turingbot 3.3.0__tar.gz → 4.0__tar.gz

{turingbot-3.3.0 → turingbot-4.0}/LICENSE

@@ -1,4 +1,4 @@
-Copyright (c) 2025 TuringBot Software. All rights reserved.
+Copyright (c) 2020-2026 TuringBot Software. All rights reserved.
 
 Permission is granted to use this software solely in conjunction with a
 licensed installation of TuringBot. Redistribution, modification, or use

turingbot-4.0/PKG-INFO

@@ -0,0 +1,119 @@
+Metadata-Version: 2.4
+Name: turingbot
+Version: 4.0
+Summary: Symbolic regression in Python: find the mathematical formula behind your data with TuringBot
+Author: TuringBot Software
+License-Expression: LicenseRef-Proprietary
+Project-URL: Homepage, https://turingbotsoftware.com
+Project-URL: Documentation, https://turingbotsoftware.com/documentation.html
+Project-URL: Repository, https://github.com/turingbotsoftware/turingbot-python
+Project-URL: Download, https://turingbotsoftware.com/download.html
+Keywords: symbolic regression,equation discovery,machine learning,interpretable machine learning,curve fitting,scikit-learn,scientific computing,simulated annealing
+Classifier: Development Status :: 5 - Production/Stable
+Classifier: Intended Audience :: Science/Research
+Classifier: Intended Audience :: Education
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Scientific/Engineering
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Scientific/Engineering :: Mathematics
+Requires-Python: >=3.7
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Dynamic: license-file
+
+# TuringBot Python Library
+
+Python interface to [TuringBot](https://turingbotsoftware.com), a symbolic regression engine: it searches for explicit formulas that fit your data, and returns the most accurate formula found for each size.
+
+The engine is compiled C++ and uses simulated annealing. There is no language runtime to set up and no GUI involved, so searches run the same in a notebook, over SSH or on a headless server. Input can be a CSV file, a numpy array or a pandas DataFrame.
+
+Coming from PySR? [`TuringBotRegressor`](#coming-from-pysr) uses the same API as `PySRRegressor`.
+
+## Installation
+
+```bash
+pip install turingbot
+```
+
+The engine comes with the TuringBot application: [download it](https://turingbotsoftware.com/download.html) and install it. The library needs Python 3.7 or newer and has no dependencies.
+
+## Quickstart
+
+```python
+import time
+import numpy as np
+from turingbot import simulation
+
+x = np.linspace(-3, 3, 100)
+data = np.column_stack([x, 2.5 * np.sin(x) + x ** 2])  # the last column is the target
+
+with simulation() as sim:
+    sim.start_process(input_file=data, column_names=["x", "y"])
+    time.sleep(10)
+    sim.refresh_functions()
+    for f in sim.functions:
+        print(f)
+```
+
+```
+[1, 3.27574, '3.0606060606060477']
+[3, 1.79969, 'x*x']
+[5, 1.03435, 'x*(0.8418349138360374+x)']
+[8, 0.93293, 'asinh(x)+x*x']
+[10, 0.0, '2.5*sin(x)+x*x']
+```
+
