex classification.
- General purpose trainer to enable training of Graph ML models in a concise fashion.
Transforms
- PyGTemporalTransform to create a sequence of subgraphs for a given batch of data, in a temporal manner.
- NodePieceMLPTransform to transform a batch produced by a NodePiece dataloader into a batch that can be fed into a PyTorch multilayer perceptron.
Additional Dataloader Support
- SSL Support: two-way SSL encryption via Kerberos.
- Collaborative dataloaders: use dataloaders on multiple machines to pull batches from the same Kafka queue. Helpful for data distributed model training.
- Datetime support in dataloaders: Output DATETIME attributes from the database using the dataloaders. Exports as UNIX epoch timestamps.
- Optional distributed_query parameter in dataloaders if running on distributed database clusters. If set to True, installs the dataloader using the DISTRIBUTED keyword in the query heading. Useful for distributed database clusters.
- stop() function in dataloaders: Kill the query producing batches for the dataloader immediately. Helpful for stopping the production of batches sent to Kafka after breaking out of a training loop.

Changed:

Dataloader factory produces multiple dataloaders if filter_by is a list of different filters.
Improved the scalability of the NodePiece dataloader.

[1.3] - 2023-02-01

Release of pyTigerGraph version 1.3.

Added:

Two new dataloaders: HGTLoader and NodePiece
- The HGTLoader is a data loader that performs stratified neighbor sampling Heterogeneous Graph Transformer.
- NodePiece is a scalable graph embedding approach introduced in NodePiece: Compositional and Parameter-Efficient Representations of Large Knowledge Graphs
Callback functions to all dataloaders: write functions to process the batch in a background thread before it is passed into the training loop.

Changed:

Added better error messaging when REST requests are incorrect.
Template query support in the featurizer (requires TigerGraph Database 3.9+)
Data splitters automatically perform a schema change if needed to add attributes to the database.

Fixed:

Fixed how custom SSL certificates were handled when instantiating the GSQL client.

[1.2] - 2022-11-09

Release of pyTigerGraph version 1.2.

Added:

The Datasets class, a way to easily import standard datasets into a database instance.
The visualizeSchema function to visualize graph schemas.
Proper deprecation warnings.
Logging capabilities using native Python logging tools.
Ability to run asynchronous queries from runInstalledQuery()

Changed:

Many changes to the featurizer capability, including:
- Automatically selecting the correct version of a graph data science algorithm given your version of the database.
- Automatically creating the schema change necessary to run the algorithm and store the results to an attribute.
- If the algorithm is not already installed at runtime, and is included in the TigerGraph Graph Data Science Library, the algorithm will be installed automatically.
- Adding more supported algorithms, in categories such as similarity and topological link prediction.

[1.1] - 2022-09-06

Release of pyTigerGraph version 1.1.

Added:

TensorFlow support for homogeneous GNNs via the Spektral library.
Heterogeneous Graph Dataloading support for DGL.
Support of lists of strings in dataloaders.

Changed:

Fixed KeyError when creating a data loader on a graph where PrimaryIdAsAttribute is False.
Error catch if Kafka dataloader doesn’t run in async mode.
Refresh schema during dataloader instantiation and featurizer attribute addition.
Reduce connection instantiation time.
Reinstall query if it is disabled.
Confirm Kafka topic is created before subscription.
More efficient use of Kafka resources.
Allow multiple consumers on the same data.
Improved deprecation warnings.

[1.0] - 2022-07-11

Release of pyTigerGraph version 1.0, in conjunction with version 1.0 of the TigerGraph Machine Learning Workbench.

Added:

Kafka authentication support for ML Workbench enterprise users.
Custom query support for Featurizer, allowing developers to generate their own graph-based features as well as use our built-in Graph Data Science algorithms.

Changed:

Additional testing of GDS functionality
More demos and tutorials for TigerGraph ML Workbench, found here.
Various bug fixes.

[0.9.1] - 2022-06-21

Added new parameter, tgCloud for when connecting to a TigerGraph Cloud instance. Set to True if using a new TigerGraph Cloud

Changed:

Deprecated gcp parameter, as tgCloud supercedes this. Existing code will be compatible.

[0.9] - 2022-05-16

We are excited to announce the pyTigerGraph v0.9 release! This release adds many new features for graph machine learning and graph data science, a refactoring of core code, and more robust testing. Additionally, we have officially “graduated” it to an official TigerGraph product. This means brand-new documentation, a new GitHub repository, and future feature enhancements.

We are committed to keeping pyTigerGraph true to its roots as an open-source project. Check out the Contributing page and our GitHub Issues page if you want to help with pyTigerGraph’s development.

pyTigerGraph 0.9 was released on May 16th, 2022.

New Features

Graph Data Science Capability

Many new capabilities added for graph data science and graph machine learning.

Data loaders for training Graph Neural Networks in DGL and PyTorch Geometric
A "featurizer" to generate graph-based features
Metric trackers for precision, recall, and accuracy
Vertex and edge splitters for generation of train/validation/testing splits.

Other Changes

Documentation

We have moved the documentation to the official TigerGraph Documentation site and updated many of the contents with type hints and more descriptive parameter explanations.

Testing

There is now well-defined testing for every function in the package. A more defined testing framework is coming soon.

Code Structure

A major refactor of the codebase was performed. No breaking changes were made to accomplish this.