dataset-complexity-profiler 0.1.0__tar.gz

dataset_complexity_profiler-0.1.0/LICENSE

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

dataset_complexity_profiler-0.1.0/MANIFEST.in

@@ -0,0 +1 @@
+include dataset_complexity_profiler/meta_model.pkl

dataset_complexity_profiler-0.1.0/PKG-INFO

@@ -0,0 +1,187 @@
+Metadata-Version: 2.4
+Name: dataset-complexity-profiler
+Version: 0.1.0
+Summary: Text dataset complexity profiler with a packaged meta-model for automatic embedding dimension recommendation
+License: MIT
+Keywords: nlp,meta-learning,dataset-profiling,embeddings
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: sentence-transformers>=2.2.2
+Requires-Dist: datasets>=2.14.5
+Requires-Dist: pymfe>=0.3.1
+Requires-Dist: scikit-learn>=1.2.2
+Requires-Dist: numpy>=1.24.0
+Requires-Dist: pandas>=1.5.3
+Requires-Dist: joblib>=1.2.0
+Dynamic: license-file
+
+# Dataset Complexity Profiling Tool
+
+A text dataset complexity profiler with a packaged default meta-model for automatic embedding dimension recommendations.
+
+This repository is organized as a package plus scripts, so you can use it as a library, run command-line flows with `uv`, and retrain the meta-model when needed.
+
+## Project structure
+
+- `dataset_complexity_profiler/`
+  - package code
+  - contains `DatasetProfiler` and packaged `meta_model.pkl`
+- `scripts/`
+  - helper scripts for demo, prediction, benchmark collection, training, and sanity checks
+- `tests/`
+  - automated pytest unit tests
+- `pyproject.toml`
+  - packaging and dependency settings for `uv`
+- `MANIFEST.in`
+  - includes `dataset_complexity_profiler/meta_model.pkl` in the package
+- `requirements.txt`
+  - runtime dependencies
+- `requirements-dev.txt`
+  - test/development dependencies
+- `uv.lock`
+  - `uv` dependency lock file
+
+## What each file does
+
+- `dataset_complexity_profiler/__init__.py`
+  - exposes `DatasetProfiler` for `from dataset_complexity_profiler import DatasetProfiler`
+- `dataset_complexity_profiler/dataset_adapter.py`
+  - the main implementation
+  - extracts meta-features, analyzes datasets, predicts embedding dimension, and trains custom meta-models
+- `scripts/main.py`
+  - demo runner for benchmark datasets
+  - good for a quick functional check
+- `scripts/predict.py`
+  - example script to predict optimal embedding dimension for one dataset and save JSON output
+- `scripts/collect_benchmarks.py`
+  - builds `benchmarks.csv` from a set of benchmark text datasets
+  - used when preparing training data for a new meta-model
+- `scripts/train_meta_model.py`
+  - trains the regression meta-model on `benchmarks.csv`
+  - saves a new `meta_model.pkl`
+- `scripts/test.py`
+  - manual demo script for running a quick prediction flow
+  - a small convenience script, separate from automated tests
+- `tests/`
+  - unit tests for package logic using `pytest`
+
+## What `scripts/main.py` does
+
+- Runs the package pipeline on a set of preconfigured benchmark datasets.
+- Uses the default packaged `meta_model.pkl`.
+- Downloads datasets, computes embeddings, predicts optimal embedding dimension, and prints recommendations.
+- Useful for a full end-to-end verification of the workflow.
+
+## How to demonstrate the model
+
