pca + update README

Author: agrimagsrl

.gitignore

@@ -1,4 +1,5 @@
 venv
 .idea
 tests
-dist
+dist
+micromlgen/__pycache__

MANIFEST

@@ -6,7 +6,7 @@ micromlgen/micromlgen.py
 micromlgen/platforms.py
 micromlgen/utils.py
 micromlgen/templates/classmap.jinja
-micromlgen/templates/xy.jinja
+micromlgen/templates/testset.jinja
 micromlgen/templates/rvm/rvm.jinja
 micromlgen/templates/svm/svm.jinja
 micromlgen/templates/svm/computations/class.jinja

README.md

@@ -8,7 +8,10 @@ to an introduction to the topic.
 
 `pip install micromlgen`
 
-## Use
+## Support (and Relevant) Vector Machines
+
+`micromlgen` can port to plain C SVM-based (SVC, LinearSVC, OneClassSVM)
+and RVM-based (from `skbayes.rvm_ard_models` package) classifiers.
 
 ```python
 from micromlgen import port
@@ -44,18 +47,18 @@ if __name__ == '__main__':
     }))
 ```
 
-You can pass a test set to generate self test code
+## PCA
+
+It can export a PCA transformer.
 
 ```python
-from micromlgen import port
-from sklearn.svm import SVC
+from sklearn.decomposition import PCA
 from sklearn.datasets import load_iris
-
+from micromlgen import port
 
 if __name__ == '__main__':
-    iris = load_iris()
-    X_train, X_test = iris.data[:-10, :], iris.data[-10:, :]
-    y_train, y_test = iris.target[:-10], iris.target[-10:]
-    clf = SVC(kernel='linear').fit(X_train, y_train)
-    print(port(clf, test_set=(X_test, y_test)))
+    X = load_iris().data
+    pca = PCA(n_components=2, whiten=False).fit(X)
+    
+    print(port(pca))
 ```

micromlgen/__pycache__/__init__.cpython-36.pyc

@@ -1,5 +1,6 @@
 from skbayes.rvm_ard_models import RVC
 from sklearn.svm import SVC, LinearSVC, OneClassSVM
+from sklearn.decomposition import PCA
 
 from micromlgen import platforms
 from micromlgen.utils import jinja
@@ -65,6 +66,18 @@ def port_svm(clf, classname=None, **kwargs):
     return jinja('svm/svm.jinja', template_data)
 
 
+def port_pca(pca, classname=None, **kwargs):
+    """Port a PCA"""
+    template_data = {
+        'arrays': {
+            'components': pca.components_,
+            'mean': pca.mean_
+        },
+        'classname': classname if classname is not None else 'PCA'
+    }
+    return jinja('pca/pca.jinja', template_data)
+
+
 def port(
         clf,
         classname=None,
@@ -76,4 +89,6 @@ def port(
         return port_svm(**locals())
     elif isinstance(clf, RVC):
         return port_rvm(**locals())
-    raise TypeError('clf MUST be one of SVC, LinearSVC, OneClassSVC, RVC')
+    elif isinstance(clf, PCA):
+        return port_pca(pca=clf, **locals())
+    raise TypeError('clf MUST be one of SVC, LinearSVC, OneClassSVC, RVC, PCA')

micromlgen/__pycache__/micromlgen.cpython-36.pyc

@@ -0,0 +1,45 @@
+#pragma once
+
+namespace Eloquent {
+    namespace ML {
+        namespace Port {
+
+            class {{ classname }} {
+                public:
+
+                    /**
+                     * Apply dimensionality reduction
+                     * @warn Will override the source vector if no dest provided!
+                     */
+                    void transform(float *x, float *dest = NULL) {
+                        static float u[{{ arrays.components|length }}] = { 0 };
+
+                        {% for i, component in f.enumerate(arrays.components) %}
+                        u[{{ i }}] = dot(x, {% for j, cj in f.enumerate(component) %} {% if j > 0 %},{% endif %} {{ f.round(cj) }} {% endfor %});
+                        {% endfor %}
+
+                        memcpy(dest != NULL ? dest : x, u, sizeof(float) * {{ arrays.components|length }});
+                    }
+
+                protected:
+
+                    /**
+                     * Compute dot product with varargs
+                     */
+                    float dot(float *x, ...) {
+                        va_list w;
+                        va_start(w, {{ arrays.components[0]|length }});
+
+                        static float mean[] = { {% for i, m in f.enumerate(arrays.mean) %}{% if i > 0 %},{% endif %} {{ f.round(m) }} {% endfor %} };
+                        float dot = 0.0;
+
+                        for (uint16_t i = 0; i < {{ arrays.components[0]|length }}; i++) {
+                            dot += (x[i] - mean[i]) * va_arg(w, double);
+                        }
+
+                        return dot;
+                    }
+            };
+        }
+    }
+}

micromlgen/__pycache__/platforms.cpython-36.pyc

@@ -10,8 +10,8 @@ namespace Eloquent {
 
                     }
 
-                    uint8_t predict(double *x) {
-                        double decision[{{ arrays.vectors|length }}] = { 0 };
+                    uint8_t predict(float *x) {
+                        float decision[{{ arrays.vectors|length }}] = { 0 };
                         {% for i, (rv, cf, act, b) in f.enumerate(f.zip(arrays.vectors, arrays.coefs, arrays.actives, arrays.intercepts)) %}
                             {% if rv.shape[0] == 0 %}
                                 decision[{{ i }}] = {{ b }};
@@ -21,7 +21,7 @@ namespace Eloquent {
                         {% endfor %}
 
                         uint8_t idx = 0;
-                        double val = decision[0];
+                        float val = decision[0];
 
                         for (uint8_t i = 1; i < {{ arrays.vectors|length }}; i++) {
                             if (decision[i] > val) {

micromlgen/__pycache__/utils.cpython-36.pyc

@@ -7,13 +7,13 @@
 setup(
   name = 'micromlgen',
   packages = ['micromlgen'],
-  version = '1.0.2',
+  version = '1.1.0',
   license='MIT',
   description = 'Generate C code for microcontrollers from Python\'s sklearn classifiers',
   author = 'Simone Salerno',
   author_email = 'eloquentarduino@gmail.com',
   url = 'https://github.com/eloquentarduino/micromlgen',
-  download_url = 'https://github.com/eloquentarduino/micromlgen/archive/v_10ter.tar.gz',
+  download_url = 'https://github.com/eloquentarduino/micromlgen/archive/v_11.tar.gz',
   keywords = [
     'ML',
     'microcontrollers',