mirror of
https://github.com/Doctorado-ML/STree.git
synced 2025-08-17 16:36:01 +00:00
Compare commits
6 Commits
enhance-pa
...
complete-s
| Author | SHA1 | Date | |
|---|---|---|---|
0340584c52
|
|||
|
9b3c7ccdfa
|
|||
|
e4ac5075e5 | ||
|
|
36816074ff | ||
|
475ad7e752
|
|||
|
|
1c869e154e |
47
.github/workflows/main.yml
vendored
Normal file
47
.github/workflows/main.yml
vendored
Normal file
@@ -0,0 +1,47 @@
|
||||
name: CI
|
||||
|
||||
on:
|
||||
push:
|
||||
branches: [master]
|
||||
pull_request:
|
||||
branches: [master]
|
||||
workflow_dispatch:
|
||||
|
||||
jobs:
|
||||
build:
|
||||
runs-on: ${{ matrix.os }}
|
||||
strategy:
|
||||
matrix:
|
||||
os: [macos-latest, ubuntu-latest]
|
||||
python: [3.8]
|
||||
|
||||
steps:
|
||||
- uses: actions/checkout@v2
|
||||
- name: Set up Python ${{ matrix.python }}
|
||||
uses: actions/setup-python@v2
|
||||
with:
|
||||
python-version: ${{ matrix.python }}
|
||||
- name: Install dependencies
|
||||
run: |
|
||||
pip install -q --upgrade pip
|
||||
pip install -q -r requirements.txt
|
||||
pip install -q --upgrade codecov coverage black flake8 codacy-coverage
|
||||
- name: Lint
|
||||
run: |
|
||||
black --check --diff stree
|
||||
flake8 --count stree
|
||||
- name: Tests
|
||||
run: |
|
||||
coverage run -m unittest -v stree.tests
|
||||
coverage xml
|
||||
- name: Upload coverage to Codecov
|
||||
uses: codecov/codecov-action@v1
|
||||
with:
|
||||
token: ${{ secrets.CODECOV_TOKEN }}
|
||||
files: ./coverage.xml
|
||||
- name: Run codacy-coverage-reporter
|
||||
if: runner.os == 'Linux'
|
||||
uses: codacy/codacy-coverage-reporter-action@master
|
||||
with:
|
||||
project-token: ${{ secrets.CODACY_PROJECT_TOKEN }}
|
||||
coverage-reports: coverage.xml
|
||||
1
.gitignore
vendored
1
.gitignore
vendored
@@ -133,3 +133,4 @@ dmypy.json
|
||||
.pre-commit-config.yaml
|
||||
|
||||
**.csv
|
||||
.virtual_documents
|
||||
46
README.md
46
README.md
@@ -1,6 +1,6 @@
|
||||
[](https://app.codeship.com/projects/399170)
|
||||
|
||||
[](https://codecov.io/gh/doctorado-ml/stree)
|
||||
[](https://www.codacy.com/gh/Doctorado-ML/STree?utm_source=github.com&utm_medium=referral&utm_content=Doctorado-ML/STree&utm_campaign=Badge_Grade)
|
||||
[](https://www.codacy.com/gh/Doctorado-ML/STree?utm_source=github.com&utm_medium=referral&utm_content=Doctorado-ML/STree&utm_campaign=Badge_Grade)
|
||||
|
||||
# Stree
|
||||
|
||||
@@ -18,23 +18,45 @@ pip install git+https://github.com/doctorado-ml/stree
|
||||
|
||||
### Jupyter notebooks
|
||||
|
||||
* [](https://mybinder.org/v2/gh/Doctorado-ML/STree/master?urlpath=lab/tree/notebooks/benchmark.ipynb) Benchmark
|
||||
- [](https://mybinder.org/v2/gh/Doctorado-ML/STree/master?urlpath=lab/tree/notebooks/benchmark.ipynb) Benchmark
|
||||
|
||||
* [](https://colab.research.google.com/github/Doctorado-ML/STree/blob/master/notebooks/benchmark.ipynb) Benchmark
|
||||
- [](https://colab.research.google.com/github/Doctorado-ML/STree/blob/master/notebooks/benchmark.ipynb) Benchmark
|
||||
|
||||
* [](https://colab.research.google.com/github/Doctorado-ML/STree/blob/master/notebooks/features.ipynb) Test features
|
||||
- [](https://colab.research.google.com/github/Doctorado-ML/STree/blob/master/notebooks/features.ipynb) Test features
|
||||
|
||||
* [](https://colab.research.google.com/github/Doctorado-ML/STree/blob/master/notebooks/adaboost.ipynb) Adaboost
|
||||
- [](https://colab.research.google.com/github/Doctorado-ML/STree/blob/master/notebooks/adaboost.ipynb) Adaboost
|
||||
|
||||
* [](https://colab.research.google.com/github/Doctorado-ML/STree/blob/master/notebooks/gridsearch.ipynb) Gridsearch
|
||||
- [](https://colab.research.google.com/github/Doctorado-ML/STree/blob/master/notebooks/gridsearch.ipynb) Gridsearch
|
||||
|
||||
* [](https://colab.research.google.com/github/Doctorado-ML/STree/blob/master/notebooks/test_graphs.ipynb) Test Graphics
|
||||
- [](https://colab.research.google.com/github/Doctorado-ML/STree/blob/master/notebooks/test_graphs.ipynb) Test Graphics
|
||||
|
||||
### Command line
|
||||
## Hyperparameters
|
||||
|
||||
```bash
|
||||
python main.py
|
||||
```
|
||||
| | **Hyperparameter** | **Type/Values** | **Default** | **Meaning** |
|
||||
| --- | ------------------ | ------------------------------------------------------ | ----------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
|
||||
| \* | C | \<float\> | 1.0 | Regularization parameter. The strength of the regularization is inversely proportional to C. Must be strictly positive. |
|
||||
| \* | kernel | {"linear", "poly", "rbf"} | linear | Specifies the kernel type to be used in the algorithm. It must be one of ‘linear’, ‘poly’ or ‘rbf’. |
|
||||
| \* | max_iter | \<int\> | 1e5 | Hard limit on iterations within solver, or -1 for no limit. |
|
||||
| \* | random_state | \<int\> | None | Controls the pseudo random number generation for shuffling the data for probability estimates. Ignored when probability is False.<br>Pass an int for reproducible output across multiple function calls |
|
||||
| | max_depth | \<int\> | None | Specifies the maximum depth of the tree |
|
||||
| \* | tol | \<float\> | 1e-4 | Tolerance for stopping criterion. |
|
||||
| \* | degree | \<int\> | 3 | Degree of the polynomial kernel function (‘poly’). Ignored by all other kernels. |
|
||||
| \* | gamma | {"scale", "auto"} or \<float\> | scale | Kernel coefficient for ‘rbf’ and ‘poly’.<br>if gamma='scale' (default) is passed then it uses 1 / (n_features \* X.var()) as value of gamma,<br>if ‘auto’, uses 1 / n_features. |
|
||||
| | split_criteria | {"impurity", "max_samples"} | impurity | Decides (just in case of a multi class classification) which column (class) use to split the dataset in a node\*\* |
|
||||
| | criterion | {“gini”, “entropy”} | entropy | The function to measure the quality of a split (only used if max_features != num_features). <br>Supported criteria are “gini” for the Gini impurity and “entropy” for the information gain. |
|
||||
| | min_samples_split | \<int\> | 0 | The minimum number of samples required to split an internal node. 0 (default) for any |
|
||||
| | max_features | \<int\>, \<float\> <br><br>or {“auto”, “sqrt”, “log2”} | None | The number of features to consider when looking for the split:<br>If int, then consider max_features features at each split.<br>If float, then max_features is a fraction and int(max_features \* n_features) features are considered at each split.<br>If “auto”, then max_features=sqrt(n_features).<br>If “sqrt”, then max_features=sqrt(n_features).<br>If “log2”, then max_features=log2(n_features).<br>If None, then max_features=n_features. |
|
||||
| | splitter | {"best", "random"} | random | The strategy used to choose the feature set at each node (only used if max_features != num_features). <br>Supported strategies are “best” to choose the best feature set and “random” to choose a random combination. <br>The algorithm generates 5 candidates at most to choose from in both strategies. |
|
||||
|
||||
\* Hyperparameter used by the support vector classifier of every node
|
||||
|
||||
\*\* **Splitting in a STree node**
|
||||
|
||||
The decision function is applied to the dataset and distances from samples to hyperplanes are computed in a matrix. This matrix has as many columns as classes the samples belongs to (if more than two, i.e. multiclass classification) or 1 column if it's a binary class dataset. In binary classification only one hyperplane is computed and therefore only one column is needed to store the distances of the samples to it. If three or more classes are present in the dataset we need as many hyperplanes as classes are there, and therefore one column per hyperplane is needed.
|
||||
|
||||
In case of multiclass classification we have to decide which column take into account to make the split, that depends on hyperparameter _split_criteria_, if "impurity" is chosen then STree computes information gain of every split candidate using each column and chooses the one that maximize the information gain, otherwise STree choses the column with more samples with a predicted class (the column with more positive numbers in it).
|
||||
|
||||
Once we have the column to take into account for the split, the algorithm splits samples with positive distances to hyperplane from the rest.
