Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,118 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import time
|
3 |
+
import numpy as np
|
4 |
+
import matplotlib.pyplot as plt
|
5 |
+
from sklearn.datasets import load_iris
|
6 |
+
from sklearn.model_selection import train_test_split
|
7 |
+
from sklearn.feature_selection import SelectKBest, f_classif
|
8 |
+
from sklearn.pipeline import make_pipeline
|
9 |
+
from sklearn.preprocessing import MinMaxScaler
|
10 |
+
from sklearn.svm import LinearSVC
|
11 |
+
|
12 |
+
theme = gr.themes.Monochrome(
|
13 |
+
primary_hue="indigo",
|
14 |
+
secondary_hue="blue",
|
15 |
+
neutral_hue="slate",
|
16 |
+
)
|
17 |
+
model_card = f"""
|
18 |
+
## Description
|
19 |
+
|
20 |
+
**Univariate feature selection** can be used to improve classification accuracy on a noisy dataset.
|
21 |
+
In **univariate feature selection**, each feature is evaluated independently, and a statistical test is used to determine its strength of association with the target variable.
|
22 |
+
The most important features are then selected based on their statistical significance, typically using a threshold p-value or a pre-defined number of top features to select.
|
23 |
+
|
24 |
+
In this demo, some noisy (non informative) features are added to the iris dataset then use **Support vector machine (SVM)** to classify the Iris dataset both before and after applying univariate feature selection.
|
25 |
+
The results of the feature selection are presented through p-values and weights of SVMs, which are plotted for comparison.
|
26 |
+
The objective of this demo is to evaluate the accuracy of the models and assess the impact of univariate feature selection on the model weights.
|
27 |
+
You can play around with different ``number of top features`` and ``random seed``.
|
28 |
+
|
29 |
+
## Dataset
|
30 |
+
|
31 |
+
Iris dataset
|
32 |
+
"""
|
33 |
+
# The iris dataset
|
34 |
+
X, y = load_iris(return_X_y=True)
|
35 |
+
|
36 |
+
# Some noisy data not correlated
|
37 |
+
E = np.random.RandomState(42).uniform(0, 0.1, size=(X.shape[0], 20))
|
38 |
+
|
39 |
+
# Add the noisy data to the informative features
|
40 |
+
X = np.hstack((X, E))
|
41 |
+
|
42 |
+
|
43 |
+
def do_train(k_features, random_state):
|
44 |
+
# Split dataset to select feature and evaluate the classifier
|
45 |
+
X_train, X_test, y_train, y_test = train_test_split(X, y, stratify=y, random_state=random_state)
|
46 |
+
selector = SelectKBest(f_classif, k=k_features)
|
47 |
+
selector.fit(X_train, y_train)
|
48 |
+
scores = -np.log10(selector.pvalues_)
|
49 |
+
scores /= scores.max()
|
50 |
+
|
51 |
+
|
52 |
+
fig1, axes1 = plt.subplots()
|
53 |
+
X_indices = np.arange(X.shape[-1])
|
54 |
+
axes1.bar(X_indices - 0.05, scores, width=0.2)
|
55 |
+
axes1.set_title("Feature univariate score")
|
56 |
+
axes1.set_xlabel("Feature number")
|
57 |
+
axes1.set_ylabel(r"Univariate score ($-Log(p_{value})$)")
|
58 |
+
|
59 |
+
clf = make_pipeline(MinMaxScaler(), LinearSVC())
|
60 |
+
clf.fit(X_train, y_train)
|
61 |
+
|
62 |
+
svm_weights = np.abs(clf[-1].coef_).sum(axis=0)
|
63 |
+
svm_weights /= svm_weights.sum()
|
64 |
+
|
65 |
+
clf_selected = make_pipeline(SelectKBest(f_classif, k=k_features), MinMaxScaler(), LinearSVC())
|
66 |
+
clf_selected.fit(X_train, y_train)
|
67 |
+
|
68 |
+
svm_weights_selected = np.abs(clf_selected[-1].coef_).sum(axis=0)
|
69 |
+
svm_weights_selected /= svm_weights_selected.sum()
|
70 |
+
|
71 |
+
fig2, axes2 = plt.subplots()
|
72 |
+
axes2.bar(
|
73 |
+
X_indices - 0.45, scores, width=0.2, label=r"Univariate score ($-Log(p_{value})$)"
|
74 |
+
)
|
75 |
+
|
76 |
+
axes2.bar(X_indices - 0.25, svm_weights, width=0.2, label="SVM weight")
|
77 |
+
|
78 |
+
axes2.bar(
|
79 |
+
X_indices[selector.get_support()] - 0.05,
|
80 |
+
svm_weights_selected,
|
81 |
+
width=0.2,
|
82 |
+
label="SVM weights after selection",
|
83 |
+
)
|
84 |
+
|
85 |
+
axes2.set_title("Comparing feature selection")
|
86 |
+
axes2.set_xlabel("Feature number")
|
87 |
+
axes2.set_yticks(())
|
88 |
+
axes2.axis("tight")
|
89 |
+
axes2.legend(loc="upper right")
|
90 |
+
|
91 |
+
text = f"Classification accuracy without selecting features: {clf.score(X_test, y_test)*100:.2f}%. Classification accuracy after univariate feature selection: {clf_selected.score(X_test, y_test)*100:.2f}%"
|
92 |
+
|
93 |
+
return fig1, fig2, text
|
94 |
+
|
95 |
+
|
96 |
+
|
97 |
+
with gr.Blocks(theme=theme) as demo:
|
98 |
+
gr.Markdown('''
|
99 |
+
<div>
|
100 |
+
<h1 style='text-align: center'>Univariate Feature Selection</h1>
|
101 |
+
</div>
|
102 |
+
''')
|
103 |
+
gr.Markdown(model_card)
|
104 |
+
gr.Markdown("Author: <a href=\"https://huggingface.co/vumichien\">Vu Minh Chien</a>. Based on the example from <a href=\"https://scikit-learn.org/stable/auto_examples/feature_selection/plot_feature_selection.html#sphx-glr-auto-examples-feature-selection-plot-feature-selection-py\">scikit-learn</a>")
|
105 |
+
k_features = gr.Slider(minimum=2, maximum=10, step=1, value=2, label="Number of top features to select")
|
106 |
+
random_state = gr.Slider(minimum=0, maximum=2000, step=1, value=0, label="Random seed")
|
107 |
+
with gr.Row():
|
108 |
+
with gr.Column():
|
109 |
+
plot_1 = gr.Plot(label="Univariate score")
|
110 |
+
with gr.Column():
|
111 |
+
plot_2 = gr.Plot(label="Comparing feature selection")
|
112 |
+
with gr.Row():
|
113 |
+
resutls = gr.Textbox(label="Results")
|
114 |
+
|
115 |
+
k_features.change(fn=do_train, inputs=[k_features, random_state], outputs=[plot_1, plot_2, resutls])
|
116 |
+
random_state.change(fn=do_train, inputs=[k_features, random_state], outputs=[plot_1, plot_2, resutls])
|
117 |
+
|
118 |
+
demo.launch()
|