Spaces:
Sleeping
Sleeping
Add application files
Browse files- app.py +126 -0
- requirements.txt +3 -0
app.py
ADDED
@@ -0,0 +1,126 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from functools import partial
|
2 |
+
|
3 |
+
import gradio as gr
|
4 |
+
import matplotlib.pyplot as plt
|
5 |
+
from matplotlib.ticker import NullFormatter
|
6 |
+
import numpy as np
|
7 |
+
from sklearn import datasets, manifold
|
8 |
+
|
9 |
+
|
10 |
+
SEED = 0
|
11 |
+
N_COMPONENTS = 2
|
12 |
+
np.random.seed(SEED)
|
13 |
+
|
14 |
+
|
15 |
+
def get_circles(n_samples):
|
16 |
+
X, color = datasets.make_circles(
|
17 |
+
n_samples=n_samples,
|
18 |
+
factor=0.5,
|
19 |
+
noise=0.05,
|
20 |
+
random_state=SEED
|
21 |
+
)
|
22 |
+
return X, color
|
23 |
+
|
24 |
+
|
25 |
+
def get_s_curve(n_samples):
|
26 |
+
X, color = datasets.make_s_curve(n_samples=n_samples, random_state=SEED)
|
27 |
+
X[:, 1], X[:, 2] = X[:, 2], X[:, 1].copy()
|
28 |
+
return X, color
|
29 |
+
|
30 |
+
|
31 |
+
def get_uniform_grid(n_samples):
|
32 |
+
x = np.linspace(0, 1, int(np.sqrt(n_samples)))
|
33 |
+
xx, yy = np.meshgrid(x, x)
|
34 |
+
X = np.hstack(
|
35 |
+
[
|
36 |
+
xx.ravel().reshape(-1, 1),
|
37 |
+
yy.ravel().reshape(-1, 1),
|
38 |
+
]
|
39 |
+
)
|
40 |
+
color = xx.ravel()
|
41 |
+
return X, color
|
42 |
+
|
43 |
+
|
44 |
+
DATA_MAPPING = {
|
45 |
+
'circles': get_circles,
|
46 |
+
's-curve': get_s_curve,
|
47 |
+
'uniform grid': get_uniform_grid,
|
48 |
+
}
|
49 |
+
|
50 |
+
|
51 |
+
|
52 |
+
def plot_data(dataset: str, perplexity: int, n_samples: int, tsne: bool):
|
53 |
+
if isinstance(perplexity, dict):
|
54 |
+
perplexity = perplexity['value']
|
55 |
+
else:
|
56 |
+
perplexity = int(perplexity)
|
57 |
+
|
58 |
+
X, color = DATA_MAPPING[dataset](n_samples)
|
59 |
+
if tsne:
|
60 |
+
tsne = manifold.TSNE(
|
61 |
+
n_components=N_COMPONENTS,
|
62 |
+
init="random",
|
63 |
+
random_state=0,
|
64 |
+
perplexity=perplexity,
|
65 |
+
n_iter=400,
|
66 |
+
)
|
67 |
+
Y = tsne.fit_transform(X)
|
68 |
+
else:
|
69 |
+
Y = X
|
70 |
+
|
71 |
+
fig, ax = plt.subplots(figsize=(7, 7))
|
72 |
+
|
73 |
+
ax.scatter(Y[:, 0], Y[:, 1], c=color)
|
74 |
+
ax.xaxis.set_major_formatter(NullFormatter())
|
75 |
+
ax.yaxis.set_major_formatter(NullFormatter())
|
76 |
+
ax.axis("tight")
|
77 |
+
|
78 |
+
return fig
|
79 |
+
|
80 |
+
|
81 |
+
title = "t-SNE: The effect of various perplexity values on the shape"
|
82 |
+
description = (
|
83 |
+
"An illustration of t-SNE on the two concentric circles and the"
|
84 |
+
"S-curve datasets for different perplexity values."
|
85 |
+
)
|
86 |
+
|
87 |
+
|
88 |
+
with gr.Blocks(title=title) as demo:
|
89 |
+
gr.HTML(f"<b>{title}</b>")
|
90 |
+
gr.Markdown(description)
|
91 |
+
|
92 |
+
input_data = gr.Radio(
|
93 |
+
list(DATA_MAPPING),
|
94 |
+
value="circles",
|
95 |
+
label="dataset"
|
96 |
+
)
|
97 |
+
n_samples = gr.Slider(
|
98 |
+
minimum=100,
|
99 |
+
maximum=1000,
|
100 |
+
value=150,
|
101 |
+
step=25,
|
102 |
+
label='Number of Samples'
|
103 |
+
)
|
104 |
+
perplexity = gr.Slider(
|
105 |
+
minimum=2,
|
106 |
+
maximum=100,
|
107 |
+
value=5,
|
108 |
+
step=1,
|
109 |
+
label='Perplexity'
|
110 |
+
)
|
111 |
+
with gr.Row():
|
112 |
+
with gr.Column():
|
113 |
+
plot = gr.Plot(label="Original data")
|
114 |
+
fn = partial(plot_data, tsne=False)
|
115 |
+
input_data.change(fn=fn, inputs=[input_data, perplexity, n_samples], outputs=plot)
|
116 |
+
perplexity.change(fn=fn, inputs=[input_data, perplexity, n_samples], outputs=plot)
|
117 |
+
n_samples.change(fn=fn, inputs=[input_data, perplexity, n_samples], outputs=plot)
|
118 |
+
with gr.Column():
|
119 |
+
plot = gr.Plot(label="t-SNE")
|
120 |
+
fn = partial(plot_data, tsne=True)
|
121 |
+
input_data.change(fn=fn, inputs=[input_data, perplexity, n_samples], outputs=plot)
|
122 |
+
perplexity.change(fn=fn, inputs=[input_data, perplexity, n_samples], outputs=plot)
|
123 |
+
n_samples.change(fn=fn, inputs=[input_data, perplexity, n_samples], outputs=plot)
|
124 |
+
|
125 |
+
|
126 |
+
demo.launch()
|
requirements.txt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
matplotlib
|
2 |
+
numpy
|
3 |
+
scikit-learn
|