+Each line is `[size, error, formula]`, the Pareto front of size versus error, from the simplest formula to the most accurate. Here, the last one is the formula that generated the data.
+
+The search runs in a background process. `refresh_functions()` reads the current front at any time, and the search stops at `terminate_process()` or when the `with` block exits.
+
+`input_file` also takes the path of a CSV or TXT file, or a pandas DataFrame; the last column is the target. TuringBot is found in its default install location; otherwise, pass its path with `path=`. On a Linux server, that can be the AppImage, which needs no installation.
+
+## Features
+
+- **Custom search metrics**, which can reference dataset columns, for example a weight column `w`: `custom_metric="sum(w*pow(actual-predicted,2))/sum(w)"`.
+- **Custom functions and constants**, used as building blocks by the search: `custom_functions="bump(x)=exp(-x*x)"`, `custom_constants="g=9.81,c=299792458"`.
+- **Time series**: `delay`, `moving_average` and `decay` give formulas access to past values of the inputs, and with `allow_target_delay=True`, of the target itself.
+- **GUI and command line**: the same engine runs in the TuringBot desktop app and from the command line, with the same options.
+
+## Search options
+
+```python
+sim.start_process(
+    input_file="data.csv",
+    search_metric=4,                       # RMS error (the default)
+    allowed_functions="+ * / sin cos exp",
+    train_test_split=80,                   # 80% of the rows for training, 20% for testing
+    threads=4,
+)
+```
+
+With `train_test_split`, each formula also has its error on the test rows: `[size, error, test_error, formula]`.
+
+Every option, with the IDs of the 16 search metrics: `help(simulation.start_process)`, or the [documentation](https://turingbotsoftware.com/documentation.html#python).
+
+## Coming from PySR
+
+`TuringBotRegressor` follows the scikit-learn API and uses `PySRRegressor`'s names where the two match: `model_selection` (`"best"`, `"accuracy"` or `"score"`, with the same definitions), `equations_`, `sympy()`, `max_evals` and `early_stop_condition`. It works in pipelines, cross-validation and grid searches.
+
+```python
+from turingbot import TuringBotRegressor
+
+model = TuringBotRegressor(max_time=30).fit(X, y)
+print(model.equations_)   # the formulas found, from the simplest to the most accurate
+print(model.sympy())      # the chosen formula as a sympy expression
+y_pred = model.predict(X_new)
+```
+
+| PySRRegressor | TuringBotRegressor |
+|---|---|
+| `niterations`, `timeout_in_seconds` | `max_time` (seconds) |
+| `binary_operators`, `unary_operators` | `allowed_functions` |
+| `maxsize` | `maximum_formula_complexity` |
+
+It needs numpy. Its options are in `help(TuringBotRegressor)`.
+
+## Citing
+
+If you use TuringBot in your research, please cite it: [CITATION.cff](CITATION.cff) has the details, which GitHub also shows under "Cite this repository".

