kalmanformer 0.0.1__tar.gz

kalmanformer-0.0.1/.gitignore

@@ -0,0 +1,209 @@
+*.sh
+
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[codz]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py.cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# UV
+#   Similar to Pipfile.lock, it is generally recommended to include uv.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#uv.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+#poetry.toml
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#   pdm recommends including project-wide configuration in pdm.toml, but excluding .pdm-python.
+#   https://pdm-project.org/en/latest/usage/project/#working-with-version-control
+#pdm.lock
+#pdm.toml
+.pdm-python
+.pdm-build/
+
+# pixi
+#   Similar to Pipfile.lock, it is generally recommended to include pixi.lock in version control.
+#pixi.lock
+#   Pixi creates a virtual environment in the .pixi directory, just like venv module creates one
+#   in the .venv directory. It is recommended not to include this directory in version control.
+.pixi
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.envrc
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/
+
+# Abstra
+# Abstra is an AI-powered process automation framework.
+# Ignore directories containing user credentials, local state, and settings.
+# Learn more at https://abstra.io/docs
+.abstra/
+
+# Visual Studio Code
+#  Visual Studio Code specific template is maintained in a separate VisualStudioCode.gitignore 
+#  that can be found at https://github.com/github/gitignore/blob/main/Global/VisualStudioCode.gitignore
+#  and can be added to the global gitignore or merged into this file. However, if you prefer, 
+#  you could uncomment the following to ignore the entire vscode folder
+# .vscode/
+
+# Ruff stuff:
+.ruff_cache/
+
+# PyPI configuration file
+.pypirc
+
+# Cursor
+#  Cursor is an AI-powered code editor. `.cursorignore` specifies files/directories to
+#  exclude from AI features like autocomplete and code analysis. Recommended for sensitive data
+#  refer to https://docs.cursor.com/context/ignore-files
+.cursorignore
+.cursorindexingignore
+
+# Marimo
+marimo/_static/
+marimo/_lsp/
+__marimo__/

kalmanformer-0.0.1/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Phil Wang
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

kalmanformer-0.0.1/PKG-INFO

@@ -0,0 +1,114 @@
+Metadata-Version: 2.4
+Name: kalmanformer
+Version: 0.0.1
+Summary: KalmanFormer - using transformer to model the Kalman Gain in Kalman Filters
+Project-URL: Homepage, https://pypi.org/project/kalmanformer/
+Project-URL: Repository, https://codeberg.org/lucidrains/kalmanformer
+Author-email: Phil Wang <lucidrains@gmail.com>
+License: MIT License
+        
+        Copyright (c) 2026 Phil Wang
+        
+        Permission is hereby granted, free of charge, to any person obtaining a copy
+        of this software and associated documentation files (the "Software"), to deal
+        in the Software without restriction, including without limitation the rights
+        to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+        copies of the Software, and to permit persons to whom the Software is
+        furnished to do so, subject to the following conditions:
+        
+        The above copyright notice and this permission notice shall be included in all
+        copies or substantial portions of the Software.
+        
+        THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+        IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+        FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+        AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+        LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+        OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+        SOFTWARE.
+License-File: LICENSE
+Keywords: artificial intelligence,deep learning,kalman filter,state estimation,transformers
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Requires-Dist: einops>=0.8.0
+Requires-Dist: torch>=2.0
+Requires-Dist: x-mlps-pytorch>=0.2.8
+Requires-Dist: x-transformers>=2.19.7
+Provides-Extra: examples
+Provides-Extra: test
+Requires-Dist: pytest; extra == 'test'
+Description-Content-Type: text/markdown
+
+<img src="./kalmanformer.png" width="400px"></img>
+
+## KalmanFormer
+
+Implementation of <a href="https://www.frontiersin.org/articles/10.3389/fnbot.2024.1460255/full">KalmanFormer</a>.
+
+The paper proposes learning the Kalman Gain directly from data using Transformers, bypassing the limitations of traditional Kalman Filters on non-linear systems.
+
+## Install
+
+```bash
+$ pip install kalmanformer
+```
+
+## Usage
+
+```python
+import torch
+from kalmanformer import KalmanFormer
+
+# kalmanformer
+
+kalmanformer = KalmanFormer(
+    state_dim = 3,
+    obs_dim = 3,
+    dim = 64,
+    depth = 2,
+    heads = 2,
+    dim_head = 32,
+    mlp_dim = 64
+)
+
+# mock observations
+
+observations = torch.randn(2, 10, 3)
+
+# state transition matrix f and observation matrix h
+
+F = torch.randn(3, 3)
+H = torch.randn(3, 3)
+
+# initial state
+
+x_0 = torch.zeros(2, 3)
+
+# tracking over sequence
+
+post_states = kalmanformer(
+    observations,
+    F,
+    H,
+    x_0 = x_0
+)
+
+assert post_states.shape == (2, 10, 3)
+```
+
+## Citations
+
+```bibtex
+@article{Shen2025KalmanFormer,
+    title   = {KalmanFormer: using transformer to model the Kalman Gain in Kalman Filters},
+    author  = {Siyuan Shen and Jichen Chen and Guanfeng Yu and Zhengjun Zhai and Pujie Han},
+    journal = {Frontiers in Neurorobotics},
+    year    = {2025},
+    volume  = {18},
+    doi     = {10.3389/fnbot.2024.1460255}
+}
+```

