diff --git a/docs/mkdocs.yml b/docs/mkdocs.yml index f4aa77710..330fcf4c4 100755 --- a/docs/mkdocs.yml +++ b/docs/mkdocs.yml @@ -47,6 +47,7 @@ pages: - user_guide/data/boston_housing_data.md - user_guide/data/iris_data.md - user_guide/data/loadlocal_mnist.md + - user_guide/data/make_multiplexer_dataset.md - user_guide/data/mnist_data.md - user_guide/data/three_blobs_data.md - user_guide/data/wine_data.md diff --git a/docs/sources/CHANGELOG.md b/docs/sources/CHANGELOG.md index cb1170e95..b03c1f453 100755 --- a/docs/sources/CHANGELOG.md +++ b/docs/sources/CHANGELOG.md @@ -19,6 +19,7 @@ The CHANGELOG for the current development version is available at - Added `'leverage'` and `'conviction` as evaluation metrics to the `frequent_patterns.association_rules` function. [#246](https://github.com/rasbt/mlxtend/pull/246) & [#247](https://github.com/rasbt/mlxtend/pull/247) - Added a `loadings_` attribute to `PrincipalComponentAnalysis` to compute the factor loadings of the features on the principal components. [#251](https://github.com/rasbt/mlxtend/pull/251) - Allow grid search over classifiers/regressors in ensemble and stacking estimators [#259](https://github.com/rasbt/mlxtend/pull/259) +- New `make_multiplexer_dataset` function that creates a dataset generated by a n-bit Boolean multiplexer for evaluating supervised learning algorithms [#263](https://github.com/rasbt/mlxtend/pull/263) ##### Changes diff --git a/docs/sources/USER_GUIDE_INDEX.md b/docs/sources/USER_GUIDE_INDEX.md index bfc0fec50..8d856da67 100755 --- a/docs/sources/USER_GUIDE_INDEX.md +++ b/docs/sources/USER_GUIDE_INDEX.md @@ -18,6 +18,7 @@ - [boston_housing_data](user_guide/data/boston_housing_data.md) - [iris_data](user_guide/data/iris_data.md) - [loadlocal_mnist](user_guide/data/loadlocal_mnist.md) +- [make_multiplexer_dataset](user_guide/data/make_multiplexer_dataset.md) - [mnist_data](user_guide/data/mnist_data.md) - [three_blobs_data](user_guide/data/three_blobs_data.md) - [wine_data](user_guide/data/wine_data.md) diff --git a/docs/sources/user_guide/data/make_multiplexer_dataset.ipynb b/docs/sources/user_guide/data/make_multiplexer_dataset.ipynb new file mode 100644 index 000000000..929c10595 --- /dev/null +++ b/docs/sources/user_guide/data/make_multiplexer_dataset.ipynb @@ -0,0 +1,235 @@ +{ + "cells": [ + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "# Make Multiplexer Dataset" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "Function that creates a dataset generated by a n-bit Boolean multiplexer for evaluating supervised learning algorithms." + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "> `from mlxtend.data import make_multiplexer_dataset` " + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## Overview" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "The `make_multiplexer_dataset` function creates a dataset generated by an n-bit Boolean multiplexer. Such dataset represents a dataset generated by a simple rule, based on the behavior of a electric multiplexer, yet presents a relatively challenging classification problem for supervised learning algorithm with interactions between features (epistasis) as it may be encountered in many real-world scenarios [1].\n", + "\n", + "The following illustration depicts a 6-bit multiplexer that consists of 2 address bits and 4 register bits. The address bits converted to decimal representation point to a position in the register bit. For example, if the address bits are \"00\" (0 in decimal), the address bits point to the register bit at position 0. The value of the register position pointed to determines the class label. For example, if the register bit at position is 0, the class label is 0. Vice versa, if the register bit at position 0 is 1, the class label is 1. \n", + "\n", + "![](make_multiplexer_dataset_data_files/6bit_multiplexer.png)\n", + "\n", + "\n", + "In the example above, the address bits \"10\" (2 in decimal) point to the 3rd register position (as we start counting from index 0), which has a bit value of 1. Hence, the class label is 1.