turingbot-4.0/README.md

@@ -0,0 +1,95 @@
+# TuringBot Python Library
+
+Python interface to [TuringBot](https://turingbotsoftware.com), a symbolic regression engine: it searches for explicit formulas that fit your data, and returns the most accurate formula found for each size.
+
+The engine is compiled C++ and uses simulated annealing. There is no language runtime to set up and no GUI involved, so searches run the same in a notebook, over SSH or on a headless server. Input can be a CSV file, a numpy array or a pandas DataFrame.
+
+Coming from PySR? [`TuringBotRegressor`](#coming-from-pysr) uses the same API as `PySRRegressor`.
+
+## Installation
+
+```bash
+pip install turingbot
+```
+
+The engine comes with the TuringBot application: [download it](https://turingbotsoftware.com/download.html) and install it. The library needs Python 3.7 or newer and has no dependencies.
+
+## Quickstart
+
+```python
+import time
+import numpy as np
+from turingbot import simulation
+
+x = np.linspace(-3, 3, 100)
+data = np.column_stack([x, 2.5 * np.sin(x) + x ** 2])  # the last column is the target
+
+with simulation() as sim:
+    sim.start_process(input_file=data, column_names=["x", "y"])
+    time.sleep(10)
+    sim.refresh_functions()
+    for f in sim.functions:
+        print(f)
+```
+
+```
+[1, 3.27574, '3.0606060606060477']
+[3, 1.79969, 'x*x']
+[5, 1.03435, 'x*(0.8418349138360374+x)']
+[8, 0.93293, 'asinh(x)+x*x']
+[10, 0.0, '2.5*sin(x)+x*x']
+```
+
+Each line is `[size, error, formula]`, the Pareto front of size versus error, from the simplest formula to the most accurate. Here, the last one is the formula that generated the data.
+
+The search runs in a background process. `refresh_functions()` reads the current front at any time, and the search stops at `terminate_process()` or when the `with` block exits.
+
+`input_file` also takes the path of a CSV or TXT file, or a pandas DataFrame; the last column is the target. TuringBot is found in its default install location; otherwise, pass its path with `path=`. On a Linux server, that can be the AppImage, which needs no installation.
+
+## Features
+
+- **Custom search metrics**, which can reference dataset columns, for example a weight column `w`: `custom_metric="sum(w*pow(actual-predicted,2))/sum(w)"`.
+- **Custom functions and constants**, used as building blocks by the search: `custom_functions="bump(x)=exp(-x*x)"`, `custom_constants="g=9.81,c=299792458"`.
+- **Time series**: `delay`, `moving_average` and `decay` give formulas access to past values of the inputs, and with `allow_target_delay=True`, of the target itself.
+- **GUI and command line**: the same engine runs in the TuringBot desktop app and from the command line, with the same options.
+
+## Search options
+
+```python
+sim.start_process(
+    input_file="data.csv",
+    search_metric=4,                       # RMS error (the default)
+    allowed_functions="+ * / sin cos exp",
+    train_test_split=80,                   # 80% of the rows for training, 20% for testing
+    threads=4,
+)
+```
+
+With `train_test_split`, each formula also has its error on the test rows: `[size, error, test_error, formula]`.
+
+Every option, with the IDs of the 16 search metrics: `help(simulation.start_process)`, or the [documentation](https://turingbotsoftware.com/documentation.html#python).
+
+## Coming from PySR
+
+`TuringBotRegressor` follows the scikit-learn API and uses `PySRRegressor`'s names where the two match: `model_selection` (`"best"`, `"accuracy"` or `"score"`, with the same definitions), `equations_`, `sympy()`, `max_evals` and `early_stop_condition`. It works in pipelines, cross-validation and grid searches.
+
+```python
+from turingbot import TuringBotRegressor
+
+model = TuringBotRegressor(max_time=30).fit(X, y)
+print(model.equations_)   # the formulas found, from the simplest to the most accurate
+print(model.sympy())      # the chosen formula as a sympy expression
+y_pred = model.predict(X_new)
+```
+
+| PySRRegressor | TuringBotRegressor |
+|---|---|
+| `niterations`, `timeout_in_seconds` | `max_time` (seconds) |
+| `binary_operators`, `unary_operators` | `allowed_functions` |
+| `maxsize` | `maximum_formula_complexity` |
+
+It needs numpy. Its options are in `help(TuringBotRegressor)`.
+
+## Citing
+
+If you use TuringBot in your research, please cite it: [CITATION.cff](CITATION.cff) has the details, which GitHub also shows under "Cite this repository".

{turingbot-3.3.0 → turingbot-4.0}/pyproject.toml

@@ -1,11 +1,11 @@
 [build-system]
-requires = ["setuptools>=61.0"]
+requires = ["setuptools>=77.0"]
 build-backend = "setuptools.build_meta"
 
 [project]
 name = "turingbot"
-version = "3.3.0"
-description = "Python interface for TuringBot, a symbolic regression software that discovers mathematical formulas from data"
+version = "4.0"
+description = "Symbolic regression in Python: find the mathematical formula behind your data with TuringBot"
 readme = "README.md"
 license = "LicenseRef-Proprietary"
 requires-python = ">=3.7"
@@ -14,7 +14,11 @@ authors = [
 ]
 keywords = [
     "symbolic regression",
+    "equation discovery",
     "machine learning",
+    "interpretable machine learning",
+    "curve fitting",
+    "scikit-learn",
     "scientific computing",
     "simulated annealing",
 ]

turingbot-4.0/turingbot/__init__.py

@@ -0,0 +1,4 @@
+from .turingbot import simulation, find_executable, TuringBotRegressor
+
+__version__ = "4.0"
+__all__ = ["simulation", "find_executable", "TuringBotRegressor"]