kalmanformer-0.0.1/README.md

@@ -0,0 +1,69 @@
+<img src="./kalmanformer.png" width="400px"></img>
+
+## KalmanFormer
+
+Implementation of <a href="https://www.frontiersin.org/articles/10.3389/fnbot.2024.1460255/full">KalmanFormer</a>.
+
+The paper proposes learning the Kalman Gain directly from data using Transformers, bypassing the limitations of traditional Kalman Filters on non-linear systems.
+
+## Install
+
+```bash
+$ pip install kalmanformer
+```
+
+## Usage
+
+```python
+import torch
+from kalmanformer import KalmanFormer
+
+# kalmanformer
+
+kalmanformer = KalmanFormer(
+    state_dim = 3,
+    obs_dim = 3,
+    dim = 64,
+    depth = 2,
+    heads = 2,
+    dim_head = 32,
+    mlp_dim = 64
+)
+
+# mock observations
+
+observations = torch.randn(2, 10, 3)
+
+# state transition matrix f and observation matrix h
+
+F = torch.randn(3, 3)
+H = torch.randn(3, 3)
+
+# initial state
+
+x_0 = torch.zeros(2, 3)
+
+# tracking over sequence
+
+post_states = kalmanformer(
+    observations,
+    F,
+    H,
+    x_0 = x_0
+)
+
+assert post_states.shape == (2, 10, 3)
+```
+
+## Citations
+
+```bibtex
+@article{Shen2025KalmanFormer,
+    title   = {KalmanFormer: using transformer to model the Kalman Gain in Kalman Filters},
+    author  = {Siyuan Shen and Jichen Chen and Guanfeng Yu and Zhengjun Zhai and Pujie Han},
+    journal = {Frontiers in Neurorobotics},
+    year    = {2025},
+    volume  = {18},
+    doi     = {10.3389/fnbot.2024.1460255}
+}
+```