|
||||
|
||||
## Tests
|
||||
|
||||
|
||||
2
main.py
2
main.py
@@ -8,7 +8,7 @@ random_state = 1
|
||||
X, y = load_iris(return_X_y=True)
|
||||
|
||||
Xtrain, Xtest, ytrain, ytest = train_test_split(
|
||||
X, y, test_size=0.2, random_state=random_state
|
||||
X, y, test_size=0.3, random_state=random_state
|
||||
)
|
||||
|
||||
now = time.time()
|
||||
|
||||
File diff suppressed because one or more lines are too long
@@ -34,9 +34,12 @@
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"import time\n",
|
||||
"import warnings\n",
|
||||
"from sklearn.ensemble import AdaBoostClassifier, BaggingClassifier\n",
|
||||
"from sklearn.model_selection import train_test_split\n",
|
||||
"from stree import Stree"
|
||||
"from sklearn.exceptions import ConvergenceWarning\n",
|
||||
"from stree import Stree\n",
|
||||
"warnings.filterwarnings(\"ignore\", category=ConvergenceWarning)"
|
||||
]
|
||||
},
|
||||
{
|
||||
@@ -59,9 +62,15 @@
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
"output_type": "stream",
|
||||
"name": "stdout",
|
||||
"text": "Fraud: 0.173% 492\nValid: 99.827% 284315\nX.shape (100492, 28) y.shape (100492,)\nFraud: 0.644% 647\nValid: 99.356% 99845\n"
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"Fraud: 0.173% 492\n",
|
||||
"Valid: 99.827% 284315\n",
|
||||
"X.shape (100492, 28) y.shape (100492,)\n",
|
||||
"Fraud: 0.651% 654\n",
|
||||
"Valid: 99.349% 99838\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
@@ -127,14 +136,18 @@
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
"output_type": "stream",
|
||||
"name": "stdout",
|
||||
"text": "Score Train: 0.9985784146480154\nScore Test: 0.9981093273185617\nTook 73.27 seconds\n"
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"Score Train: 0.9984504719663368\n",
|
||||
"Score Test: 0.9983415151917209\n",
|
||||
"Took 26.09 seconds\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"now = time.time()\n",
|
||||
"clf = Stree(max_depth=3, random_state=random_state)\n",
|
||||
"clf = Stree(max_depth=3, random_state=random_state, max_iter=1e3)\n",
|
||||
"clf.fit(Xtrain, ytrain)\n",
|
||||
"print(\"Score Train: \", clf.score(Xtrain, ytrain))\n",
|
||||
"print(\"Score Test: \", clf.score(Xtest, ytest))\n",
|
||||
@@ -167,15 +180,19 @@
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
"output_type": "stream",
|
||||
"name": "stdout",
|
||||
"text": "Kernel: linear\tTime: 93.78 seconds\tScore Train: 0.9983083\tScore Test: 0.9983083\nKernel: rbf\tTime: 18.32 seconds\tScore Train: 0.9935602\tScore Test: 0.9935651\nKernel: poly\tTime: 69.68 seconds\tScore Train: 0.9973132\tScore Test: 0.9972801\n"
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"Kernel: linear\tTime: 43.49 seconds\tScore Train: 0.9980098\tScore Test: 0.9980762\n",
|
||||
"Kernel: rbf\tTime: 8.86 seconds\tScore Train: 0.9934891\tScore Test: 0.9934987\n",
|
||||
"Kernel: poly\tTime: 41.14 seconds\tScore Train: 0.9972279\tScore Test: 0.9973133\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"for kernel in ['linear', 'rbf', 'poly']:\n",
|
||||
" now = time.time()\n",
|
||||
" clf = AdaBoostClassifier(base_estimator=Stree(C=C, kernel=kernel, max_depth=max_depth, random_state=random_state), algorithm=\"SAMME\", n_estimators=n_estimators, random_state=random_state)\n",
|
||||
" clf = AdaBoostClassifier(base_estimator=Stree(C=C, kernel=kernel, max_depth=max_depth, random_state=random_state, max_iter=1e3), algorithm=\"SAMME\", n_estimators=n_estimators, random_state=random_state)\n",
|
||||
" clf.fit(Xtrain, ytrain)\n",
|
||||
" score_train = clf.score(Xtrain, ytrain)\n",
|
||||
" score_test = clf.score(Xtest, ytest)\n",
|
||||
@@ -208,15 +225,19 @@
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
"output_type": "stream",
|
||||
"name": "stdout",
|
||||
"text": "Kernel: linear\tTime: 387.06 seconds\tScore Train: 0.9985784\tScore Test: 0.9981093\nKernel: rbf\tTime: 144.00 seconds\tScore Train: 0.9992750\tScore Test: 0.9983415\nKernel: poly\tTime: 101.78 seconds\tScore Train: 0.9992466\tScore Test: 0.9981757\n"
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"Kernel: linear\tTime: 187.51 seconds\tScore Train: 0.9984505\tScore Test: 0.9983083\n",
|
||||
"Kernel: rbf\tTime: 73.65 seconds\tScore Train: 0.9993461\tScore Test: 0.9985074\n",
|
||||
"Kernel: poly\tTime: 52.19 seconds\tScore Train: 0.9993461\tScore Test: 0.9987727\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"for kernel in ['linear', 'rbf', 'poly']:\n",
|
||||
" now = time.time()\n",
|
||||
" clf = BaggingClassifier(base_estimator=Stree(C=C, kernel=kernel, max_depth=max_depth, random_state=random_state), n_estimators=n_estimators, random_state=random_state)\n",
|
||||
" clf = BaggingClassifier(base_estimator=Stree(C=C, kernel=kernel, max_depth=max_depth, random_state=random_state, max_iter=1e3), n_estimators=n_estimators, random_state=random_state)\n",
|
||||
" clf.fit(Xtrain, ytrain)\n",
|
||||
" score_train = clf.score(Xtrain, ytrain)\n",
|
||||
" score_test = clf.score(Xtest, ytest)\n",
|
||||
@@ -225,6 +246,11 @@
|
||||
}
|
||||
],
|
||||
"metadata": {
|
||||
"kernelspec": {
|
||||
"display_name": "Python 3",
|
||||
"language": "python",
|
||||
"name": "python3"
|
||||
},
|
||||
"language_info": {
|
||||
"codemirror_mode": {
|
||||
"name": "ipython",
|
||||
@@ -235,14 +261,9 @@
|
||||
"name": "python",
|
||||
"nbconvert_exporter": "python",
|
||||
"pygments_lexer": "ipython3",
|
||||
"version": "3.7.6-final"
|
||||
},
|
||||
"orig_nbformat": 2,
|
||||
"kernelspec": {
|
||||
"name": "python37664bitgeneralvenve3128601eb614c5da59c5055670b6040",
|
||||
"display_name": "Python 3.7.6 64-bit ('general': venv)"
|
||||
"version": "3.8.2"
|
||||
}
|
||||
},
|
||||
"nbformat": 4,
|
||||
"nbformat_minor": 2
|
||||
"nbformat_minor": 4
|
||||
}
|
||||
File diff suppressed because one or more lines are too long
@@ -30,45 +30,59 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 2,
|
||||
"metadata": {
|
||||
"id": "zIHKVxthDZEa",
|
||||
"colab": {},
|
||||
"colab_type": "code",
|
||||
"colab": {}
|
||||
"id": "zIHKVxthDZEa"
|
||||
},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"from sklearn.ensemble import AdaBoostClassifier\n",
|
||||
"from sklearn.svm import LinearSVC\n",
|
||||
"from sklearn.model_selection import GridSearchCV, train_test_split\n",
|
||||
"from stree import Stree"
|
||||
],
|
||||
"execution_count": 2,
|
||||
"outputs": []
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 3,
|
||||
"metadata": {
|
||||
"id": "IEmq50QgDZEi",
|
||||
"colab": {},
|
||||
"colab_type": "code",
|
||||
"colab": {}
|
||||
"id": "IEmq50QgDZEi"
|
||||
},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"import os\n",
|
||||
"if not os.path.isfile('data/creditcard.csv'):\n",
|
||||
" !wget --no-check-certificate --content-disposition http://nube.jccm.es/index.php/s/Zs7SYtZQJ3RQ2H2/download\n",
|
||||
" !tar xzf creditcard.tgz"
|
||||
],
|
||||
"execution_count": 3,
|
||||
"outputs": []
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 4,
|
||||
"metadata": {
|
||||
"id": "z9Q-YUfBDZEq",
|
||||
"colab_type": "code",
|
||||
"colab": {},
|
||||
"colab_type": "code",
|
||||
"id": "z9Q-YUfBDZEq",
|
||||
"outputId": "afc822fb-f16a-4302-8a67-2b9e2880159b",
|
||||
"tags": []
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stdout",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"Fraud: 0.173% 492\n",
|
||||
"Valid: 99.827% 284315\n",
|
||||
"X.shape (1492, 28) y.shape (1492,)\n",
|
||||
"Fraud: 33.177% 495\n",
|
||||
"Valid: 66.823% 997\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"random_state=1\n",
|
||||
"\n",
|
||||
@@ -107,14 +121,6 @@
|
||||
"Xtest = data[1]\n",
|
||||
"ytrain = data[2]\n",
|
||||
"ytest = data[3]"
|
||||
],
|
||||
"execution_count": 4,
|
||||
"outputs": [
|
||||
{
|
||||
"output_type": "stream",
|
||||
"name": "stdout",
|
||||
"text": "Fraud: 0.173% 492\nValid: 99.827% 284315\nX.shape (1492, 28) y.shape (1492,)\nFraud: 32.976% 492\nValid: 67.024% 1000\n"
|
||||
}
|
||||
]
|
||||
},
|
||||
{
|
||||
@@ -126,24 +132,47 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 5,
|
||||
"metadata": {
|
||||
"id": "HmX3kR4PDZEw",
|
||||
"colab": {},
|
||||
"colab_type": "code",
|
||||
"colab": {}
|
||||
"id": "HmX3kR4PDZEw"
|
||||
},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"parameters = {\n",
|
||||
" 'base_estimator': [Stree()],\n",
|
||||
"parameters = [{\n",
|
||||
" 'base_estimator': [Stree(random_state=random_state)],\n",
|
||||
" 'n_estimators': [10, 25],\n",
|
||||
" 'learning_rate': [.5, 1],\n",
|
||||
" 'base_estimator__split_criteria': ['max_samples', 'impurity'],\n",
|
||||
" 'base_estimator__tol': [.1, 1e-02],\n",
|
||||
" 'base_estimator__max_depth': [3, 5],\n",
|
||||
" 'base_estimator__C': [7, 55],\n",
|
||||
" 'base_estimator__kernel': ['linear', 'poly', 'rbf']\n",
|
||||
"}"
|
||||
],
|
||||
"execution_count": 5,
|
||||
"outputs": []
|
||||
" 'base_estimator__max_depth': [3, 5, 7],\n",
|
||||
" 'base_estimator__C': [1, 7, 55],\n",
|
||||
" 'base_estimator__kernel': ['linear']\n",
|
||||
"},\n",
|
||||
"{\n",
|
||||
" 'base_estimator': [Stree(random_state=random_state)],\n",
|
||||
" 'n_estimators': [10, 25],\n",
|
||||
" 'learning_rate': [.5, 1],\n",
|
||||
" 'base_estimator__split_criteria': ['max_samples', 'impurity'],\n",
|
||||
" 'base_estimator__tol': [.1, 1e-02],\n",
|
||||
" 'base_estimator__max_depth': [3, 5, 7],\n",
|
||||
" 'base_estimator__C': [1, 7, 55],\n",
|
||||
" 'base_estimator__degree': [3, 5, 7],\n",
|
||||
" 'base_estimator__kernel': ['poly']\n",
|
||||
"},\n",
|
||||
"{\n",
|
||||
" 'base_estimator': [Stree(random_state=random_state)],\n",
|
||||
