[Diff since v0.6.2](v0.6.2...v0.6.3)

**Merged pull requests:**
- Support rescaled array in sampling (#103) (@GiggleLiu)

**Closed issues:**
- Rescaled Array for Sampling (#102)

[Diff since v0.6.1](v0.6.1...v0.6.2)

**Merged pull requests:**
- Upgrade omeinsum to version 0.9.1 (#101) (@GiggleLiu)

**Closed issues:**
- Complex entries need to be conjugated after cost_and_gradient (#98)

[Diff since v0.6.0](v0.6.0...v0.6.1)

**Merged pull requests:**
- Demo: fix tests in complex BP (#100) (@GiggleLiu)

[Diff since v0.5.0](v0.5.0...v0.6.0)

- Changed the `mars` keyword of `TensorNetworkModel` argument to `unity_tensors_labels` to be more accurate.
- Changed the `vars` field of `TensorNetworkModel` to `nvars`. It seems not nessesary to name the variables.
- Removed a lot of unused interfaces for `TensorNetworkModel`.
- Move the autodiff related feastures to `OMEinsum`.

The following APIs are added:
- `BeliefPropgation`: create a belief propagation model instance, which can be derived from the UAI model.
- `belief_propagate`: perform message passing until convergence, returns a BPState object and extra info.
- `random_tensor_train_uai` and `random_matrix_product_uai`: two UAI models for testing.

**Merged pull requests:**
- Implement Belief propagation as a new inference backend (#97) (@GiggleLiu)

**Closed issues:**
- Precompilation is commented out (#64)

[Diff since v0.4.2](v0.4.2...v0.5.0)

- Add ProblemReductions as a dependency. Now we can convert a statistical model into a tensor network with a single function call.
- Add the CUDA extension, replacing the previous Requires.

**Merged pull requests:**
- Switch from Requires to extensions (#81) (@GiggleLiu)
- Fix show method for the MMAP model (#95) (@GiggleLiu)
- Fix sampling algorithm (#96) (@GiggleLiu)

**Closed issues:**
- Issue with `MMAPModel` and `most_probable_config` Function (#93)

[Diff since v0.4.1](v0.4.1...v0.4.2)

**Merged pull requests:**
- CompatHelper: add new compat entry for Artifacts at version 1, (keep existing compat) (#89) (@github-actions[bot])
- Add a function to generate tensor network model by optimized EinCode (#90) (@ArrogantGao)
- update generic tensor network to version 2.0 (#92) (@GiggleLiu)

**Closed issues:**
- different behavior of `argmax` for Vector and Matrix leads to error (#91)

[Diff since v0.4.0](v0.4.0...v0.4.1)

[Diff since v0.3.0](v0.3.0...v0.4.0)

**Merged pull requests:**
- Let marginals return dict (#61) (@GiggleLiu)
- Fix issues: 62 69 66 (#70) (@GiggleLiu)
- docs: add a section on performance evaluation (#74) (@mroavi)
- Fix the overflow issue in probability (#78) (@GiggleLiu)
- CompatHelper: bump compat for TropicalNumbers to 0.6, (keep existing compat) (#82) (@github-actions[bot])
- CompatHelper: bump compat for CUDA to 5, (keep existing compat) (#83) (@github-actions[bot])

**Closed issues:**
- Improve convering combinatorial optimization problems (#59)
- Asia network log partition is zero in the docs (#62)
- UAI reference comparison tests often fail (#65)
- Tensor networks are never defined (#66)
- State of the field is missing (#67)
- Performance evaluation is missing (#68)
- Quirks in the hard-core lattice gas example (#69)
- Code example in paper (#76)
- Arithmetic overflow or underflow when solving PR tasks (#77)

[Diff since v0.2.1](v0.2.1...v0.3.0)

**Closed issues:**
- Combinatorial optimization example (#55)
- Joint marginal probability (#56)

**Merged pull requests:**
- Example: hard core lattice gas (#58) (@GiggleLiu)
- Return `Samples` type instead of a matrix (#60) (@GiggleLiu)

[Diff since v0.2.0](v0.2.0...v0.2.1)

**Merged pull requests:**
- Port GenericTensorNetworks (#57) (@GiggleLiu)