+Use one of these scripts to show the package working:
+
+- `uv run python scripts/main.py`
+  - runs a complete benchmark demo on several datasets
+- `uv run python scripts/predict.py`
+  - predicts optimal embedding dimension for a single dataset and writes `prediction_result.json`
+- `uv run python scripts/test.py`
+  - quick manual check script for a simple prediction flow
+
+## Why `benchmarks.csv` exists
+
+- `benchmarks.csv` is a training dataset for the meta-model.
+- It is generated by `scripts/collect_benchmarks.py` from several benchmark text datasets.
+- It is not required for normal use of the package because the repository already ships with a default `meta_model.pkl`.
+
+In other words:
+
+- `benchmarks.csv` is used for meta-model training and experimentation.
+- `DatasetProfiler` can work immediately without it.
+
+## How the default model works
+
+The package includes `dataset_complexity_profiler/meta_model.pkl`.
+If you create `DatasetProfiler()` normally, it loads that model automatically.
+That means users can get recommendations without training anything.
+
+## How to train your own meta-model
+
+If you want a model tuned to your own data distribution, you do not need to pass a CSV file directly to `DatasetProfiler`.
+Instead, use Python data structures with raw text and labels.
+
+Example:
+
+```python
+from dataset_complexity_profiler import DatasetProfiler
+
+profiler = DatasetProfiler(auto_load_meta_model=False)
+
+datasets = [
+    {"texts": ["sample 1", "sample 2"], "labels": [0, 1]},
+    {"texts": ["sample 3", "sample 4"], "labels": [1, 0]},
+]
+
+profiler.train_custom_meta_model(datasets)
+profiler.save_meta_model("custom_meta_model.pkl")
+```
+
+Then later:
+
+```python
+profiler = DatasetProfiler(auto_load_meta_model=False)
+profiler.load_meta_model("custom_meta_model.pkl")
+```
+
+If you want to generate a training dataset from benchmark datasets, use:
+
+```bash
+uv run python scripts/collect_benchmarks.py
+uv run python scripts/train_meta_model.py
+```
+
+Then the resulting `meta_model.pkl` will be placed in `dataset_complexity_profiler/`.
+
+## How to use the package with `uv`
+
+Install dependencies:
+
+```bash
+uv sync
+```
+
+Run the demo script:
+
+```bash
+uv run python scripts/main.py
+```
+
+Predict on a new dataset:
+
+```bash
+uv run python scripts/predict.py
+```
+
+Collect benchmark training data:
+
+```bash
+uv run python scripts/collect_benchmarks.py
+```
+
+Train a new meta-model:
+
+```bash
+uv run python scripts/train_meta_model.py
+```
+
+## Core API
+
+- `analyze_text_dataset(texts, labels, ...)` — full dataset analysis report
+- `analyze_and_adapt(X, y, ...)` — analyze precomputed embeddings
+- `fit_transform(texts, labels, ...)` — text → embeddings → compressed vectors
+- `predict_embedding_dim(X, y, ...)` — recommended embedding dimension
+- `train_custom_meta_model(datasets, ...)` — train a custom meta-model on user datasets
+- `save_meta_model(path)` and `load_meta_model(path)`
+
+## Notes
+
+- `scripts/test.py` is optional and only for manual sanity checking.
+- `tests/` is the real automated test suite.
+- `benchmarks.csv` is only needed when you want to retrain or expand the meta-model.
+- For user-specific training, pass text/label pairs in Python, not a single CSV file.