" 'n_estimators': [10, 25],\n",
|
||||
" 'learning_rate': [.5, 1],\n",
|
||||
" 'base_estimator__split_criteria': ['max_samples', 'impurity'],\n",
|
||||
" 'base_estimator__tol': [.1, 1e-02],\n",
|
||||
" 'base_estimator__max_depth': [3, 5, 7],\n",
|
||||
" 'base_estimator__C': [1, 7, 55],\n",
|
||||
" 'base_estimator__gamma': [.1, 1, 10],\n",
|
||||
" 'base_estimator__kernel': ['rbf']\n",
|
||||
"}]"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
@@ -151,12 +180,26 @@
|
||||
"metadata": {},
|
||||
"outputs": [
|
||||
{
|
||||
"output_type": "execute_result",
|
||||
"data": {
|
||||
"text/plain": "{'C': 1.0,\n 'criterion': 'gini',\n 'degree': 3,\n 'gamma': 'scale',\n 'kernel': 'linear',\n 'max_depth': None,\n 'max_features': None,\n 'max_iter': 1000,\n 'min_samples_split': 0,\n 'random_state': None,\n 'split_criteria': 'max_samples',\n 'splitter': 'random',\n 'tol': 0.0001}"
|
||||
"text/plain": [
|
||||
"{'C': 1.0,\n",
|
||||
" 'criterion': 'entropy',\n",
|
||||
" 'degree': 3,\n",
|
||||
" 'gamma': 'scale',\n",
|
||||
" 'kernel': 'linear',\n",
|
||||
" 'max_depth': None,\n",
|
||||
" 'max_features': None,\n",
|
||||
" 'max_iter': 100000.0,\n",
|
||||
" 'min_samples_split': 0,\n",
|
||||
" 'random_state': None,\n",
|
||||
" 'split_criteria': 'impurity',\n",
|
||||
" 'splitter': 'random',\n",
|
||||
" 'tol': 0.0001}"
|
||||
]
|
||||
},
|
||||
"execution_count": 6,
|
||||
"metadata": {},
|
||||
"execution_count": 6
|
||||
"output_type": "execute_result"
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
@@ -165,61 +208,142 @@
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 7,
|
||||
"metadata": {
|
||||
"id": "CrcB8o6EDZE5",
|
||||
"colab_type": "code",
|
||||
"colab": {},
|
||||
"colab_type": "code",
|
||||
"id": "CrcB8o6EDZE5",
|
||||
"outputId": "7703413a-d563-4289-a13b-532f38f82762",
|
||||
"tags": []
|
||||
},
|
||||
"source": [
|
||||
"random_state=2020\n",
|
||||
"clf = AdaBoostClassifier(random_state=random_state, algorithm=\"SAMME\")\n",
|
||||
"grid = GridSearchCV(clf, parameters, verbose=10, n_jobs=-1, return_train_score=True)\n",
|
||||
"grid.fit(Xtrain, ytrain)"
|
||||
],
|
||||
"execution_count": 7,
|
||||
"outputs": [
|
||||
{
|
||||
"output_type": "stream",
|
||||
"name": "stdout",
|
||||
"text": "Fitting 5 folds for each of 96 candidates, totalling 480 fits\n[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n[Parallel(n_jobs=-1)]: Done 2 tasks | elapsed: 2.0s\n[Parallel(n_jobs=-1)]: Done 9 tasks | elapsed: 2.4s\n[Parallel(n_jobs=-1)]: Done 16 tasks | elapsed: 2.7s\n[Parallel(n_jobs=-1)]: Done 25 tasks | elapsed: 3.3s\n[Parallel(n_jobs=-1)]: Done 34 tasks | elapsed: 4.3s\n[Parallel(n_jobs=-1)]: Done 45 tasks | elapsed: 5.3s\n[Parallel(n_jobs=-1)]: Done 56 tasks | elapsed: 6.6s\n[Parallel(n_jobs=-1)]: Done 69 tasks | elapsed: 8.1s\n[Parallel(n_jobs=-1)]: Done 82 tasks | elapsed: 9.4s\n[Parallel(n_jobs=-1)]: Done 97 tasks | elapsed: 10.1s\n[Parallel(n_jobs=-1)]: Done 112 tasks | elapsed: 11.1s\n[Parallel(n_jobs=-1)]: Done 129 tasks | elapsed: 12.3s\n[Parallel(n_jobs=-1)]: Done 146 tasks | elapsed: 13.6s\n[Parallel(n_jobs=-1)]: Done 165 tasks | elapsed: 14.9s\n[Parallel(n_jobs=-1)]: Done 184 tasks | elapsed: 16.2s\n[Parallel(n_jobs=-1)]: Done 205 tasks | elapsed: 17.6s\n[Parallel(n_jobs=-1)]: Done 226 tasks | elapsed: 19.1s\n[Parallel(n_jobs=-1)]: Done 249 tasks | elapsed: 21.6s\n[Parallel(n_jobs=-1)]: Done 272 tasks | elapsed: 25.9s\n[Parallel(n_jobs=-1)]: Done 297 tasks | elapsed: 30.4s\n[Parallel(n_jobs=-1)]: Done 322 tasks | elapsed: 36.7s\n[Parallel(n_jobs=-1)]: Done 349 tasks | elapsed: 38.1s\n[Parallel(n_jobs=-1)]: Done 376 tasks | elapsed: 39.6s\n[Parallel(n_jobs=-1)]: Done 405 tasks | elapsed: 41.9s\n[Parallel(n_jobs=-1)]: Done 434 tasks | elapsed: 44.9s\n[Parallel(n_jobs=-1)]: Done 465 tasks | elapsed: 48.2s\n[Parallel(n_jobs=-1)]: Done 480 out of 480 | elapsed: 49.2s finished\n"
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"Fitting 5 folds for each of 1008 candidates, totalling 5040 fits\n"
|
||||
]
|
||||
},
|
||||
{
|
||||
"output_type": "execute_result",
|
||||
"data": {
|
||||
"text/plain": "GridSearchCV(estimator=AdaBoostClassifier(algorithm='SAMME', random_state=2020),\n n_jobs=-1,\n param_grid={'base_estimator': [Stree(C=55, max_depth=3, tol=0.01)],\n 'base_estimator__C': [7, 55],\n 'base_estimator__kernel': ['linear', 'poly', 'rbf'],\n 'base_estimator__max_depth': [3, 5],\n 'base_estimator__tol': [0.1, 0.01],\n 'learning_rate': [0.5, 1], 'n_estimators': [10, 25]},\n return_train_score=True, verbose=10)"
|
||||
"name": "stderr",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"[Parallel(n_jobs=-1)]: Using backend LokyBackend with 16 concurrent workers.\n",
|
||||
"[Parallel(n_jobs=-1)]: Done 40 tasks | elapsed: 1.6s\n",
|
||||
"[Parallel(n_jobs=-1)]: Done 130 tasks | elapsed: 3.1s\n",
|
||||
"[Parallel(n_jobs=-1)]: Done 256 tasks | elapsed: 5.5s\n",
|
||||
"[Parallel(n_jobs=-1)]: Done 418 tasks | elapsed: 9.3s\n",
|
||||
"[Parallel(n_jobs=-1)]: Done 616 tasks | elapsed: 18.6s\n",
|
||||
"[Parallel(n_jobs=-1)]: Done 850 tasks | elapsed: 28.2s\n",
|
||||
"[Parallel(n_jobs=-1)]: Done 1120 tasks | elapsed: 35.4s\n",
|
||||
"[Parallel(n_jobs=-1)]: Done 1426 tasks | elapsed: 43.5s\n",
|
||||
"[Parallel(n_jobs=-1)]: Done 1768 tasks | elapsed: 51.3s\n",
|
||||
"[Parallel(n_jobs=-1)]: Done 2146 tasks | elapsed: 1.0min\n",
|
||||
"[Parallel(n_jobs=-1)]: Done 2560 tasks | elapsed: 1.2min\n",
|
||||
"[Parallel(n_jobs=-1)]: Done 3010 tasks | elapsed: 1.4min\n",
|
||||
"[Parallel(n_jobs=-1)]: Done 3496 tasks | elapsed: 1.7min\n",
|
||||
"[Parallel(n_jobs=-1)]: Done 4018 tasks | elapsed: 2.1min\n",
|
||||
"[Parallel(n_jobs=-1)]: Done 4576 tasks | elapsed: 2.6min\n",
|
||||
"[Parallel(n_jobs=-1)]: Done 5040 out of 5040 | elapsed: 2.9min finished\n"
|
||||
]
|
||||
},
|
||||
{
|
||||
"data": {
|
||||
"text/plain": [
|
||||
"GridSearchCV(estimator=AdaBoostClassifier(algorithm='SAMME', random_state=1),\n",
|
||||
" n_jobs=-1,\n",
|
||||
" param_grid=[{'base_estimator': [Stree(C=55, max_depth=7,\n",
|
||||
" random_state=1,\n",
|
||||
" split_criteria='max_samples',\n",
|
||||
" tol=0.1)],\n",
|
||||
" 'base_estimator__C': [1, 7, 55],\n",
|
||||
" 'base_estimator__kernel': ['linear'],\n",
|
||||
" 'base_estimator__max_depth': [3, 5, 7],\n",
|
||||
" 'base_estimator__split_criteria': ['max_samples',\n",
|
||||
" 'impuri...\n",
|
||||
" {'base_estimator': [Stree(random_state=1)],\n",
|
||||
" 'base_estimator__C': [1, 7, 55],\n",
|
||||
" 'base_estimator__gamma': [0.1, 1, 10],\n",
|
||||
" 'base_estimator__kernel': ['rbf'],\n",
|
||||
" 'base_estimator__max_depth': [3, 5, 7],\n",
|
||||
" 'base_estimator__split_criteria': ['max_samples',\n",
|
||||
" 'impurity'],\n",
|
||||
" 'base_estimator__tol': [0.1, 0.01],\n",
|
||||
" 'learning_rate': [0.5, 1],\n",
|
||||
" 'n_estimators': [10, 25]}],\n",
|
||||
" return_train_score=True, verbose=5)"
|
||||
]
|
||||
},
|
||||
"execution_count": 7,
|
||||
"metadata": {},
|
||||
"execution_count": 7
|
||||
"output_type": "execute_result"
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"clf = AdaBoostClassifier(random_state=random_state, algorithm=\"SAMME\")\n",
|
||||
"grid = GridSearchCV(clf, parameters, verbose=5, n_jobs=-1, return_train_score=True)\n",
|
||||
"grid.fit(Xtrain, ytrain)"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 8,
|
||||
"metadata": {
|
||||
"id": "ZjX88NoYDZE8",
|
||||
"colab_type": "code",
|
||||
"colab": {},
|
||||
"colab_type": "code",
|
||||
"id": "ZjX88NoYDZE8",
|
||||
"outputId": "285163c8-fa33-4915-8ae7-61c4f7844344",
|
||||
"tags": []
|
||||
},
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stdout",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"Best estimator: AdaBoostClassifier(algorithm='SAMME',\n",
|
||||
" base_estimator=Stree(C=55, max_depth=7, random_state=1,\n",
|
||||
" split_criteria='max_samples', tol=0.1),\n",
|
||||
" learning_rate=0.5, n_estimators=25, random_state=1)\n",
|
||||
"Best hyperparameters: {'base_estimator': Stree(C=55, max_depth=7, random_state=1, split_criteria='max_samples', tol=0.1), 'base_estimator__C': 55, 'base_estimator__kernel': 'linear', 'base_estimator__max_depth': 7, 'base_estimator__split_criteria': 'max_samples', 'base_estimator__tol': 0.1, 'learning_rate': 0.5, 'n_estimators': 25}\n",
|
||||
"Best accuracy: 0.9511777695988222\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"print(\"Best estimator: \", grid.best_estimator_)\n",
|
||||
"print(\"Best hyperparameters: \", grid.best_params_)\n",
|
||||
"print(\"Best accuracy: \", grid.best_score_)"
|
||||
],
|
||||
"execution_count": 8,
|
||||
"outputs": [
|
||||
]
|
||||
},
|
||||
{
|
||||
"output_type": "stream",
|
||||
"name": "stdout",
|
||||
"text": "Best estimator: AdaBoostClassifier(algorithm='SAMME',\n base_estimator=Stree(C=55, max_depth=3, tol=0.01),\n learning_rate=0.5, n_estimators=25, random_state=2020)\nBest hyperparameters: {'base_estimator': Stree(C=55, max_depth=3, tol=0.01), 'base_estimator__C': 55, 'base_estimator__kernel': 'linear', 'base_estimator__max_depth': 3, 'base_estimator__tol': 0.01, 'learning_rate': 0.5, 'n_estimators': 25}\nBest accuracy: 0.9559440559440558\n"
|
||||
}
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Best estimator: AdaBoostClassifier(algorithm='SAMME',\n",
|
||||
" base_estimator=Stree(C=55, max_depth=7, random_state=1,\n",
|
||||
" split_criteria='max_samples', tol=0.1),\n",
|
||||
" learning_rate=0.5, n_estimators=25, random_state=1)\n",
|
||||
