saturncube 0.7.0__tar.gz

saturncube-0.7.0/CHANGELOG.md

@@ -0,0 +1,46 @@
+# Changelog
+
+## Unreleased
+- **Behaviour change:** `patience` now defaults to `"auto"`: 10 for regression, off for classification. Regression fits that used to
+  run all `epochs` now hold out `val_split` (10%) of the training rows and may stop early with the best weights restored. The
+  automatic default needs at least 20 held-out rows (`AUTO_MIN_VAL_ROWS`) and never warns; `patience=None` restores the old behaviour.
+- `patience` now always has data to monitor: without `val_data=`, models hold out `val_split` (default 0.1, stratified for
+  classification) of the training rows; `val_split=0` disables it and `patience` warns. New attributes `n_val_`, `best_epoch_`,
+  `stopped_epoch_`. `partial_fit` never holds rows out. Early stopping is still off by default.
+- `sc.warmup()` pays PyTorch's one-time optimizer start-up cost; `sc.benchmark` calls it so the first model's timing is fair.
+- docs/TUNING.md and scripts/tune_defaults.py, scripts/tune_followup.py: measured evidence for the training defaults. No default
+  learning rate or epoch count was changed.
+- `sc.benchmark(models, X, y, groups=None, seeds=5, baselines=True)`: fair model comparison with group-disjoint splits, mean ± std
+  over seeds, automatic majority and scaled scikit-learn baselines, and a printed table. Returns a `BenchmarkResult`.
+- New optional extra `bench` (scikit-learn); `dev` now includes it.
+- Text input: `TextVectorizer` (TF-IDF, optional SVD), `TokenEncoder` (padded token ids) and `TextModel`, which puts any
+  registry model behind raw-text `fit` / `predict` / `evaluate` / `save` / `load`. Encoders are fitted on training text only
+  and saved as JSON.
+- The README model table is generated from the registry (`python scripts/update_readme.py`); a test fails if it drifts.
+- Every registered class now has its own docstring (38 were missing); a test enforces it.
+- `sc.available()` no longer lists `unit_tests` (not a model). `register(..., hidden=True)` and
+  `available(include_hidden=True)` keep such helpers creatable by name.
+
+## 0.7.0
+- Automatic data preparation: `load_dataset`, `Dataset.report/prepare/prepare_sequences/prepare_images`, a saved-and-reloadable
+  `Pipeline`, `auto_model`, `check_arrays`. Handles messy CSVs (formats, missing values, categories, dates), splits without leakage.
+- Models now raise a clear error for text or NaN input instead of failing obscurely.
+
+## 0.6.0
+- Added `CNN`, a compact ResNet-style convolutional network for images (classification and regression).
+- Added `SpikingNetwork`, a leaky integrate-and-fire network trained with surrogate gradients (direct or rate encoding).
+- Added `SelfOrganizingMap`, with quantization / topographic error, U-matrix, hit map and unit labelling.
+- Removed the experimental C-language helpers; program tools are Python-only.
+- Packaging: full project metadata, `slow` test marker, CI and publish workflows.
+
+## 0.5.0
+- Search (genetic algorithm, successive halving, simulated annealing), program synthesis and Datalog,
+  retrieval and case memory, learning strategies, verification and property testing, code generation,
+  experiment design and statistics, and classical ML (random forest, k-means, naive Bayes).
+
+## 0.4.0
+- Sequence, attention, memory, metric / few-shot, graph, tree and pointer models; deep RL; meta-learning;
+  more learning rules and optimizers.
+
+## 0.1.0 - 0.3.0
+- Vector/tabular networks, optimizers, unsupervised models, Hebbian rules, continual learning, tabular RL.

saturncube-0.7.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 ChristopherDarien-3
+
+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.

saturncube-0.7.0/MANIFEST.in

@@ -0,0 +1,7 @@
+include README.md CHANGELOG.md RELEASING.md
+include LICENSE*
+recursive-include tests *.py
+recursive-include examples *.py
+recursive-include docs *.md
+recursive-include scripts *.py
+global-exclude __pycache__ *.py[cod]