Merge pull request #25 from opendilab/doc/practice

doc(hansbug): add practice pages

.gitignore

@@ -60,8 +60,6 @@ gen
 
 ### Images template
 # JPEG
-*.jpg
-*.jpeg
 *.jpe
 *.jif
 *.jfif

docs/source/best_practice/sklearn/index.rst

@@ -0,0 +1,35 @@
+Apply into Scikit-Learn
+===========================
+
+Actually, ``TreeValue`` can be used in practice with not only ``numpy`` or ``torch`` library, such as ``scikit-learn``.
+In the following part, a demo of PCA to tree-structured arrays will be shown.
+
+In the field of traditional machine learning, PCA (Principal Component Analysis) is often used to preprocess data,
+by normalizing the data range, and trying to reduce the dimensionality of the data, so as to reduce the complexity
+of the input data and improve machine learning's efficiency and quality. Just as the following image
+
+.. figure:: heading_of_pca.jpg
+    :alt: PCA Principle
+
+    PCA in a nutshell. Source: Lavrenko and Sutton 2011, slide 13.
+
+In the scikit-learn library, the PCA class is provided to support this function, and the function ``fit_transform``
+can be used to simplify the data. For a set of ``np.array`` format data that presents a tree structure,
+we can implement the operation support for the tree structure by quickly wrapping the function ``fit_transform``.
+The specific code is as follows
+
+.. literalinclude:: sklearn.demo.py
+    :language: python
+    :linenos:
+
+The output should be
+
+.. literalinclude:: sklearn.demo.py.txt
+    :language: text
+    :linenos:
+
+For further information, see the links below:
+
+* `Official documentation of PCA in scikit-learn <https://scikit-learn.org/stable/modules/generated/sklearn.decomposition.PCA.html?highlight=pca#sklearn.decomposition.PCA>`_
+* `Details of PCA <https://devopedia.org/principal-component-analysis>`_
+

docs/source/best_practice/sklearn/sklearn.demo.py

@@ -0,0 +1,20 @@
+import numpy as np
+from sklearn.decomposition import PCA
+
+from treevalue import FastTreeValue
+
+fit_transform = FastTreeValue.func()(lambda x: PCA(min(*x.shape)).fit_transform(x))
+
+if __name__ == '__main__':
+    data = FastTreeValue({
+        'a': np.random.randint(-5, 15, (4, 3)),
+        'x': {
+            'c': np.random.randint(-15, 5, (5, 4)),
+        }
+    })
+    print("Original int data:")
+    print(data)
+
+    pdata = fit_transform(data)
+    print("Fit transformed data:")
+    print(pdata)

docs/source/index.rst

@@ -26,6 +26,7 @@ structure processing when the calculation is tree-based.
     :caption: Best Practice
 
     best_practice/numpy/index
+    best_practice/sklearn/index
 
 .. toctree::
     :maxdepth: 2

requirements-doc.txt

@@ -7,4 +7,5 @@ packaging
 sphinx-multiversion~=0.2.4
 where~=1.0.2
 numpy>=1.19,<2
-easydict>=1.7,<2
+easydict>=1.7,<2
+scikit-learn>=0.24.2