"Best hyperparameters: {'base_estimator': Stree(C=55, max_depth=7, random_state=1, split_criteria='max_samples', tol=0.1), 'base_estimator__C': 55, 'base_estimator__kernel': 'linear', 'base_estimator__max_depth': 7, 'base_estimator__split_criteria': 'max_samples', 'base_estimator__tol': 0.1, 'learning_rate': 0.5, 'n_estimators': 25}"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "markdown",
|
||||
"metadata": {},
|
||||
"source": [
|
||||
"Best accuracy: 0.9511777695988222"
|
||||
]
|
||||
}
|
||||
],
|
||||
"metadata": {
|
||||
"colab": {
|
||||
"name": "gridsearch.ipynb",
|
||||
"provenance": []
|
||||
},
|
||||
"kernelspec": {
|
||||
"display_name": "Python 3",
|
||||
"language": "python",
|
||||
"name": "python3"
|
||||
},
|
||||
"language_info": {
|
||||
"codemirror_mode": {
|
||||
"name": "ipython",
|
||||
@@ -230,18 +354,9 @@
|
||||
"name": "python",
|
||||
"nbconvert_exporter": "python",
|
||||
"pygments_lexer": "ipython3",
|
||||
"version": "3.7.6-final"
|
||||
},
|
||||
"orig_nbformat": 2,
|
||||
"kernelspec": {
|
||||
"name": "python37664bitgeneralvenvfbd0a23e74cf4e778460f5ffc6761f39",
|
||||
"display_name": "Python 3.7.6 64-bit ('general': venv)"
|
||||
},
|
||||
"colab": {
|
||||
"name": "gridsearch.ipynb",
|
||||
"provenance": []
|
||||
"version": "3.8.2"
|
||||
}
|
||||
},
|
||||
"nbformat": 4,
|
||||
"nbformat_minor": 0
|
||||
"nbformat_minor": 4
|
||||
}
|
||||
@@ -1,4 +1,4 @@
|
||||
numpy
|
||||
scikit-learn
|
||||
scikit-learn==0.23.2
|
||||
pandas
|
||||
ipympl
|
||||
1
runtime.txt
Normal file
1
runtime.txt
Normal file
@@ -0,0 +1 @@
|
||||
python-3.8
|
||||
6
setup.py
6
setup.py
@@ -1,6 +1,6 @@
|
||||
import setuptools
|
||||
|
||||
__version__ = "0.9rc5"
|
||||
__version__ = "0.9rc6"
|
||||
__author__ = "Ricardo Montañana Gómez"
|
||||
|
||||
|
||||
@@ -25,12 +25,12 @@ setuptools.setup(
|
||||
classifiers=[
|
||||
"Development Status :: 4 - Beta",
|
||||
"License :: OSI Approved :: MIT License",
|
||||
"Programming Language :: Python :: 3.7",
|
||||
"Programming Language :: Python :: 3.8",
|
||||
"Natural Language :: English",
|
||||
"Topic :: Scientific/Engineering :: Artificial Intelligence",
|
||||
"Intended Audience :: Science/Research",
|
||||
],
|
||||
install_requires=["scikit-learn>=0.23.0", "numpy", "ipympl"],
|
||||
install_requires=["scikit-learn==0.23.2", "numpy", "ipympl"],
|
||||
test_suite="stree.tests",
|
||||
zip_safe=False,
|
||||
)
|
||||
|
||||
450
stree/Strees.py
450
stree/Strees.py
@@ -3,15 +3,15 @@ __author__ = "Ricardo Montañana Gómez"
|
||||
__copyright__ = "Copyright 2020, Ricardo Montañana Gómez"
|
||||
__license__ = "MIT"
|
||||
__version__ = "0.9"
|
||||
Build an oblique tree classifier based on SVM Trees
|
||||
Build an oblique tree classifier based on SVM nodes
|
||||
"""
|
||||
|
||||
import os
|
||||
import numbers
|
||||
import random
|
||||
import warnings
|
||||
from math import log
|
||||
from itertools import combinations
|
||||
from math import log, factorial
|
||||
from typing import Optional
|
||||
import numpy as np
|
||||
from sklearn.base import BaseEstimator, ClassifierMixin
|
||||
from sklearn.svm import SVC, LinearSVC
|
||||
@@ -57,6 +57,7 @@ class Snode:
|
||||
)
|
||||
self._features = features
|
||||
self._impurity = impurity
|
||||
self._partition_column: int = -1
|
||||
|
||||
@classmethod
|
||||
def copy(cls, node: "Snode") -> "Snode":
|
||||
@@ -69,6 +70,12 @@ class Snode:
|
||||
node._title,
|
||||
)
|
||||
|
||||
def set_partition_column(self, col: int):
|
||||
self._partition_column = col
|
||||
|
||||
def get_partition_column(self) -> int:
|
||||
return self._partition_column
|
||||
|
||||
def set_down(self, son):
|
||||
self._down = son
|
||||
|
||||
@@ -93,9 +100,8 @@ class Snode:
|
||||
classes, card = np.unique(self._y, return_counts=True)
|
||||
if len(classes) > 1:
|
||||
max_card = max(card)
|
||||
min_card = min(card)
|
||||
self._class = classes[card == max_card][0]
|
||||
self._belief = max_card / (max_card + min_card)
|
||||
self._belief = max_card / np.sum(card)
|
||||
else:
|
||||
self._belief = 1
|
||||
try:
|
||||
@@ -104,24 +110,23 @@ class Snode:
|
||||
self._class = None
|
||||
|
||||
def __str__(self) -> str:
|
||||
if self.is_leaf():
|
||||
count_values = np.unique(self._y, return_counts=True)
|
||||
result = (
|
||||
if self.is_leaf():
|
||||
return (
|
||||
f"{self._title} - Leaf class={self._class} belief="
|
||||
f"{self._belief: .6f} impurity={self._impurity:.4f} "
|
||||
f"counts={count_values}"
|
||||
)
|
||||
return result
|
||||
else:
|
||||
return (
|
||||
f"{self._title} feaures={self._features} impurity="
|
||||
f"{self._impurity:.4f} "
|
||||
f"counts={count_values}"
|
||||
)
|
||||
|
||||
|
||||
class Siterator:
|
||||
"""Stree preorder iterator
|
||||
"""
|
||||
"""Stree preorder iterator"""
|
||||
|
||||
def __init__(self, tree: Snode):
|
||||
self._stack = []
|
||||
@@ -167,20 +172,22 @@ class Splitter:
|
||||
f"criterion must be gini or entropy got({criterion})"
|
||||
)
|
||||
|
||||
if criteria not in ["min_distance", "max_samples", "max_distance"]:
|
||||
if criteria not in [
|
||||
"max_samples",
|
||||
"impurity",
|
||||
]:
|
||||
raise ValueError(
|
||||
"split_criteria has to be min_distance "
|
||||
f"max_distance or max_samples got ({criteria})"
|
||||
f"criteria has to be max_samples or impurity; got ({criteria})"
|
||||
)
|
||||
|
||||
if splitter_type not in ["random", "best"]:
|
||||
raise ValueError(
|
||||
f"splitter must be either random or best got({splitter_type})"
|
||||
f"splitter must be either random or best, got({splitter_type})"
|
||||
)
|
||||
self.criterion_function = getattr(self, f"_{self._criterion}")
|
||||
self.decision_criteria = getattr(self, f"_{self._criteria}")
|
||||
|
||||
def impurity(self, y: np.array) -> np.array:
|
||||
def partition_impurity(self, y: np.array) -> np.array:
|
||||
return self.criterion_function(y)
|
||||
|
||||
@staticmethod
|
||||
@@ -190,6 +197,18 @@ class Splitter:
|
||||
|
||||
@staticmethod
|
||||
def _entropy(y: np.array) -> float:
|
||||
"""Compute entropy of a labels set
|
||||
|
||||
Parameters
|
||||
----------
|
||||
y : np.array
|
||||
set of labels
|
||||
|
||||
Returns
|
||||
-------
|
||||
float
|
||||
entropy
|
||||
"""
|
||||
n_labels = len(y)
|
||||
if n_labels <= 1:
|
||||
return 0
|
||||
@@ -208,6 +227,22 @@ class Splitter:
|
||||
def information_gain(
|
||||
self, labels: np.array, labels_up: np.array, labels_dn: np.array
|
||||
) -> float:
|
||||
"""Compute information gain of a split candidate
|
||||
|
||||
Parameters
|
||||
----------
|
||||
labels : np.array
|
||||
labels of the dataset
|
||||
labels_up : np.array
|
||||
labels of one side
|
||||
labels_dn : np.array
|
||||
labels on the other side
|
||||
|
||||
Returns
|
||||
-------
|
||||
float
|
||||
information gain
|
||||
"""
|
||||
imp_prev = self.criterion_function(labels)
|
||||
card_up = card_dn = imp_up = imp_dn = 0
|
||||
if labels_up is not None:
|
||||
@@ -238,7 +273,7 @@ class Splitter:
|
||||
node = Snode(
|
||||
self._clf, dataset, labels, feature_set, 0.0, "subset"
|
||||
)
|
||||
self.partition(dataset, node)
|
||||
self.partition(dataset, node, train=True)
|
||||
y1, y2 = self.part(labels)
|
||||
gain = self.information_gain(labels, y1, y2)
|
||||
if gain > max_gain:
|
||||
@@ -246,124 +281,206 @@ class Splitter:
|
||||
selected = feature_set
|
||||
return selected if selected is not None else feature_set
|
||||
|
||||
@staticmethod
|
||||
def _generate_spaces(features: int, max_features: int) -> list:
|
||||
"""Generate at most 5 feature random combinations
|
||||
|
||||
Parameters
|
||||
----------
|
||||
features : int
|
||||
number of features in each combination
|
||||
max_features : int
|
||||
number of features in dataset
|
||||
|
||||
Returns
|
||||
-------
|
||||
list
|
||||
list with up to 5 combination of features randomly selected
|
||||
"""
|
||||
comb = set()
|
||||
# Generate at most 5 combinations
|
||||
if max_features == features:
|
||||
set_length = 1
|
||||
else:
|
||||
number = factorial(features) / (
|
||||
factorial(max_features) * factorial(features - max_features)
|
||||
)
|
||||
set_length = min(5, number)
|
||||
while len(comb) < set_length:
|
||||
comb.add(
|
||||
tuple(sorted(random.sample(range(features), max_features)))
|
||||
)
|
||||
return list(comb)
|
||||
|
||||
def _get_subspaces_set(
|
||||
self, dataset: np.array, labels: np.array, max_features: int
|
||||
) -> np.array:
|
||||
features = range(dataset.shape[1])
|
||||
features_sets = list(combinations(features, max_features))
|
||||
"""Compute the indices of the features selected by splitter depending
|
||||
on the self._splitter_type hyper parameter
|
||||
|
||||
Parameters
|
||||
----------
|
||||
dataset : np.array
|
||||
array of samples
|
||||
labels : np.array
|
||||
labels of the dataset
|
||||
max_features : int
|
||||
number of features of the subspace
|
||||
(<= number of features in dataset)
|
||||
|
||||
Returns
|
||||
-------
|
||||
np.array
|
||||
indices of the features selected
|
||||
"""
|
||||
features_sets = self._generate_spaces(dataset.shape[1], max_features)
|
||||
if len(features_sets) > 1:
|
||||
if self._splitter_type == "random":
|
||||
index = random.randint(0, len(features_sets) - 1)
|
||||
return features_sets[index]
|
||||
else:
|
||||
# get only 3 sets at most
|
||||
if len(features_sets) > 3:
|
||||
features_sets = random.sample(features_sets, 3)
|
||||
return self._select_best_set(dataset, labels, features_sets)
|
||||
else:
|
||||
return features_sets[0]
|
||||
|
||||
def get_subspace(
|
||||
self, dataset: np.array, labels: np.array, max_features: int
|
||||
) -> list:
|
||||
"""Return the best subspace to make a split
|
||||
) -> tuple:
|
||||
"""Return a subspace of the selected dataset of max_features length.
