4.3.0

1. IterativeAlgorithmController to monitor training progress and early stop.
2. Refactor learning algorithm hyperparameters with record Options
3. Refactor partitioning clustering algorithms
4. Rename CLARANS to KMedoids
5. SpectralClustering for sparse feature count matrix with cosine similarity.
6. Move lambda function interface into smile.util.function package
7. Update scala code to Scala 3 syntax
8. block quote representation for LLM reasoning block
9. Add --host and --port options for inference endpoint
10. Add --line option to render inference output JSON objects in a line-by-line fashion
11. gradle build scripts

4.2.0

1. DataFrame is fully re-designed. See Data Processing for a tutorial of the new API.
2. Random Forest and Gradient Boosting are 30% faster.
3. LLM frontend supports MathML.
4. Add smile.datasets package including many open benchmark datasets and training formula.
5. Bug fixes.

4.1.0

1. Refactor Graph API
2. Add Prim's algorithm for MST
3. Add Held-Karp algorithm, Christofides algorithm, branch-and-bound TSP, 2-opt algorithm, and various heuristics algorithms for TSP
4. Add random projection tree and random projection forest
5. Add nearest neighbor descent algorithm
6. UMAP is 6X faster on large data
7. Refactor manifold algorithm APIs
8. Add Pairing Heap data structure
9. Bug fixes

4.0.0

1. smile-deep package for deep learning
2. Llama 3.1 model in Java
3. Native Java implementation of tiktoken tokenizer
4. EfficientNet model for image classification
5. smile-shell has built-in training and inference functionalities, including streaming APIs.
6. smile-serve is an LLM inference server with OpenAI-compatible APIs and fully functional frontend.
7. Gradient boost is 10X faster on very large dataset
8. Code refresh to leverage latest Java features.
9. Various plain value classes are converted to records
10. Smile shell for Java and Kotlin
11. Java 21 required

3.1.1

Bug fixes.

3.1.0

• Declarative Data Visualization for Java

3.0.3

Bug fixes.

3.0.2

1. Minor bug fixes.
2. Improve flaky tests.

3.0.1

1. Remove XStream dependency as it exposes many vulnerabilities
2. Bug fixes with ICA, MCC, Shap Value, etc.

3.0.0

1. Switch to dual license model to meet open-source projects (GPL) and the development and distribution needs of commercial distributors (such as OEMs, ISVs and VARs).
2. Java Module friendly with auto module name
3. Redesigned feature engineering packages (missing value imputation, transform, selection, extraction, importance)
4. One-class SVM
5. Isolation forest
6. Feature Hashing
7. One-way ANOVA
8. BigMatrix supporting more than 2 billion elements
9. Latin hypercube sampling
10. CLI supports training, batch prediction, endpoint, etc.
11. Bug fixes