\n", + "\n", + "Below are a few more examples:\n", + "\n", + "1. Address bits: [0, 1], register bits: [1, 0, 1, 1], class label: 0\n", + "2. Address bits: [0, 1], register bits: [1, 1, 1, 0], class label: 1\n", + "3. Address bits: [1, 0], register bits: [1, 0, 0, 1], class label: 0\n", + "4. Address bits: [1, 1], register bits: [1, 1, 1, 0], class label: 0\n", + "5. Address bits: [0, 1], register bits: [0, 1, 1, 0], class label: 1\n", + "6. Address bits: [0, 1], register bits: [1, 0, 0, 1], class label: 0\n", + "7. Address bits: [0, 1], register bits: [0, 1, 1, 1], class label: 1\n", + "8. Address bits: [0, 1], register bits: [0, 0, 0, 0], class label: 0\n", + "9. Address bits: [1, 0], register bits: [1, 0, 1, 1], class label: 1\n", + "10. Address bits: [0, 1], register bits: [1, 1, 1, 1], class label: 1\n", + "\n", + "Note that in the implementation of the multiplexer function, if the number of address bits is set to 2, this results in a 6 bit multiplexer as two bit can have 2^2=4 different register positions (2 bit + 4 bit = 6 bit). However, if we choose 3 address bits instead, 2^3=8 positions would be covered, resulting in a 11 bit (3 bit + 8 bit = 11 bit) multiplexer, and so forth.\n", + "\n" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "### References\n", + "\n", + "- [1] Urbanowicz, R. J., & Browne, W. N. (2017). *Introduction to Learning Classifier Systems*. Springer." + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## Example 1 -- 6-bit multiplexer" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "This simple example illustrates how to create dataset from a 6-bit multiplexer" + ] + }, + { + "cell_type": "code", + "execution_count": 1, + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Features:\n", + " [[0 1 0 1 0 1]\n", + " [1 0 0 0 1 1]\n", + " [0 1 1 1 0 0]\n", + " [0 1 1 1 0 0]\n", + " [0 0 1 1 0 0]\n", + " [0 1 0 0 0 0]\n", + " [0 1 1 0 1 1]\n", + " [1 0 1 0 0 0]\n", + " [1 0 0 1 0 1]\n", + " [1 0 1 0 0 1]]\n", + "\n", + "Class labels:\n", + " [1 1 1 1 1 0 0 0 0 0]\n" + ] + } + ], + "source": [ + "import numpy as np\n", + "from mlxtend.data import make_multiplexer_dataset\n", + "\n", + "\n", + "X, y = make_multiplexer_dataset(address_bits=2, \n", + " sample_size=10,\n", + " positive_class_ratio=0.5, \n", + " shuffle=False,\n", + " random_seed=123)\n", + "\n", + "print('Features:\\n', X)\n", + "print('\\nClass labels:\\n', y)" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## API" + ] + }, + { + "cell_type": "code", + "execution_count": 2, + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "## make_multiplexer_dataset\n", + "\n", + "*make_multiplexer_dataset(address_bits=2, sample_size=100, positive_class_ratio=0.5, shuffle=False, random_seed=None)*\n", + "\n", + "Function to create a binary n-bit multiplexer dataset.\n", + "\n", + "New in mlxtend v0.9\n", + "\n", + "**Parameters**\n", + "\n", + "- `address_bits` : int (default: 2)\n", + "\n", + " A positive integer that determines the number of address\n", + " bits in the multiplexer, which in turn determine the\n", + " n-bit capacity of the multiplexer and therefore the\n", + " number of features. The number of features is determined by\n", + " the number of address bits. For example, 2 address bits\n", + " will result in a 6 bit multiplexer and consequently\n", + " 6 features (2 + 2^2 = 6). If `address_bits=3`, then\n", + " this results in an 11-bit multiplexer as (2 + 2^3 = 11)\n", + " with 11 features.\n", + "\n", + "\n", + "- `sample_size` : int (default: 100)\n", + "\n", + " The total number of samples generated.\n", + "\n", + "\n", + "- `positive_class_ratio` : float (default: 0.5)\n", + "\n", + " The fraction (a float between 0 and 1)\n", + " of samples in the `sample_size`d dataset\n", + " that have class label 1.\n", + " If `positive_class_ratio=0.5` (default), then\n", + " the ratio of class 0 and class 1 samples is perfectly balanced.