|
||||
Depending on hyperparmeter
|
||||
|
||||
Parameters
|
||||
----------
|
||||
dataset : np.array
|
||||
array of samples (# samples, # features)
|
||||
labels : np.array
|
||||
labels of the dataset
|
||||
max_features : int
|
||||
number of features to form the subspace
|
||||
|
||||
Returns
|
||||
-------
|
||||
tuple
|
||||
tuple with the dataset with only the features selected and the
|
||||
indices of the features selected
|
||||
"""
|
||||
indices = self._get_subspaces_set(dataset, labels, max_features)
|
||||
return dataset[:, indices], indices
|
||||
|
||||
@staticmethod
|
||||
def _min_distance(data: np.array, _) -> np.array:
|
||||
"""Assign class to min distances
|
||||
def _impurity(self, data: np.array, y: np.array) -> np.array:
|
||||
"""return column of dataset to be taken into account to split dataset
|
||||
|
||||
return a vector of classes so partition can separate class 0 from
|
||||
the rest of classes, ie. class 0 goes to one splitted node and the
|
||||
rest of classes go to the other
|
||||
:param data: distances to hyper plane of every class
|
||||
:type data: np.array (m, n_classes)
|
||||
:param _: enable call compat with other measures
|
||||
:type _: None
|
||||
:return: vector with the class assigned to each sample
|
||||
:rtype: np.array shape (m,)
|
||||
Parameters
|
||||
----------
|
||||
data : np.array
|
||||
distances to hyper plane of every class
|
||||
y : np.array
|
||||
vector of labels (classes)
|
||||
|
||||
Returns
|
||||
-------
|
||||
np.array
|
||||
column of dataset to be taken into account to split dataset
|
||||
"""
|
||||
return np.argmin(data, axis=1)
|
||||
|
||||
@staticmethod
|
||||
def _max_distance(data: np.array, _) -> np.array:
|
||||
"""Assign class to max distances
|
||||
|
||||
return a vector of classes so partition can separate class 0 from
|
||||
the rest of classes, ie. class 0 goes to one splitted node and the
|
||||
rest of classes go to the other
|
||||
:param data: distances to hyper plane of every class
|
||||
:type data: np.array (m, n_classes)
|
||||
:param _: enable call compat with other measures
|
||||
:type _: None
|
||||
:return: vector with the class assigned to each sample values
|
||||
(can be 0, 1, ...)
|
||||
:rtype: np.array shape (m,)
|
||||
"""
|
||||
return np.argmax(data, axis=1)
|
||||
max_gain = 0
|
||||
selected = -1
|
||||
for col in range(data.shape[1]):
|
||||
tup = y[data[:, col] > 0]
|
||||
tdn = y[data[:, col] <= 0]
|
||||
info_gain = self.information_gain(y, tup, tdn)
|
||||
if info_gain > max_gain:
|
||||
selected = col
|
||||
max_gain = info_gain
|
||||
return selected
|
||||
|
||||
@staticmethod
|
||||
def _max_samples(data: np.array, y: np.array) -> np.array:
|
||||
"""return distances of the class with more samples
|
||||
"""return column of dataset to be taken into account to split dataset
|
||||
|
||||
:param data: distances to hyper plane of every class
|
||||
:type data: np.array (m, n_classes)
|
||||
:param y: vector of labels (classes)
|
||||
:type y: np.array (m,)
|
||||
:return: vector with distances to hyperplane (can be positive or neg.)
|
||||
:rtype: np.array shape (m,)
|
||||
Parameters
|
||||
----------
|
||||
data : np.array
|
||||
distances to hyper plane of every class
|
||||
y : np.array
|
||||
column of dataset to be taken into account to split dataset
|
||||
|
||||
Returns
|
||||
-------
|
||||
np.array
|
||||
column of dataset to be taken into account to split dataset
|
||||
"""
|
||||
# select the class with max number of samples
|
||||
_, samples = np.unique(y, return_counts=True)
|
||||
selected = np.argmax(samples)
|
||||
return data[:, selected]
|
||||
return np.argmax(samples)
|
||||
|
||||
def partition(self, samples: np.array, node: Snode):
|
||||
def partition(self, samples: np.array, node: Snode, train: bool):
|
||||
"""Set the criteria to split arrays. Compute the indices of the samples
|
||||
that should go to one side of the tree (down)
|
||||
|
||||
that should go to one side of the tree (up)
|
||||
"""
|
||||
# data contains the distances of every sample to every class hyperplane
|
||||
# array of (m, nc) nc = # classes
|
||||
data = self._distances(node, samples)
|
||||
if data.shape[0] < self._min_samples_split:
|
||||
self._down = np.ones((data.shape[0]), dtype=bool)
|
||||
# there aren't enough samples to split
|
||||
self._up = np.ones((data.shape[0]), dtype=bool)
|
||||
return
|
||||
if data.ndim > 1:
|
||||
# split criteria for multiclass
|
||||
data = self.decision_criteria(data, node._y)
|
||||
self._down = data > 0
|
||||
# Convert data to a (m, 1) array selecting values for samples
|
||||
if train:
|
||||
# in train time we have to compute the column to take into
|
||||
# account to split the dataset
|
||||
col = self.decision_criteria(data, node._y)
|
||||
node.set_partition_column(col)
|
||||
else:
|
||||
# in predcit time just use the column computed in train time
|
||||
# is taking the classifier of class <col>
|
||||
col = node.get_partition_column()
|
||||
if col == -1:
|
||||
# No partition is producing information gain
|
||||
data = np.ones(data.shape)
|
||||
data = data[:, col]
|
||||
self._up = data > 0
|
||||
|
||||
def part(self, origin: np.array) -> list:
|
||||
"""Split an array in two based on indices (self._up) and its complement
|
||||
partition has to be called first to establish up indices
|
||||
|
||||
Parameters
|
||||
----------
|
||||
origin : np.array
|
||||
dataset to split
|
||||
|
||||
Returns
|
||||
-------
|
||||
list
|
||||
list with two splits of the array
|
||||
"""
|
||||
down = ~self._up
|
||||
return [
|
||||
origin[self._up] if any(self._up) else None,
|
||||
origin[down] if any(down) else None,
|
||||
]
|
||||
|
||||
@staticmethod
|
||||
def _distances(node: Snode, data: np.ndarray) -> np.array:
|
||||
"""Compute distances of the samples to the hyperplane of the node
|
||||
|
||||
:param node: node containing the svm classifier
|
||||
:type node: Snode
|
||||
:param data: samples to find out distance to hyperplane
|
||||
:type data: np.ndarray
|
||||
:return: array of shape (m, 1) with the distances of every sample to
|
||||
the hyperplane of the node
|
||||
:rtype: np.array
|
||||
Parameters
|
||||
----------
|
||||
node : Snode
|
||||
node containing the svm classifier
|
||||
data : np.ndarray
|
||||
samples to compute distance to hyperplane
|
||||
|
||||
Returns
|
||||
-------
|
||||
np.array
|
||||
array of shape (m, nc) with the distances of every sample to
|
||||
the hyperplane of every class. nc = # of classes
|
||||
"""
|
||||
return node._clf.decision_function(data[:, node._features])
|
||||
|
||||
def part(self, origin: np.array) -> list:
|
||||
"""Split an array in two based on indices (down) and its complement
|
||||
|
||||
:param origin: dataset to split
|
||||
:type origin: np.array
|
||||
:param down: indices to use to split array
|
||||
:type down: np.array
|
||||
:return: list with two splits of the array
|
||||
:rtype: list
|
||||
"""
|
||||
up = ~self._down
|
||||
return [
|
||||
origin[up] if any(up) else None,
|
||||
origin[self._down] if any(self._down) else None,
|
||||
]
|
||||
|
||||
|
||||
class Stree(BaseEstimator, ClassifierMixin):
|
||||
"""Estimator that is based on binary trees of svm nodes
|
||||
@@ -377,14 +494,14 @@ class Stree(BaseEstimator, ClassifierMixin):
|
||||
self,
|
||||
C: float = 1.0,
|
||||
kernel: str = "linear",
|
||||
max_iter: int = 1000,
|
||||
max_iter: int = 1e5,
|
||||
random_state: int = None,
|
||||
max_depth: int = None,
|
||||
tol: float = 1e-4,
|
||||
degree: int = 3,
|
||||
gamma="scale",
|
||||
split_criteria: str = "max_samples",
|
||||
criterion: str = "gini",
|
||||
split_criteria: str = "impurity",
|
||||
criterion: str = "entropy",
|
||||
min_samples_split: int = 0,
|
||||
max_features=None,
|
||||
splitter: str = "random",
|
||||
@@ -405,6 +522,7 @@ class Stree(BaseEstimator, ClassifierMixin):
|
||||
|
||||
def _more_tags(self) -> dict:
|
||||
"""Required by sklearn to supply features of the classifier
|
||||
make mandatory the labels array
|
||||
|
||||
:return: the tag required
|
||||
:rtype: dict
|
||||
@@ -416,16 +534,19 @@ class Stree(BaseEstimator, ClassifierMixin):
|
||||
) -> "Stree":
|
||||
"""Build the tree based on the dataset of samples and its labels
|
||||
|
||||
:param X: dataset of samples to make predictions
|
||||
:type X: np.array
|
||||
:param y: samples labels
|
||||
:type y: np.array
|
||||
:param sample_weight: weights of the samples. Rescale C per sample.
|
||||
Hi' weights force the classifier to put more emphasis on these points
|
||||
:type sample_weight: np.array optional
|
||||
:raises ValueError: if parameters C or max_depth are out of bounds
|
||||
:return: itself to be able to chain actions: fit().predict() ...
|
||||
:rtype: Stree
|
||||
Returns
|
||||
-------
|
||||
Stree
|
||||
itself to be able to chain actions: fit().predict() ...
|
||||
|
||||
Raises
|
||||
------
|
||||
ValueError
|
||||
if C < 0
|
||||
ValueError
|
||||
if max_depth < 1
|
||||
ValueError
|
||||
if all samples have 0 or negative weights
|
||||
"""
|
||||
# Check parameters are Ok.
|
||||
if self.C < 0:
|
||||
@@ -448,6 +569,10 @@ class Stree(BaseEstimator, ClassifierMixin):
|
||||
sample_weight = _check_sample_weight(
|
||||
sample_weight, X, dtype=np.float64
|
||||
)
|
||||
if not any(sample_weight):
|
||||
raise ValueError(
|
||||
"Invalid input - all samples have zero or negative weights."
|
||||
)
|
||||
check_classification_targets(y)
|
||||
# Initialize computed parameters
|
||||
self.splitter_ = Splitter(
|
||||
@@ -469,6 +594,8 @@ class Stree(BaseEstimator, ClassifierMixin):
|
||||
self.max_features_ = self._initialize_max_features()
|
||||
self.tree_ = self.train(X, y, sample_weight, 1, "root")
|
||||
self._build_predictor()
|
||||
self.X_ = X
|
||||
self.y_ = y
|
||||
return self
|
||||
|
||||
def train(
|
||||
@@ -478,23 +605,27 @@ class Stree(BaseEstimator, ClassifierMixin):
|
||||
sample_weight: np.ndarray,
|
||||
depth: int,
|
||||
title: str,
|
||||
) -> Snode:
|
||||
) -> Optional[Snode]:
|
||||
"""Recursive function to split the original dataset into predictor
|
||||
nodes (leaves)
|
||||
|
||||
:param X: samples dataset
|
||||
:type X: np.ndarray
|
||||
:param y: samples labels
|
||||
:type y: np.ndarray
|
||||
:param sample_weight: weight of samples. Rescale C per sample.