dataset_complexity_profiler-0.1.0/README.md

@@ -0,0 +1,169 @@
+# Dataset Complexity Profiling Tool
+
+A text dataset complexity profiler with a packaged default meta-model for automatic embedding dimension recommendations.
+
+This repository is organized as a package plus scripts, so you can use it as a library, run command-line flows with `uv`, and retrain the meta-model when needed.
+
+## Project structure
+
+- `dataset_complexity_profiler/`
+  - package code
+  - contains `DatasetProfiler` and packaged `meta_model.pkl`
+- `scripts/`
+  - helper scripts for demo, prediction, benchmark collection, training, and sanity checks
+- `tests/`
+  - automated pytest unit tests
+- `pyproject.toml`
+  - packaging and dependency settings for `uv`
+- `MANIFEST.in`
+  - includes `dataset_complexity_profiler/meta_model.pkl` in the package
+- `requirements.txt`
+  - runtime dependencies
+- `requirements-dev.txt`
+  - test/development dependencies
+- `uv.lock`
+  - `uv` dependency lock file
+
+## What each file does
+
+- `dataset_complexity_profiler/__init__.py`
+  - exposes `DatasetProfiler` for `from dataset_complexity_profiler import DatasetProfiler`
+- `dataset_complexity_profiler/dataset_adapter.py`
+  - the main implementation
+  - extracts meta-features, analyzes datasets, predicts embedding dimension, and trains custom meta-models
+- `scripts/main.py`
+  - demo runner for benchmark datasets
+  - good for a quick functional check
+- `scripts/predict.py`
+  - example script to predict optimal embedding dimension for one dataset and save JSON output
+- `scripts/collect_benchmarks.py`
+  - builds `benchmarks.csv` from a set of benchmark text datasets
+  - used when preparing training data for a new meta-model
+- `scripts/train_meta_model.py`
+  - trains the regression meta-model on `benchmarks.csv`
+  - saves a new `meta_model.pkl`
+- `scripts/test.py`
+  - manual demo script for running a quick prediction flow
+  - a small convenience script, separate from automated tests
+- `tests/`
+  - unit tests for package logic using `pytest`
+
+## What `scripts/main.py` does
+
+- Runs the package pipeline on a set of preconfigured benchmark datasets.
+- Uses the default packaged `meta_model.pkl`.
+- Downloads datasets, computes embeddings, predicts optimal embedding dimension, and prints recommendations.
+- Useful for a full end-to-end verification of the workflow.
+
+## How to demonstrate the model
+
+Use one of these scripts to show the package working:
+
+- `uv run python scripts/main.py`
+  - runs a complete benchmark demo on several datasets
+- `uv run python scripts/predict.py`
+  - predicts optimal embedding dimension for a single dataset and writes `prediction_result.json`
+- `uv run python scripts/test.py`
+  - quick manual check script for a simple prediction flow
+
+## Why `benchmarks.csv` exists
+
+- `benchmarks.csv` is a training dataset for the meta-model.
+- It is generated by `scripts/collect_benchmarks.py` from several benchmark text datasets.
+- It is not required for normal use of the package because the repository already ships with a default `meta_model.pkl`.
+
+In other words:
+
+- `benchmarks.csv` is used for meta-model training and experimentation.
+- `DatasetProfiler` can work immediately without it.
+
+## How the default model works
+
+The package includes `dataset_complexity_profiler/meta_model.pkl`.
+If you create `DatasetProfiler()` normally, it loads that model automatically.
+That means users can get recommendations without training anything.
+
+## How to train your own meta-model
+
+If you want a model tuned to your own data distribution, you do not need to pass a CSV file directly to `DatasetProfiler`.
+Instead, use Python data structures with raw text and labels.
+
+Example:
+
+```python
+from dataset_complexity_profiler import DatasetProfiler
+
+profiler = DatasetProfiler(auto_load_meta_model=False)
+
+datasets = [
+    {"texts": ["sample 1", "sample 2"], "labels": [0, 1]},
+    {"texts": ["sample 3", "sample 4"], "labels": [1, 0]},
+]
+
+profiler.train_custom_meta_model(datasets)
+profiler.save_meta_model("custom_meta_model.pkl")
+```
+
+Then later:
+
+```python
+profiler = DatasetProfiler(auto_load_meta_model=False)
+profiler.load_meta_model("custom_meta_model.pkl")
+```
+
+If you want to generate a training dataset from benchmark datasets, use:
+
+```bash
+uv run python scripts/collect_benchmarks.py
+uv run python scripts/train_meta_model.py
+```
+
+Then the resulting `meta_model.pkl` will be placed in `dataset_complexity_profiler/`.
+
+## How to use the package with `uv`
+
+Install dependencies:
+
+```bash
+uv sync
+```
+
+Run the demo script:
+
+```bash
+uv run python scripts/main.py
+```
+
+Predict on a new dataset:
+
+```bash
+uv run python scripts/predict.py
+```
+
+Collect benchmark training data:
+
+```bash
+uv run python scripts/collect_benchmarks.py
+```
+
+Train a new meta-model:
+
+```bash
+uv run python scripts/train_meta_model.py
+```
+
+## Core API
+
+- `analyze_text_dataset(texts, labels, ...)` — full dataset analysis report
+- `analyze_and_adapt(X, y, ...)` — analyze precomputed embeddings
+- `fit_transform(texts, labels, ...)` — text → embeddings → compressed vectors
+- `predict_embedding_dim(X, y, ...)` — recommended embedding dimension
+- `train_custom_meta_model(datasets, ...)` — train a custom meta-model on user datasets
+- `save_meta_model(path)` and `load_meta_model(path)`
+
+## Notes
+
+- `scripts/test.py` is optional and only for manual sanity checking.
+- `tests/` is the real automated test suite.
+- `benchmarks.csv` is only needed when you want to retrain or expand the meta-model.
+- For user-specific training, pass text/label pairs in Python, not a single CSV file.

dataset_complexity_profiler-0.1.0/dataset_complexity_profiler/__init__.py

@@ -0,0 +1,9 @@
+"""Пакет dataset_complexity_profiler.
+
+Предоставляет основной класс `DatasetProfiler` для анализа текстовых датасетов
+и рекомендации оптимальной размерности эмбеддингов.
+"""
+
+from .dataset_adapter import DatasetProfiler
+
+__all__ = ["DatasetProfiler"]