0.8.0 beta

Changelog:

• hosted machine learning service w/ support for deep neural networks, including convolutional layers, gradient descent, and Adam optimization
• hosted linear algebra service with support for QR factorization, PLU factorization, and SVD
• new math package featuring automatic differentiation of numerical operators
• new App and Library classes to define microservices
• dynamic Chain support (i.e. Map and Tuple chains)
• Model class, in preparation for object-relational mapping
• Dynamic models with support for constructing method definitions based on compile-time instance constants
• bugfixes for einsum
• random and concatenation constructors for Dense tensors
• support for typing.Tuple and typing.Dict type hints
• performance improvements for multi-stage Ops
• benchmark tests
• support for rustc v1.60

SHA256 checksum: b7740786771dcf4ec8c453645245ab8eb420ef0f4f8293de8c48d03f00a7939d

0.7.0 beta

Changelog:

• New ML model architecture in tinychain.ml
• ML models now support AdamOptimizer
• New convenience functions for Tensor: log, mean, round, argmax, argsort, trigonometry functions
• New initializers for Tensor: random_normal, random_uniform
• Improved closure usability
• Improved cache performance
• Improved Stream usability
• Uniform hash function for State

SHA256 checksum: 27524a16a2f6f1cc6429206cfe82a5728c61ef490838ad047e4955f4ae794257

0.6.0 beta

Changelog:

• Map & Tuple now preserve item type information
• einsum now supports elided dimensions (e.g. einsum("...ij,...jk->...ik", ...))
• tc.ml.conv now supports evaluating a 1D or 2D convolution
• Number & Tensor now support trigonometric operations
• New linear algebra functions: diagonal, identity, set_diagonal, QR decomposition, Householder reflection
• New semantic version type Version
• Tensor now supports concatenate, flip, and reshape
• Improved Docker support
• Graph is now a subclass of Map rather than Cluster
• Tuple.fold method signature now matches Stream.fold

0.5.1 beta

Changelog:

• New Dockerfile, new install script, new installation documentation
• Tensor performance improvements
• First draft of deep neural net support w/ backpropagation
• Chain now supports a Tuple of subject Collections
• Reorganize client tests into their own package

SHA256 checksum: d4bcf2726563fca08ebca44e456c9520ac650090f2b3d4211a15c2c8173d62a0

Beginning with this release, only one binary is published, which includes a dependency on ArrayFire for hardware acceleration. Installing TinyChain without hardware acceleration is still supported using cargo install tinychain. A new Dockerfile and install script are provided to support automatic installation of all dependencies.

0.5.0 beta

Changelog:

• Filesystem cache now supports truly concurrent eviction with freqfs: https://github.com/haydnv/freqfs/
• Improved transaction lock performance under load
• Improved block lock performance under load
• Tensor bugfixes
• Improved anonymous class support in the Python client

SHA256 checksums:

• tinychain_gpu: 4b01a81a2e3311d3e3d2492fc62f1d56be6d16c246ad4afb212136f9fd71e0f9
• tinychain: 0f6bcaaf35ecba38262569f08607412c82b7839b81f14a8e05cee57d5575b80a

0.4.1 beta

Changelog:

• Graph database with support for edge & foreign key relationships
• Improved support for Stream data structure
• While loop
• String rendering support with Handlebars

SHA256 checksum (tinychain): 5b2506edf553f881d6aa2e6b76813c731bcf9007e015fafcaa64d81ed06ed7a5
SHA256 checksum (tinychain_gpu): 94162e68a2e50acffd518b10b48b8e4a7c9a5027f4141662bfd61bbcaf4e325d

0.4.0 beta

Changelog:

• add support for new Stream and Closure data structures
• define Dockerfile to build Tinychain with ArrayFire support automatically
• move implementations of Table.aggregate, Table.delete, and Table.update to the Python client
• improve usability of automatic replication of Collection ops

SHA256 checksum (tinychain): 9c7178b20ff625e36ff77d0f862fe44ff6d17e762181ba4e626c47192687bba3
SHA256 checksum (tinychain_gpu): 11caf7dcad431d6701f217d2a07609a6da5bb811e9c88ea0df7e1d87de78f6b9

0.3.2 beta

Changelog:

• improved Tensor performance
• improved replication performance
• stream encoding errors now propagate

SHA256 checksum (tinychain): d4b3b7b9c45b2c1341f0e73c2e17ee2d4c6dde226e20d43f77b0ed339b1c3d2b
SHA256 checksum (tinychain_gpu): 69ad05626d2e74ccf9f83850a51db7939c3ba929ffadcb77d0a87f2987481e7b

0.3.1 beta

Changelog:

• new SparseTensor data structure
• replication performance improvements, especially for multi-cluster replication
• assorted bugfixes

SHA256 checksum (tinychain): 29a166bef5b67cf9023fbab7e2cabd4b737be1a9f768480e81de49cb12ef853a
SHA256 checksum (tinychain_gpu): 5dd7c5fe5181bbf7b3c9730bda12c46e099fb2137fcfffbd18a15a50488db696

Please note that all Tensor features (including SparseTensor) require the ArrayFire library. There are instructions to install ArrayFire in the README: https://github.com/haydnv/tinychain

0.2.3 beta

Note: this release introduces Tinychain's automatic GPU acceleration, which requires the ArrayFire library. There are instructions in the "Getting Started" section of the README to install ArrayFire.

Changelog:

• Add preliminary support for the Tensor feature (dense Tensor only)
• Add update method to the public API of Table

SHA256 checksum (tinychain): 1e67ba5b281ba77b32702352d8840bb6ac30711a0e5e698c12ec640c3b70de5f
SHA256 checksum (tinychain_gpu): 29b4d0860477bc976ed1a528277932616eeab5fbf668baaf6bb1bbe77814a9cd