|
||||
Hi weights force the classifier to put more emphasis on these points.
|
||||
:type sample_weight: np.ndarray
|
||||
:param depth: actual depth in the tree
|
||||
:type depth: int
|
||||
:param title: description of the node
|
||||
:type title: str
|
||||
:return: binary tree
|
||||
:rtype: Snode
|
||||
Parameters
|
||||
----------
|
||||
X : np.ndarray
|
||||
samples dataset
|
||||
y : np.ndarray
|
||||
samples labels
|
||||
sample_weight : np.ndarray
|
||||
weight of samples. Rescale C per sample.
|
||||
depth : int
|
||||
actual depth in the tree
|
||||
title : str
|
||||
description of the node
|
||||
|
||||
Returns
|
||||
-------
|
||||
Optional[Snode]
|
||||
binary tree
|
||||
"""
|
||||
if depth > self.__max_depth:
|
||||
return None
|
||||
@@ -521,10 +652,10 @@ class Stree(BaseEstimator, ClassifierMixin):
|
||||
if np.unique(y_next).shape[0] != self.n_classes_:
|
||||
sample_weight += 1e-5
|
||||
clf.fit(Xs, y, sample_weight=sample_weight)
|
||||
impurity = self.splitter_.impurity(y)
|
||||
impurity = self.splitter_.partition_impurity(y)
|
||||
node = Snode(clf, X, y, features, impurity, title, sample_weight)
|
||||
self.depth_ = max(depth, self.depth_)
|
||||
self.splitter_.partition(X, node)
|
||||
self.splitter_.partition(X, node, True)
|
||||
X_U, X_D = self.splitter_.part(X)
|
||||
y_u, y_d = self.splitter_.part(y)
|
||||
sw_u, sw_d = self.splitter_.part(sample_weight)
|
||||
@@ -544,8 +675,7 @@ class Stree(BaseEstimator, ClassifierMixin):
|
||||
return node
|
||||
|
||||
def _build_predictor(self):
|
||||
"""Process the leaves to make them predictors
|
||||
"""
|
||||
"""Process the leaves to make them predictors"""
|
||||
|
||||
def run_tree(node: Snode):
|
||||
if node.is_leaf():
|
||||
@@ -557,8 +687,7 @@ class Stree(BaseEstimator, ClassifierMixin):
|
||||
run_tree(self.tree_)
|
||||
|
||||
def _build_clf(self):
|
||||
""" Build the correct classifier for the node
|
||||
"""
|
||||
"""Build the correct classifier for the node"""
|
||||
return (
|
||||
LinearSVC(
|
||||
max_iter=self.max_iter,
|
||||
@@ -581,12 +710,17 @@ class Stree(BaseEstimator, ClassifierMixin):
|
||||
def _reorder_results(y: np.array, indices: np.array) -> np.array:
|
||||
"""Reorder an array based on the array of indices passed
|
||||
|
||||
:param y: data untidy
|
||||
:type y: np.array
|
||||
:param indices: indices used to set order
|
||||
:type indices: np.array
|
||||
:return: array y ordered
|
||||
:rtype: np.array
|
||||
Parameters
|
||||
----------
|
||||
y : np.array
|
||||
data untidy
|
||||
indices : np.array
|
||||
indices used to set order
|
||||
|
||||
Returns
|
||||
-------
|
||||
np.array
|
||||
array y ordered
|
||||
"""
|
||||
# return array of same type given in y
|
||||
y_ordered = y.copy()
|
||||
@@ -598,10 +732,22 @@ class Stree(BaseEstimator, ClassifierMixin):
|
||||
def predict(self, X: np.array) -> np.array:
|
||||
"""Predict labels for each sample in dataset passed
|
||||
|
||||
:param X: dataset of samples
|
||||
:type X: np.array
|
||||
:return: array of labels
|
||||
:rtype: np.array
|
||||
Parameters
|
||||
----------
|
||||
X : np.array
|
||||
dataset of samples
|
||||
|
||||
Returns
|
||||
-------
|
||||
np.array
|
||||
array of labels
|
||||
|
||||
Raises
|
||||
------
|
||||
ValueError
|
||||
if dataset with inconsistent number of features
|
||||
NotFittedError
|
||||
if model is not fitted
|
||||
"""
|
||||
|
||||
def predict_class(
|
||||
@@ -613,7 +759,7 @@ class Stree(BaseEstimator, ClassifierMixin):
|
||||
# set a class for every sample in dataset
|
||||
prediction = np.full((xp.shape[0], 1), node._class)
|
||||
return prediction, indices
|
||||
self.splitter_.partition(xp, node)
|
||||
self.splitter_.partition(xp, node, train=False)
|
||||
x_u, x_d = self.splitter_.part(xp)
|
||||
i_u, i_d = self.splitter_.part(indices)
|
||||
prx_u, prin_u = predict_class(x_u, i_u, node.get_up())
|
||||
@@ -643,15 +789,19 @@ class Stree(BaseEstimator, ClassifierMixin):
|
||||
) -> float:
|
||||
"""Compute accuracy of the prediction
|
||||
|
||||
:param X: dataset of samples to make predictions
|
||||
:type X: np.array
|
||||
:param y_true: samples labels
|
||||
:type y_true: np.array
|
||||
:param sample_weight: weights of the samples. Rescale C per sample.
|
||||
Hi' weights force the classifier to put more emphasis on these points
|
||||
:type sample_weight: np.array optional
|
||||
:return: accuracy of the prediction
|
||||
:rtype: float
|
||||
Parameters
|
||||
----------
|
||||
X : np.array
|
||||
dataset of samples to make predictions
|
||||
y : np.array
|
||||
samples labels
|
||||
sample_weight : np.array, optional
|
||||
weights of the samples. Rescale C per sample, by default None
|
||||
|
||||
Returns
|
||||
-------
|
||||
float
|
||||
accuracy of the prediction
|
||||
"""
|
||||
# sklearn check
|
||||
check_is_fitted(self)
|
||||
@@ -668,8 +818,10 @@ class Stree(BaseEstimator, ClassifierMixin):
|
||||
"""Create an iterator to be able to visit the nodes of the tree in
|
||||
preorder, can make a list with all the nodes in preorder
|
||||
|
||||
:return: an iterator, can for i in... and list(...)
|
||||
:rtype: Siterator
|
||||
Returns
|
||||
-------
|
||||
Siterator
|
||||
an iterator, can for i in... and list(...)
|
||||
"""
|
||||
try:
|
||||
tree = self.tree_
|
||||
@@ -680,8 +832,10 @@ class Stree(BaseEstimator, ClassifierMixin):
|
||||
def __str__(self) -> str:
|
||||
"""String representation of the tree
|
||||
|
||||
:return: description of nodes in the tree in preorder
|
||||
:rtype: str
|
||||
Returns
|
||||
-------
|
||||
str
|
||||
description of nodes in the tree in preorder
|
||||
"""
|
||||
output = ""
|
||||
for i in self:
|
||||
|
||||
@@ -40,12 +40,13 @@ class Snode_test(unittest.TestCase):
|
||||
# Check Class
|
||||
class_computed = classes[card == max_card]
|
||||
self.assertEqual(class_computed, node._class)
|
||||
# Check Partition column
|
||||
self.assertEqual(node._partition_column, -1)
|
||||
|
||||
check_leave(self._clf.tree_)
|
||||
|
||||
def test_nodes_coefs(self):
|
||||
"""Check if the nodes of the tree have the right attributes filled
|
||||
"""
|
||||
"""Check if the nodes of the tree have the right attributes filled"""
|
||||
|
||||
def run_tree(node: Snode):
|
||||
if node._belief < 1:
|
||||
@@ -54,16 +55,19 @@ class Snode_test(unittest.TestCase):
|
||||
self.assertIsNotNone(node._clf.coef_)
|
||||
if node.is_leaf():
|
||||
return
|
||||
run_tree(node.get_down())
|
||||
run_tree(node.get_up())
|
||||
run_tree(node.get_down())
|
||||
|
||||
run_tree(self._clf.tree_)
|
||||
model = Stree(self._random_state)
|
||||
model.fit(*load_dataset(self._random_state, 3, 4))
|
||||
run_tree(model.tree_)
|
||||
|
||||
def test_make_predictor_on_leaf(self):
|
||||
test = Snode(None, [1, 2, 3, 4], [1, 0, 1, 1], [], 0.0, "test")
|
||||
test.make_predictor()
|
||||
self.assertEqual(1, test._class)
|
||||
self.assertEqual(0.75, test._belief)
|
||||
self.assertEqual(-1, test._partition_column)
|
||||
|
||||
def test_make_predictor_on_not_leaf(self):
|
||||
test = Snode(None, [1, 2, 3, 4], [1, 0, 1, 1], [], 0.0, "test")
|
||||
@@ -71,11 +75,14 @@ class Snode_test(unittest.TestCase):
|
||||
test.make_predictor()
|
||||
self.assertIsNone(test._class)
|
||||
self.assertEqual(0, test._belief)
|
||||
self.assertEqual(-1, test._partition_column)
|
||||
self.assertEqual(-1, test.get_up()._partition_column)
|
||||
|
||||
def test_make_predictor_on_leaf_bogus_data(self):
|
||||
test = Snode(None, [1, 2, 3, 4], [], [], 0.0, "test")
|
||||
test.make_predictor()
|
||||
self.assertIsNone(test._class)
|
||||
self.assertEqual(-1, test._partition_column)
|
||||
|
||||
def test_copy_node(self):
|
||||
px = [1, 2, 3, 4]
|
||||
@@ -86,3 +93,4 @@ class Snode_test(unittest.TestCase):
|
||||
self.assertListEqual(computed._y, py)
|
||||
self.assertEqual("test", computed._title)
|
||||
self.assertIsInstance(computed._clf, Stree)
|
||||
self.assertEqual(test._partition_column, computed._partition_column)
|
||||
|
||||
@@ -19,7 +19,7 @@ class Splitter_test(unittest.TestCase):
|
||||
min_samples_split=0,
|
||||
splitter_type="random",
|
||||
criterion="gini",
|
||||
criteria="min_distance",
|
||||
criteria="max_samples",
|
||||
random_state=None,
|
||||
):
|
||||
return Splitter(
|
||||
@@ -46,11 +46,7 @@ class Splitter_test(unittest.TestCase):
|
||||
_ = Splitter(clf=None)
|
||||
for splitter_type in ["best", "random"]:
|
||||
for criterion in ["gini", "entropy"]:
|
||||
for criteria in [
|
||||
"min_distance",
|
||||
"max_samples",
|
||||
"max_distance",
|
||||
]:
|
||||
for criteria in ["max_samples", "impurity"]:
|
||||
tcl = self.build(
|
||||
splitter_type=splitter_type,
|
||||
criterion=criterion,
|
||||
@@ -138,43 +134,45 @@ class Splitter_test(unittest.TestCase):
|
||||
[0.7, 0.01, -0.1],
|
||||
[0.7, -0.9, 0.5],
|
||||
[0.1, 0.2, 0.3],
|
||||
[-0.1, 0.2, 0.3],
|
||||
[-0.1, 0.2, 0.3],
|
||||
]
|
||||
)
|
||||
expected = np.array([0.2, 0.01, -0.9, 0.2])