kalmanformer-0.0.1/kalmanformer/__init__.py

@@ -0,0 +1 @@
+from kalmanformer.kalmanformer import KalmanFormer

kalmanformer-0.0.1/kalmanformer/kalmanformer.py

@@ -0,0 +1,182 @@
+import torch
+from torch import nn, einsum, stack
+from torch.nn import Module
+
+from einops import rearrange, pack
+
+from x_transformers import Encoder, Decoder
+from x_mlps_pytorch import create_mlp
+
+# helper functions
+
+def exists(v):
+    return v is not None
+
+def default(v, d):
+    return v if exists(v) else d
+
+# main class
+
+class KalmanFormer(Module):
+    def __init__(
+        self,
+        *,
+        state_dim,
+        obs_dim,
+        dim = 64,
+        depth = 2,
+        heads = 2,
+        dim_head = 32,
+        ff_mult = 1,
+        mlp_depth = 2,
+        mlp_dim = 64,
+        enc_kwargs: dict = dict(),
+        dec_kwargs: dict = dict(),
+        learn_dynamics: bool = False
+    ):
+        super().__init__()
+        self.state_dim = state_dim
+        self.obs_dim = obs_dim
+        self.learn_dynamics = learn_dynamics
+
+        if learn_dynamics:
+            self.learned_F = create_mlp(
+                dim = dim,
+                depth = 2,
+                dim_in = state_dim,
+                dim_out = state_dim
+            )
+            self.learned_H = nn.Linear(state_dim, obs_dim, bias=False)
+
+        self.proj_obs_diff = nn.Linear(obs_dim, dim)
+        self.proj_innov_diff = nn.Linear(obs_dim, dim)
+
+        self.proj_state_evol_diff = nn.Linear(state_dim, dim)
+        self.proj_state_upd_diff = nn.Linear(state_dim, dim)
+
+        self.encoder = Encoder(
+            dim = dim,
+            depth = depth,
+            heads = heads,
+            attn_dim_head = dim_head,
+            ff_mult = ff_mult,
+            verbose = False,
+            **enc_kwargs
+        )
+
+        self.decoder = Decoder(
+            dim = dim,
+            depth = depth,
+            heads = heads,
+            attn_dim_head = dim_head,
+            cross_attend = True,
+            ff_mult = ff_mult,
+            verbose = False,
+            **dec_kwargs
+        )
+
+        self.to_kalman_gain = create_mlp(
+            dim = mlp_dim,
+            depth = mlp_depth,
+            dim_in = dim * 2,
+            dim_out = state_dim * obs_dim
+        )
+
+        nn.init.zeros_(self.to_kalman_gain.layers[-1].weight)
+        nn.init.zeros_(self.to_kalman_gain.layers[-1].bias)
+
+    def step(
+        self,
+        z_k,            # (b, obs_dim)
+        z_prev,         # (b, obs_dim)
+        x_prev_post,    # (b, state_dim)
+        x_prev_prior,   # (b, state_dim)
+        F_k,            # (b, state_dim, state_dim)
+        H_k             # (b, obs_dim, state_dim)
+    ):
+        if exists(F_k):
+            if F_k.ndim == 2:
+                F_k = rearrange(F_k, 'i j -> 1 i j')
+            x_prior = einsum('b i j, b j -> b i', F_k, x_prev_post)
+        else:
+            x_prior = self.learned_F(x_prev_post)
+
+        if exists(H_k):
+            if H_k.ndim == 2:
+                H_k = rearrange(H_k, 'i j -> 1 i j')
+            z_prior = einsum('b i j, b j -> b i', H_k, x_prior)
+        else:
+            z_prior = self.learned_H(x_prior)
+
+        obs_diff = z_k - z_prev
+        innov_diff = z_k - z_prior
+        state_evol_diff = x_prior - x_prev_post
+        state_upd_diff = x_prev_post - x_prev_prior
+
+        token_obs_diff = self.proj_obs_diff(obs_diff)
+        token_innov_diff = self.proj_innov_diff(innov_diff)
+
+        token_state_evol_diff = self.proj_state_evol_diff(state_evol_diff)
+        token_state_upd_diff = self.proj_state_upd_diff(state_upd_diff)
+
+        enc_in, _ = pack([token_obs_diff, token_innov_diff], 'b * d')
+        enc_out = self.encoder(enc_in)
+
+        dec_in, _ = pack([token_state_evol_diff, token_state_upd_diff], 'b * d')
+        dec_out = self.decoder(dec_in, context = enc_out)
+
+        dec_flat = rearrange(dec_out, 'b n d -> b (n d)')
+
+        K_k_flat = self.to_kalman_gain(dec_flat)
+        K_k = rearrange(K_k_flat, 'b (i j) -> b i j', i = self.state_dim, j = self.obs_dim)
+
+        innovation = z_k - z_prior
+        update = einsum('b i j, b j -> b i', K_k, innovation)
+        x_post = x_prior + update
+
+        return x_post, x_prior, K_k
+
+    def forward(
+        self,
+        observations,   # (b, seq, obs_dim)
+        F,              # (state_dim, state_dim) or (b, seq, state_dim, state_dim)
+        H,              # (obs_dim, state_dim) or (b, seq, obs_dim, state_dim)
+        x_0 = None      # (b, state_dim)
+    ):
+        b, seq_len, _, device = *observations.shape, observations.device
+
+        x_prev_post = default(x_0, torch.zeros(b, self.state_dim, device = device))
+        x_prev_prior = x_prev_post.clone()
+
+        if exists(H):
+            if H.ndim == 4:
+                z_prev = einsum('b i j, b j -> b i', H[:, 0], x_prev_post)
+            else:
+                z_prev = einsum('i j, b j -> b i', H, x_prev_post)
+        else:
+            z_prev = self.learned_H(x_prev_post)
+
+        post_states = []
+
+        for k in range(seq_len):
+            z_k = observations[:, k]
+
+            F_k = H_k = None
+
+            if exists(F):
+                F_k = F[:, k] if F.ndim == 4 else F
+
+            if exists(H):
+                H_k = H[:, k] if H.ndim == 4 else H
+
+            x_post, x_prior, K_k = self.step(
+                z_k, z_prev, x_prev_post, x_prev_prior, F_k, H_k
+            )
+
+            post_states.append(x_post)
+
+            z_prev = z_k
+            x_prev_post = x_post
+            x_prev_prior = x_prior
+
+        return stack(post_states, dim = 1)

