# Torchwnn: *Weightless Neural Network*
Torchwnn is a Python library for *Weightless Neural Network* (also known as *RAM-based* and *N-tuple based Neural Network* ).
# Usage
## Installation
First, install PyTorch using their [installation instructions](https://pytorch.org/get-started/locally/). Then, use the following command to install Torchwnn:
```bash
pip install torchwnn
```
Requirements: PyTorch and ucimlrepo to load datasets from UCI repository.
## Quick Start
### Iris Example
To quickly get started with Torchwnn, here's an example using the Iris dataset. Full training code is available in the [examples/iris.py](examples/iris.py) file.
```python
import torch
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
from torchwnn.datasets.iris import Iris
from torchwnn.classifiers import Wisard
from torchwnn.encoding import Thermometer
# Use the GPU if available
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print("Using {} device".format(device))
iris = Iris()
X = iris.features
X = torch.tensor(X.values).to(device)
y = list(iris.labels)
y = torch.tensor(y).squeeze().to(device)
bits_encoding = 20
encoding = Thermometer(bits_encoding).fit(X)
X_bin = encoding.binarize(X).flatten(start_dim=1)
X_train, X_test, y_train, y_test = train_test_split(X_bin, y, test_size=0.3, random_state = 0)
entry_size = X_train.shape[1]
tuple_size = 8
model = Wisard(entry_size, iris.num_classes, tuple_size)
with torch.no_grad():
model.fit(X_train,y_train)
predictions = model.predict(X_test)
acc = accuracy_score(predictions, y_test)
print("Wisard: Accuracy = ", acc)
```
## Examples
There are several examples in the repository.
### Bleaching
```python
import torch
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
from torchwnn.datasets.iris import Iris
from torchwnn.classifiers import Wisard
from torchwnn.encoding import Thermometer
# Use the GPU if available
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print("Using {} device".format(device))
iris = Iris()
X = iris.features
X = torch.tensor(X.values).to(device)
y = list(iris.labels)
y = torch.tensor(y).squeeze().to(device)
bits_encoding = 20
encoding = Thermometer(bits_encoding).fit(X)
X_bin = encoding.binarize(X).flatten(start_dim=1)
X_train, X_test, y_train, y_test = train_test_split(X_bin, y, test_size=0.3, random_state = 0)
entry_size = X_train.shape[1]
tuple_size = 8
model = Wisard(entry_size, iris.num_classes, tuple_size, bleaching=True)
with torch.no_grad():
model.fit(X_train,y_train)
predictions = model.predict(X_test)
acc = accuracy_score(predictions, y_test)
print("Wisard: Accuracy = ", acc)
# Applying bleaching
model.fit_bleach(X_train,y_train)
print("Selected bleach: ", model.bleach)
predictions = model.predict(X_test)
acc = accuracy_score(predictions, y_test)
print("Wisard with bleaching = ", model.bleach,": Accuracy = ", acc)
```
### BloomWisard
Example using BloomWisard is available in the [examples/iris_filters.py](examples/iris_filters.py) file.
## Supported WNN models
Currently, the library supports the following WNN models:
- WiSARD - Neurons based on dictionary.
- BloomWiSARD - Neurons based on Bloom Filters. Reference: [Weightless Neural Networks as Memory Segmented Bloom Filters](https://www.sciencedirect.com/science/article/abs/pii/S0925231220305105?via%3Dihub)
Supported techniques:
- B-bleaching - Bleaching based on binary search. Reference: *B-bleaching : Agile Overtraining Avoidance in the WiSARD Weightless Neural Classifier*.