|
||||
y = [1, 2, 1, 0]
|
||||
expected = data[:, 0]
|
||||
y = [1, 2, 1, 0, 0, 0]
|
||||
computed = tcl._max_samples(data, y)
|
||||
self.assertEqual((4,), computed.shape)
|
||||
self.assertListEqual(expected.tolist(), computed.tolist())
|
||||
self.assertEqual(0, computed)
|
||||
computed_data = data[:, computed]
|
||||
self.assertEqual((6,), computed_data.shape)
|
||||
self.assertListEqual(expected.tolist(), computed_data.tolist())
|
||||
|
||||
def test_min_distance(self):
|
||||
tcl = self.build()
|
||||
def test_impurity(self):
|
||||
tcl = self.build(criteria="impurity")
|
||||
data = np.array(
|
||||
[
|
||||
[-0.1, 0.2, -0.3],
|
||||
[0.7, 0.01, -0.1],
|
||||
[0.7, -0.9, 0.5],
|
||||
[0.1, 0.2, 0.3],
|
||||
[-0.1, 0.2, 0.3],
|
||||
[-0.1, 0.2, 0.3],
|
||||
]
|
||||
)
|
||||
expected = np.array([2, 2, 1, 0])
|
||||
computed = tcl._min_distance(data, None)
|
||||
self.assertEqual((4,), computed.shape)
|
||||
self.assertListEqual(expected.tolist(), computed.tolist())
|
||||
expected = data[:, 2]
|
||||
y = np.array([1, 2, 1, 0, 0, 0])
|
||||
computed = tcl._impurity(data, y)
|
||||
self.assertEqual(2, computed)
|
||||
computed_data = data[:, computed]
|
||||
self.assertEqual((6,), computed_data.shape)
|
||||
self.assertListEqual(expected.tolist(), computed_data.tolist())
|
||||
|
||||
def test_max_distance(self):
|
||||
tcl = self.build(criteria="max_distance")
|
||||
data = np.array(
|
||||
[
|
||||
[-0.1, 0.2, -0.3],
|
||||
[0.7, 0.01, -0.1],
|
||||
[0.7, -0.9, 0.5],
|
||||
[0.1, 0.2, 0.3],
|
||||
]
|
||||
)
|
||||
expected = np.array([1, 0, 0, 2])
|
||||
computed = tcl._max_distance(data, None)
|
||||
self.assertEqual((4,), computed.shape)
|
||||
self.assertListEqual(expected.tolist(), computed.tolist())
|
||||
def test_generate_subspaces(self):
|
||||
features = 250
|
||||
for max_features in range(2, features):
|
||||
num = len(Splitter._generate_spaces(features, max_features))
|
||||
self.assertEqual(5, num)
|
||||
self.assertEqual(3, len(Splitter._generate_spaces(3, 2)))
|
||||
self.assertEqual(4, len(Splitter._generate_spaces(4, 3)))
|
||||
|
||||
def test_best_splitter_few_sets(self):
|
||||
X, y = load_iris(return_X_y=True)
|
||||
@@ -186,27 +184,22 @@ class Splitter_test(unittest.TestCase):
|
||||
|
||||
def test_splitter_parameter(self):
|
||||
expected_values = [
|
||||
[2, 3, 5, 7], # best entropy min_distance
|
||||
[0, 2, 4, 5], # best entropy max_samples
|
||||
[0, 2, 8, 12], # best entropy max_distance
|
||||
[1, 2, 5, 12], # best gini min_distance
|
||||
[0, 3, 4, 10], # best gini max_samples
|
||||
[1, 2, 9, 12], # best gini max_distance
|
||||
[3, 9, 11, 12], # random entropy min_distance
|
||||
[1, 5, 6, 9], # random entropy max_samples
|
||||
[1, 2, 4, 8], # random entropy max_distance
|
||||
[2, 6, 7, 12], # random gini min_distance
|
||||
[3, 9, 10, 11], # random gini max_samples
|
||||
[2, 5, 8, 12], # random gini max_distance
|
||||
[1, 4, 9, 12], # best entropy max_samples
|
||||
[1, 3, 6, 10], # best entropy impurity
|
||||
[6, 8, 10, 12], # best gini max_samples
|
||||
[7, 8, 10, 11], # best gini impurity
|
||||
[0, 3, 8, 12], # random entropy max_samples
|
||||
[0, 3, 9, 11], # random entropy impurity
|
||||
[0, 4, 7, 12], # random gini max_samples
|
||||
[0, 2, 5, 6], # random gini impurity
|
||||
]
|
||||
X, y = load_wine(return_X_y=True)
|
||||
rn = 0
|
||||
for splitter_type in ["best", "random"]:
|
||||
for criterion in ["entropy", "gini"]:
|
||||
for criteria in [
|
||||
"min_distance",
|
||||
"max_samples",
|
||||
"max_distance",
|
||||
"impurity",
|
||||
]:
|
||||
tcl = self.build(
|
||||
splitter_type=splitter_type,
|
||||
@@ -219,7 +212,9 @@ class Splitter_test(unittest.TestCase):
|
||||
dataset, computed = tcl.get_subspace(X, y, max_features=4)
|
||||
# print(
|
||||
# "{}, # {:7s}{:8s}{:15s}".format(
|
||||
# list(computed), splitter_type, criterion,
|
||||
# list(computed),
|
||||
# splitter_type,
|
||||
# criterion,
|
||||
# criteria,
|
||||
# )
|
||||
# )
|
||||
|
||||
@@ -5,6 +5,7 @@ import warnings
|
||||
import numpy as np
|
||||
from sklearn.datasets import load_iris, load_wine
|
||||
from sklearn.exceptions import ConvergenceWarning
|
||||
from sklearn.svm import LinearSVC
|
||||
|
||||
from stree import Stree, Snode
|
||||
from .utils import load_dataset
|
||||
@@ -25,8 +26,10 @@ class Stree_test(unittest.TestCase):
|
||||
correct number of labels and its sons have the right number of elements
|
||||
in their dataset
|
||||
|
||||
Arguments:
|
||||
node {Snode} -- node to check
|
||||
Parameters
|
||||
----------
|
||||
node : Snode
|
||||
node to check
|
||||
"""
|
||||
if node.is_leaf():
|
||||
return
|
||||
@@ -41,23 +44,22 @@ class Stree_test(unittest.TestCase):
|
||||
_, count_u = np.unique(y_up, return_counts=True)
|
||||
#
|
||||
for i in unique_y:
|
||||
number_down = count_d[i]
|
||||
try:
|
||||
number_up = count_u[i]
|
||||
try:
|
||||
number_down = count_d[i]
|
||||
except IndexError:
|
||||
number_up = 0
|
||||
number_down = 0
|
||||
self.assertEqual(count_y[i], number_down + number_up)
|
||||
# Is the partition made the same as the prediction?
|
||||
# as the node is not a leaf...
|
||||
_, count_yp = np.unique(y_prediction, return_counts=True)
|
||||
self.assertEqual(count_yp[0], y_up.shape[0])
|
||||
self.assertEqual(count_yp[1], y_down.shape[0])
|
||||
self.assertEqual(count_yp[1], y_up.shape[0])
|
||||
self.assertEqual(count_yp[0], y_down.shape[0])
|
||||
self._check_tree(node.get_down())
|
||||
self._check_tree(node.get_up())
|
||||
|
||||
def test_build_tree(self):
|
||||
"""Check if the tree is built the same way as predictions of models
|
||||
"""
|
||||
"""Check if the tree is built the same way as predictions of models"""
|
||||
warnings.filterwarnings("ignore")
|
||||
for kernel in self._kernels:
|
||||
clf = Stree(kernel=kernel, random_state=self._random_state)
|
||||
@@ -99,20 +101,22 @@ class Stree_test(unittest.TestCase):
|
||||
self.assertListEqual(yp_line.tolist(), yp_once.tolist())
|
||||
|
||||
def test_iterator_and_str(self):
|
||||
"""Check preorder iterator
|
||||
"""
|
||||
"""Check preorder iterator"""
|
||||
expected = [
|
||||
"root feaures=(0, 1, 2) impurity=0.5000",
|
||||
"root - Down feaures=(0, 1, 2) impurity=0.0671",
|
||||
"root - Down - Down, <cgaf> - Leaf class=1 belief= 0.975989 "
|
||||
"impurity=0.0469 counts=(array([0, 1]), array([ 17, 691]))",
|
||||
"root - Down - Up feaures=(0, 1, 2) impurity=0.3967",
|
||||
"root - Down - Up - Down, <cgaf> - Leaf class=1 belief= 0.750000 "
|
||||
"impurity=0.3750 counts=(array([0, 1]), array([1, 3]))",
|
||||
"root - Down - Up - Up, <pure> - Leaf class=0 belief= 1.000000 "
|
||||
"impurity=0.0000 counts=(array([0]), array([7]))",
|
||||
"root - Up, <cgaf> - Leaf class=0 belief= 0.928297 impurity=0.1331"
|
||||
" counts=(array([0, 1]), array([725, 56]))",
|
||||
"root feaures=(0, 1, 2) impurity=1.0000 counts=(array([0, 1]), arr"
|
||||
"ay([750, 750]))",
|
||||
"root - Down, <cgaf> - Leaf class=0 belief= 0.928297 impurity=0.37"
|
||||
"22 counts=(array([0, 1]), array([725, 56]))",
|
||||
"root - Up feaures=(0, 1, 2) impurity=0.2178 counts=(array([0, 1])"
|
||||
", array([ 25, 694]))",
|
||||
"root - Up - Down feaures=(0, 1, 2) impurity=0.8454 counts=(array("
|
||||
"[0, 1]), array([8, 3]))",
|
||||
"root - Up - Down - Down, <pure> - Leaf class=0 belief= 1.000000 i"
|
||||
"mpurity=0.0000 counts=(array([0]), array([7]))",
|
||||
"root - Up - Down - Up, <cgaf> - Leaf class=1 belief= 0.750000 imp"
|
||||
"urity=0.8113 counts=(array([0, 1]), array([1, 3]))",
|
||||
"root - Up - Up, <cgaf> - Leaf class=1 belief= 0.975989 impurity=0"
|
||||
".1634 counts=(array([0, 1]), array([ 17, 691]))",
|
||||
]
|
||||
computed = []
|
||||
expected_string = ""
|
||||
@@ -188,44 +192,43 @@ class Stree_test(unittest.TestCase):
|
||||
def test_muticlass_dataset(self):
|
||||
datasets = {
|
||||
"Synt": load_dataset(random_state=self._random_state, n_classes=3),
|
||||
"Iris": load_iris(return_X_y=True),
|
||||
"Iris": load_wine(return_X_y=True),
|
||||
}
|
||||
outcomes = {
|
||||
"Synt": {
|
||||
"max_samples linear": 0.9533333333333334,
|
||||
"max_samples rbf": 0.836,
|
||||
"max_samples poly": 0.9473333333333334,
|
||||
"min_distance linear": 0.9533333333333334,
|
||||
"min_distance rbf": 0.836,
|
||||
"min_distance poly": 0.9473333333333334,
|
||||
"max_distance linear": 0.9533333333333334,
|
||||
"max_distance rbf": 0.836,
|
||||
"max_distance poly": 0.9473333333333334,
|
||||
"max_samples linear": 0.9606666666666667,
|
||||
"max_samples rbf": 0.7133333333333334,
|
||||
"max_samples poly": 0.49066666666666664,
|
||||
"impurity linear": 0.9606666666666667,
|
||||
"impurity rbf": 0.7133333333333334,
|
||||
"impurity poly": 0.49066666666666664,
|
||||
},
|
||||
"Iris": {
|
||||
"max_samples linear": 0.98,
|
||||
"max_samples rbf": 1.0,
|
||||
"max_samples poly": 1.0,
|
||||
"min_distance linear": 0.98,
|
||||