\n", + "\n", + "\n", + "- `shuffle` : Bool (default: False)\n", + "\n", + " Whether or not to shuffle the features and labels.\n", + " If `False` (default), the samples are returned in sorted\n", + " order starting with `sample_size`/2 samples with class label 0\n", + " and followed by `sample_size`/2 samples with class label 1.\n", + "\n", + "\n", + "- `random_seed` : int (default: None)\n", + "\n", + " Random seed used for generating the multiplexer samples and shuffling.\n", + "\n", + "**Returns**\n", + "\n", + "- `X, y` : [n_samples, n_features], [n_class_labels]\n", + "\n", + " X is the feature matrix with the number of samples equal\n", + " to `sample_size`. The number of features is determined by\n", + " the number of address bits. For instance, 2 address bits\n", + " will result in a 6 bit multiplexer and consequently\n", + " 6 features (2 + 2^2 = 6).\n", + " All features are binary (values in {0, 1}).\n", + " y is a 1-dimensional array of class labels in {0, 1}.\n", + "\n", + "\n" + ] + } + ], + "source": [ + "with open('../../api_modules/mlxtend.data/make_multiplexer_dataset.md', 'r') as f:\n", + " s = f.read() \n", + "print(s)" + ] + } + ], + "metadata": { + "anaconda-cloud": {}, + "kernelspec": { + "display_name": "Python 3", + "language": "python", + "name": "python3" + }, + "language_info": { + "codemirror_mode": { + "name": "ipython", + "version": 3 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython3", + "version": "3.6.1" + } + }, + "nbformat": 4, + "nbformat_minor": 1 +} diff --git a/docs/sources/user_guide/data/make_multiplexer_dataset_data_files/6bit_multiplexer.pdf b/docs/sources/user_guide/data/make_multiplexer_dataset_data_files/6bit_multiplexer.pdf new file mode 100644 index 000000000..b0072ec18 Binary files /dev/null and b/docs/sources/user_guide/data/make_multiplexer_dataset_data_files/6bit_multiplexer.pdf differ diff --git a/docs/sources/user_guide/data/make_multiplexer_dataset_data_files/6bit_multiplexer.png b/docs/sources/user_guide/data/make_multiplexer_dataset_data_files/6bit_multiplexer.png new file mode 100644 index 000000000..45069e9f9 Binary files /dev/null and b/docs/sources/user_guide/data/make_multiplexer_dataset_data_files/6bit_multiplexer.png differ diff --git a/mlxtend/data/__init__.py b/mlxtend/data/__init__.py index aad9b546c..be71429b5 100644 --- a/mlxtend/data/__init__.py +++ b/mlxtend/data/__init__.py @@ -11,7 +11,9 @@ from .local_mnist import loadlocal_mnist from .boston_housing import boston_housing_data from .three_blobs import three_blobs_data +from .multiplexer import make_multiplexer_dataset __all__ = ["iris_data", "wine_data", "autompg_data", "loadlocal_mnist", "mnist_data", - "boston_housing_data", "three_blobs_data"] + "boston_housing_data", "three_blobs_data", + "make_multiplexer_dataset"] diff --git a/mlxtend/data/multiplexer.py b/mlxtend/data/multiplexer.py new file mode 100644 index 000000000..97aa419d3 --- /dev/null +++ b/mlxtend/data/multiplexer.py @@ -0,0 +1,110 @@ +# Sebastian Raschka 2014-2017 +# mlxtend Machine Learning Library Extensions +# +# A function for creating a multiplexer dataset for classification. +# Author: Sebastian Raschka +# +# License: BSD 3 clause + +import numpy as np + + +def make_multiplexer_dataset(address_bits=2, sample_size=100, + positive_class_ratio=0.5, shuffle=False, + random_seed=None): + """ + Function to create a binary n-bit multiplexer dataset. + + New in mlxtend v0.9 + + Parameters + --------------- + address_bits : int (default: 2) + A positive integer that determines the number of address + bits in the multiplexer, which in turn determine the + n-bit capacity of the multiplexer and therefore the + number of features. The number of features is determined by + the number of address bits. For example, 2 address bits + will result in a 6 bit multiplexer and consequently + 6 features (2 + 2^2 = 6). If `address_bits=3`, then + this results in an 11-bit multiplexer as (2 + 2^3 = 11) + with 11 features. + + sample_size : int (default: 100) + The total