- WiSARD
- BloomWiSARD
Raw data
{
"_id": null,
"home_page": "https://github.com/leandro-santiago/torchwnn",
"name": "torchwnn",
"maintainer": null,
"docs_url": null,
"requires_python": null,
"maintainer_email": null,
"keywords": "wisard, weithgless, neural, network",
"author": "Leandro Santiago de Ara\u00fajo",
"author_email": null,
"download_url": "https://files.pythonhosted.org/packages/36/06/6f7f7ed2338f7a8baf692ff5408e7fe1270a5a35d26383ff71ad734a45c7/torchwnn-0.0.0.tar.gz",
"platform": null,
"description": "# Torchwnn: *Weightless Neural Network*\n\nTorchwnn is a Python library for *Weightless Neural Network* (also known as *RAM-based* and *N-tuple based Neural Network* ).\n\n# Usage\n## Installation\n\nFirst, install PyTorch using their [installation instructions](https://pytorch.org/get-started/locally/). Then, use the following command to install Torchwnn:\n\n\n```bash\npip install torchwnn\n```\n\nRequirements: PyTorch and ucimlrepo to load datasets from UCI repository.\n\n\n## Quick Start\n\n### Iris Example\n\nTo quickly get started with Torchwnn, here's an example using the Iris dataset. Full training code is available in the [examples/iris.py](examples/iris.py) file.\n\n```python\nimport torch\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.metrics import accuracy_score\n\nfrom torchwnn.datasets.iris import Iris\nfrom torchwnn.classifiers import Wisard\nfrom torchwnn.encoding import Thermometer\n\n# Use the GPU if available\ndevice = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\nprint(\"Using {} device\".format(device))\n\niris = Iris()\nX = iris.features\nX = torch.tensor(X.values).to(device)\ny = list(iris.labels)\ny = torch.tensor(y).squeeze().to(device)\n\nbits_encoding = 20\nencoding = Thermometer(bits_encoding).fit(X) \nX_bin = encoding.binarize(X).flatten(start_dim=1)\n\nX_train, X_test, y_train, y_test = train_test_split(X_bin, y, test_size=0.3, random_state = 0) \n\nentry_size = X_train.shape[1]\ntuple_size = 8\nmodel = Wisard(entry_size, iris.num_classes, tuple_size)\n\nwith torch.no_grad():\n model.fit(X_train,y_train)\n predictions = model.predict(X_test) \n acc = accuracy_score(predictions, y_test)\n print(\"Wisard: Accuracy = \", acc)\n\n```\n## Examples\n\nThere are several examples in the repository. \n\n### Bleaching\n\n```python\nimport torch\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.metrics import accuracy_score\n\nfrom torchwnn.datasets.iris import Iris\nfrom torchwnn.classifiers import Wisard\nfrom torchwnn.encoding import Thermometer\n\n# Use the GPU if available\ndevice = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\nprint(\"Using {} device\".format(device))\n\niris = Iris()\nX = iris.features\nX = torch.tensor(X.values).to(device)\ny = list(iris.labels)\ny = torch.tensor(y).squeeze().to(device)\n\nbits_encoding = 20\nencoding = Thermometer(bits_encoding).fit(X) \nX_bin = encoding.binarize(X).flatten(start_dim=1)\n\nX_train, X_test, y_train, y_test = train_test_split(X_bin, y, test_size=0.3, random_state = 0) \n\nentry_size = X_train.shape[1]\ntuple_size = 8\nmodel = Wisard(entry_size, iris.num_classes, tuple_size, bleaching=True)\n\nwith torch.no_grad():\n model.fit(X_train,y_train)\n predictions = model.predict(X_test) \n acc = accuracy_score(predictions, y_test)\n print(\"Wisard: Accuracy = \", acc)\n \n # Applying bleaching\n model.fit_bleach(X_train,y_train)\n print(\"Selected bleach: \", model.bleach)\n predictions = model.predict(X_test) \n acc = accuracy_score(predictions, y_test)\n print(\"Wisard with bleaching = \", model.bleach,\": Accuracy = \", acc)\n\n```\n### BloomWisard\n\nExample using BloomWisard is available in the [examples/iris_filters.py](examples/iris_filters.py) file.\n\n## Supported WNN models\nCurrently, the library supports the following WNN models:\n\n- WiSARD - Neurons based on dictionary.\n- BloomWiSARD - Neurons based on Bloom Filters. Reference: [Weightless Neural Networks as Memory Segmented Bloom Filters](https://www.sciencedirect.com/science/article/abs/pii/S0925231220305105?via%3Dihub)\n\nSupported techniques:\n- B-bleaching - Bleaching based on binary search. Reference: *B-bleaching : Agile Overtraining Avoidance in the WiSARD Weightless Neural Classifier*.\n - WiSARD\n - BloomWiSARD\n",
"bugtrack_url": null,
"license": "MIT",
"summary": "Torcwnn is a Python library for Weightless Neural Network",
"version": "0.0.0",
"project_urls": {
"Homepage": "https://github.com/leandro-santiago/torchwnn"
},
"split_keywords": [
"wisard",
" weithgless",
" neural",
" network"
],
"urls": [
{
"comment_text": null,
"digests": {
"blake2b_256": "36066f7f7ed2338f7a8baf692ff5408e7fe1270a5a35d26383ff71ad734a45c7",
"md5": "beded90c1b5bc6a4b43cafcf8b2c014a",
"sha256": "4717291ce7f15d1278b1bfb6366e5dd5f1086ccae0a27b2095e2e4557f9eaf33"
},
"downloads": -1,
"filename": "torchwnn-0.0.0.tar.gz",
"has_sig": false,
"md5_digest": "beded90c1b5bc6a4b43cafcf8b2c014a",
"packagetype": "sdist",
"python_version": "source",
"requires_python": null,
"size": 12336,
"upload_time": "2025-02-13T05:35:16",
"upload_time_iso_8601": "2025-02-13T05:35:16.926547Z",
"url": "https://files.pythonhosted.org/packages/36/06/6f7f7ed2338f7a8baf692ff5408e7fe1270a5a35d26383ff71ad734a45c7/torchwnn-0.0.0.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2025-02-13 05:35:16",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "leandro-santiago",
"github_project": "torchwnn",
"travis_ci": false,
"coveralls": false,
"github_actions": true,
"requirements": [
{
"name": "ucimlrepo",
"specs": []
},
{
"name": "torch",
"specs": []
},
{
"name": "scikit-learn",
"specs": []
}
],
"lcname": "torchwnn"
}
JSON