"min_distance rbf": 1.0,
|
||||
"min_distance poly": 1.0,
|
||||
"max_distance linear": 0.98,
|
||||
"max_distance rbf": 1.0,
|
||||
"max_distance poly": 1.0,
|
||||
"max_samples linear": 1.0,
|
||||
"max_samples rbf": 0.6910112359550562,
|
||||
"max_samples poly": 0.6966292134831461,
|
||||
"impurity linear": 1,
|
||||
"impurity rbf": 0.6910112359550562,
|
||||
"impurity poly": 0.6966292134831461,
|
||||
},
|
||||
}
|
||||
|
||||
for name, dataset in datasets.items():
|
||||
px, py = dataset
|
||||
for criteria in ["max_samples", "min_distance", "max_distance"]:
|
||||
for criteria in ["max_samples", "impurity"]:
|
||||
for kernel in self._kernels:
|
||||
clf = Stree(
|
||||
C=1e4,
|
||||
max_iter=1e4,
|
||||
C=55,
|
||||
max_iter=1e5,
|
||||
kernel=kernel,
|
||||
random_state=self._random_state,
|
||||
)
|
||||
clf.fit(px, py)
|
||||
outcome = outcomes[name][f"{criteria} {kernel}"]
|
||||
# print(
|
||||
# f"{name} {criteria} {kernel} {outcome} {clf.score(px"
|
||||
# ", py)}"
|
||||
# )
|
||||
self.assertAlmostEqual(outcome, clf.score(px, py))
|
||||
|
||||
def test_max_features(self):
|
||||
@@ -297,7 +300,10 @@ class Stree_test(unittest.TestCase):
|
||||
0.9433333333333334,
|
||||
]
|
||||
for kernel, accuracy_expected in zip(self._kernels, accuracies):
|
||||
clf = Stree(random_state=self._random_state, kernel=kernel,)
|
||||
clf = Stree(
|
||||
random_state=self._random_state,
|
||||
kernel=kernel,
|
||||
)
|
||||
clf.fit(X, y)
|
||||
accuracy_score = clf.score(X, y)
|
||||
yp = clf.predict(X)
|
||||
@@ -309,81 +315,104 @@ class Stree_test(unittest.TestCase):
|
||||
X, y = load_dataset(self._random_state)
|
||||
clf = Stree(random_state=self._random_state, max_features=2)
|
||||
clf.fit(X, y)
|
||||
self.assertAlmostEqual(0.9426666666666667, clf.score(X, y))
|
||||
|
||||
def test_score_multi_class(self):
|
||||
warnings.filterwarnings("ignore")
|
||||
accuracies = [
|
||||
0.8258427, # Wine linear min_distance
|
||||
0.6741573, # Wine linear max_distance
|
||||
0.8314607, # Wine linear max_samples
|
||||
0.6629213, # Wine rbf min_distance
|
||||
1.0000000, # Wine rbf max_distance
|
||||
0.4044944, # Wine rbf max_samples
|
||||
0.9157303, # Wine poly min_distance
|
||||
1.0000000, # Wine poly max_distance
|
||||
0.7640449, # Wine poly max_samples
|
||||
0.9933333, # Iris linear min_distance
|
||||
0.9666667, # Iris linear max_distance
|
||||
0.9666667, # Iris linear max_samples
|
||||
0.9800000, # Iris rbf min_distance
|
||||
0.9800000, # Iris rbf max_distance
|
||||
0.9800000, # Iris rbf max_samples
|
||||
1.0000000, # Iris poly min_distance
|
||||
1.0000000, # Iris poly max_distance
|
||||
1.0000000, # Iris poly max_samples
|
||||
0.8993333, # Synthetic linear min_distance
|
||||
0.6533333, # Synthetic linear max_distance
|
||||
0.9313333, # Synthetic linear max_samples
|
||||
0.8320000, # Synthetic rbf min_distance
|
||||
0.6660000, # Synthetic rbf max_distance
|
||||
0.8320000, # Synthetic rbf max_samples
|
||||
0.6066667, # Synthetic poly min_distance
|
||||
0.6840000, # Synthetic poly max_distance
|
||||
0.6340000, # Synthetic poly max_samples
|
||||
]
|
||||
datasets = [
|
||||
("Wine", load_wine(return_X_y=True)),
|
||||
("Iris", load_iris(return_X_y=True)),
|
||||
(
|
||||
"Synthetic",
|
||||
load_dataset(self._random_state, n_classes=3, n_features=5),
|
||||
),
|
||||
]
|
||||
for dataset_name, dataset in datasets:
|
||||
X, y = dataset
|
||||
for kernel in self._kernels:
|
||||
for criteria in [
|
||||
"min_distance",
|
||||
"max_distance",
|
||||
"max_samples",
|
||||
]:
|
||||
clf = Stree(
|
||||
C=17,
|
||||
random_state=self._random_state,
|
||||
kernel=kernel,
|
||||
split_criteria=criteria,
|
||||
degree=5,
|
||||
gamma="auto",
|
||||
)
|
||||
clf.fit(X, y)
|
||||
accuracy_score = clf.score(X, y)
|
||||
yp = clf.predict(X)
|
||||
accuracy_computed = np.mean(yp == y)
|
||||
# print(
|
||||
# "{:.7f}, # {:7} {:5} {}".format(
|
||||
# accuracy_score, dataset_name, kernel, criteria
|
||||
# )
|
||||
# )
|
||||
accuracy_expected = accuracies.pop(0)
|
||||
self.assertEqual(accuracy_score, accuracy_computed)
|
||||
self.assertAlmostEqual(accuracy_expected, accuracy_score)
|
||||
self.assertAlmostEqual(0.9246666666666666, clf.score(X, y))
|
||||
|
||||
def test_bogus_splitter_parameter(self):
|
||||
clf = Stree(splitter="duck")
|
||||
with self.assertRaises(ValueError):
|
||||
clf.fit(*load_dataset())
|
||||
|
||||
def test_multiclass_classifier_integrity(self):
|
||||
"""Checks if the multiclass operation is done right"""
|
||||
X, y = load_iris(return_X_y=True)
|
||||
clf = Stree(random_state=0)
|
||||
clf.fit(X, y)
|
||||
score = clf.score(X, y)
|
||||
# Check accuracy of the whole model
|
||||
self.assertAlmostEquals(0.98, score, 5)
|
||||
svm = LinearSVC(random_state=0)
|
||||
svm.fit(X, y)
|
||||
self.assertAlmostEquals(0.9666666666666667, svm.score(X, y), 5)
|
||||
data = svm.decision_function(X)
|
||||
expected = [
|
||||
0.4444444444444444,
|
||||
0.35777777777777775,
|
||||
0.4569777777777778,
|
||||
]
|
||||
ty = data.copy()
|
||||
ty[data <= 0] = 0
|
||||
ty[data > 0] = 1
|
||||
ty = ty.astype(int)
|
||||
for i in range(3):
|
||||
self.assertAlmostEquals(
|
||||
expected[i],
|
||||
clf.splitter_._gini(ty[:, i]),
|
||||
)
|
||||
# 1st Branch
|
||||
# up has to have 50 samples of class 0
|
||||
# down should have 100 [50, 50]
|
||||
up = data[:, 2] > 0
|
||||
resup = np.unique(y[up], return_counts=True)
|
||||
resdn = np.unique(y[~up], return_counts=True)
|
||||
self.assertListEqual([1, 2], resup[0].tolist())
|
||||
self.assertListEqual([3, 50], resup[1].tolist())
|
||||
self.assertListEqual([0, 1], resdn[0].tolist())
|
||||
self.assertListEqual([50, 47], resdn[1].tolist())
|
||||
# 2nd Branch
|
||||
# up should have 53 samples of classes [1, 2] [3, 50]
|
||||
# down shoud have 47 samples of class 1
|
||||
node_up = clf.tree_.get_down().get_up()
|
||||
node_dn = clf.tree_.get_down().get_down()
|
||||
resup = np.unique(node_up._y, return_counts=True)
|
||||
resdn = np.unique(node_dn._y, return_counts=True)
|
||||
self.assertListEqual([1, 2], resup[0].tolist())
|
||||
self.assertListEqual([3, 50], resup[1].tolist())
|
||||
self.assertListEqual([1], resdn[0].tolist())
|
||||
self.assertListEqual([47], resdn[1].tolist())
|
||||
|
||||
def test_score_multiclass_rbf(self):
|
||||
X, y = load_dataset(
|
||||
random_state=self._random_state,
|
||||
n_classes=3,
|
||||
n_features=5,
|
||||
n_samples=500,
|
||||
)
|
||||
clf = Stree(kernel="rbf", random_state=self._random_state)
|
||||
self.assertEqual(0.824, clf.fit(X, y).score(X, y))
|
||||
X, y = load_wine(return_X_y=True)
|
||||
self.assertEqual(0.6741573033707865, clf.fit(X, y).score(X, y))
|
||||
|
||||
def test_score_multiclass_poly(self):
|
||||
X, y = load_dataset(
|
||||
random_state=self._random_state,
|
||||
n_classes=3,
|
||||
n_features=5,
|
||||
n_samples=500,
|
||||
)
|
||||
clf = Stree(
|
||||
kernel="poly", random_state=self._random_state, C=10, degree=5
|
||||
)
|
||||
self.assertEqual(0.786, clf.fit(X, y).score(X, y))
|
||||
X, y = load_wine(return_X_y=True)
|
||||
self.assertEqual(0.702247191011236, clf.fit(X, y).score(X, y))
|
||||
|
||||
def test_score_multiclass_linear(self):
|
||||
X, y = load_dataset(
|
||||
random_state=self._random_state,
|
||||
n_classes=3,
|
||||
n_features=5,
|
||||
n_samples=1500,
|
||||
)
|
||||
clf = Stree(kernel="linear", random_state=self._random_state)
|
||||
self.assertEqual(0.9533333333333334, clf.fit(X, y).score(X, y))
|
||||
X, y = load_wine(return_X_y=True)
|
||||
self.assertEqual(0.9550561797752809, clf.fit(X, y).score(X, y))
|
||||
|
||||
def test_zero_all_sample_weights(self):
|
||||
X, y = load_dataset(self._random_state)
|
||||
with self.assertRaises(ValueError):
|
||||
Stree().fit(X, y, np.zeros(len(y)))
|
||||
|
||||
def test_weights_removing_class(self):
|
||||
# This patch solves an stderr message from sklearn svm lib
|
||||
# "WARNING: class label x specified in weight is not found"
|
||||
@@ -407,7 +436,13 @@ class Stree_test(unittest.TestCase):
|
||||
original = weights_no_zero.copy()
|
||||
clf = Stree()
|
||||
clf.fit(X, y)
|
||||
node = clf.train(X, y, weights, 1, "test",)
|
||||
node = clf.train(
|
||||
X,
|
||||
y,
|
||||
weights,
|
||||
1,
|
||||
"test",
|
||||
)
|
||||
# if a class is lost with zero weights the patch adds epsilon
|
||||
self.assertListEqual(weights.tolist(), weights_epsilon)
|
||||
self.assertListEqual(node._sample_weight.tolist(), weights_epsilon)
|
||||
|
||||
@@ -1,9 +1,9 @@
|
||||
from sklearn.datasets import make_classification
|
||||
|
||||
|
||||
def load_dataset(random_state=0, n_classes=2, n_features=3):
|
||||
def load_dataset(random_state=0, n_classes=2, n_features=3, n_samples=1500):
|
||||
X, y = make_classification(
|
||||
n_samples=1500,
|
||||
n_samples=n_samples,
|
||||
n_features=n_features,
|
||||
n_informative=3,
|
||||
n_redundant=0,
|
||||
|
||||
Reference in New Issue
Block a user