number of samples generated. + + positive_class_ratio : float (default: 0.5) + The fraction (a float between 0 and 1) + of samples in the `sample_size`d dataset + that have class label 1. + If `positive_class_ratio=0.5` (default), then + the ratio of class 0 and class 1 samples is perfectly balanced. + + shuffle : Bool (default: False) + Whether or not to shuffle the features and labels. + If `False` (default), the samples are returned in sorted + order starting with `sample_size`/2 samples with class label 0 + and followed by `sample_size`/2 samples with class label 1. + + random_seed : int (default: None) + Random seed used for generating the multiplexer samples and shuffling. + + Returns + -------- + X, y : [n_samples, n_features], [n_class_labels] + X is the feature matrix with the number of samples equal + to `sample_size`. The number of features is determined by + the number of address bits. For instance, 2 address bits + will result in a 6 bit multiplexer and consequently + 6 features (2 + 2^2 = 6). + All features are binary (values in {0, 1}). + y is a 1-dimensional array of class labels in {0, 1}. + + """ + + if not isinstance(address_bits, int): + raise AttributeError('address_bits' + ' must be an integer. Got %s.' % + type(address_bits)) + if address_bits < 1: + raise AttributeError('Number of address_bits' + ' must be greater than 0. Got %s.' % address_bits) + register_bits = 2**address_bits + total_bits = address_bits + register_bits + X_pos, y_pos = [], [] + X_neg, y_neg = [], [] + + # use numpy's instead of python's round because of consistent + # banker's rounding behavior across versions + n_positives = np.round(sample_size*positive_class_ratio).astype(np.int) + n_negatives = sample_size - n_positives + + rng = np.random.RandomState(random_seed) + + def gen_randsample(): + all_bits = [rng.randint(0, 2) for i in range(total_bits)] + address_str = ''.join(str(c) for c in all_bits[:address_bits]) + register_pos = int(address_str, base=2) + class_label = all_bits[address_bits:][register_pos] + return all_bits, class_label + + while len(y_pos) < n_positives or len(y_neg) < n_negatives: + + all_bits, class_label = gen_randsample() + + if class_label and len(y_pos) < n_positives: + X_pos.append(all_bits) + y_pos.append(class_label) + + elif not class_label and len(y_neg) < n_negatives: + X_neg.append(all_bits) + y_neg.append(class_label) + + X, y = X_pos + X_neg, y_pos + y_neg + X, y = np.array(X, dtype=np.int), np.array(y, dtype=np.int) + + if shuffle: + p = rng.permutation(y.shape[0]) + X, y = X[p], y[p] + + return X, y diff --git a/mlxtend/data/tests/test_data.py b/mlxtend/data/tests/test_datasets.py similarity index 87% rename from mlxtend/data/tests/test_data.py rename to mlxtend/data/tests/test_datasets.py index f28e3e24c..98150b542 100644 --- a/mlxtend/data/tests/test_data.py +++ b/mlxtend/data/tests/test_datasets.py @@ -1,3 +1,11 @@ +# Sebastian Raschka 2014-2017 +# mlxtend Machine Learning Library Extensions +# +# Author: Sebastian Raschka +# +# License: BSD 3 clause + + from mlxtend.data import iris_data from mlxtend.data import wine_data from mlxtend.data import autompg_data diff --git a/mlxtend/data/tests/test_multiplexer.py b/mlxtend/data/tests/test_multiplexer.py new file mode 100644 index 000000000..7faeb28ce --- /dev/null +++ b/mlxtend/data/tests/test_multiplexer.py @@ -0,0 +1,97 @@ +# Sebastian Raschka 2014-2017 +# mlxtend Machine Learning Library Extensions +# +# Author: Sebastian Raschka +# +# License: BSD 3 clause + + +import numpy as np +import sys +from mlxtend.data import make_multiplexer_dataset +from mlxtend.utils import assert_raises + + +def test_defaults(): + + X, y = make_multiplexer_dataset() + + assert X.shape == (100, 6), X.shape + assert X.dtype == np.int + assert y.shape == (100, ), y.shape + assert y.dtype == np.int + + +def test_invalid_address_bits(): + + msg_1 = "address_bits must be an integer. Got ." + + # for Python 2.7: + if sys.version_info[0] == 2: + msg_1 = msg_1.replace('