kalmanformer-0.0.1/pyproject.toml

@@ -0,0 +1,69 @@
+[project]
+name = "kalmanformer"
+version = "0.0.1"
+description = "KalmanFormer - using transformer to model the Kalman Gain in Kalman Filters"
+authors = [
+    { name = "Phil Wang", email = "lucidrains@gmail.com" }
+]
+readme = "README.md"
+requires-python = ">= 3.10"
+license = { file = "LICENSE" }
+keywords = [
+    'artificial intelligence',
+    'deep learning',
+    'transformers',
+    'kalman filter',
+    'state estimation'
+]
+
+classifiers=[
+    'Development Status :: 4 - Beta',
+    'Intended Audience :: Developers',
+    'Topic :: Scientific/Engineering :: Artificial Intelligence',
+    'License :: OSI Approved :: MIT License',
+    'Programming Language :: Python :: 3.10',
+]
+
+dependencies = [
+    "einops>=0.8.0",
+    "torch>=2.0",
+    "x-mlps-pytorch>=0.2.8",
+    "x-transformers>=2.19.7"
+]
+
+[project.urls]
+Homepage = "https://pypi.org/project/kalmanformer/"
+Repository = "https://codeberg.org/lucidrains/kalmanformer"
+
+[project.optional-dependencies]
+examples = []
+test = [
+    "pytest"
+]
+
+[tool.pytest.ini_options]
+pythonpath = [
+    "."
+]
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.rye]
+managed = true
+dev-dependencies = []
+
+[tool.hatch.metadata]
+allow-direct-references = true
+
+[tool.hatch.build]
+include = [
+    "kalmanformer",
+    "pyproject.toml",
+    "README.md",
+    "LICENSE"
+]
+
+[tool.hatch.build.targets.wheel]
+packages = ["kalmanformer"]