Spaces:
Build error
Build error
Commit
•
58047f4
1
Parent(s):
d53bfda
Upload app.py
Browse files
app.py
ADDED
@@ -0,0 +1,294 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""Gradio demo for different clustering techiniques
|
2 |
+
|
3 |
+
Derived from https://scikit-learn.org/stable/auto_examples/cluster/plot_cluster_comparison.html
|
4 |
+
|
5 |
+
"""
|
6 |
+
|
7 |
+
import math
|
8 |
+
from functools import partial
|
9 |
+
|
10 |
+
import gradio as gr
|
11 |
+
import matplotlib.pyplot as plt
|
12 |
+
import numpy as np
|
13 |
+
from sklearn.cluster import (
|
14 |
+
AgglomerativeClustering, Birch, DBSCAN, KMeans, MeanShift, OPTICS, SpectralClustering, estimate_bandwidth
|
15 |
+
)
|
16 |
+
from sklearn.datasets import make_blobs, make_circles, make_moons
|
17 |
+
from sklearn.mixture import GaussianMixture
|
18 |
+
from sklearn.neighbors import kneighbors_graph
|
19 |
+
from sklearn.preprocessing import StandardScaler
|
20 |
+
|
21 |
+
|
22 |
+
plt.style.use('seaborn')
|
23 |
+
|
24 |
+
|
25 |
+
SEED = 0
|
26 |
+
MAX_CLUSTERS = 10
|
27 |
+
N_SAMPLES = 1000
|
28 |
+
N_COLS = 3
|
29 |
+
FIGSIZE = 7, 7 # does not affect size in webpage
|
30 |
+
COLORS = [
|
31 |
+
'blue', 'orange', 'green', 'red', 'purple', 'brown', 'pink', 'gray', 'olive', 'cyan'
|
32 |
+
]
|
33 |
+
assert len(COLORS) >= MAX_CLUSTERS, "Not enough different colors for all clusters"
|
34 |
+
np.random.seed(SEED)
|
35 |
+
|
36 |
+
|
37 |
+
def normalize(X):
|
38 |
+
return StandardScaler().fit_transform(X)
|
39 |
+
|
40 |
+
|
41 |
+
def get_regular(n_clusters):
|
42 |
+
# spiral pattern
|
43 |
+
centers = [
|
44 |
+
[0, 0],
|
45 |
+
[1, 0],
|
46 |
+
[1, 1],
|
47 |
+
[0, 1],
|
48 |
+
[-1, 1],
|
49 |
+
[-1, 0],
|
50 |
+
[-1, -1],
|
51 |
+
[0, -1],
|
52 |
+
[1, -1],
|
53 |
+
[2, -1],
|
54 |
+
][:n_clusters]
|
55 |
+
assert len(centers) == n_clusters
|
56 |
+
X, labels = make_blobs(n_samples=N_SAMPLES, centers=centers, cluster_std=0.25, random_state=SEED)
|
57 |
+
return normalize(X), labels
|
58 |
+
|
59 |
+
|
60 |
+
def get_circles(n_clusters):
|
61 |
+
X, labels = make_circles(n_samples=N_SAMPLES, factor=0.5, noise=0.05, random_state=SEED)
|
62 |
+
return normalize(X), labels
|
63 |
+
|
64 |
+
|
65 |
+
def get_moons(n_clusters):
|
66 |
+
X, labels = make_moons(n_samples=N_SAMPLES, noise=0.05, random_state=SEED)
|
67 |
+
return normalize(X), labels
|
68 |
+
|
69 |
+
|
70 |
+
def get_noise(n_clusters):
|
71 |
+
np.random.seed(SEED)
|
72 |
+
X, labels = np.random.rand(N_SAMPLES, 2), np.random.randint(0, n_clusters, size=(N_SAMPLES,))
|
73 |
+
return normalize(X), labels
|
74 |
+
|
75 |
+
|
76 |
+
def get_anisotropic(n_clusters):
|
77 |
+
X, labels = make_blobs(n_samples=N_SAMPLES, centers=n_clusters, random_state=170)
|
78 |
+
transformation = [[0.6, -0.6], [-0.4, 0.8]]
|
79 |
+
X = np.dot(X, transformation)
|
80 |
+
return X, labels
|
81 |
+
|
82 |
+
|
83 |
+
def get_varied(n_clusters):
|
84 |
+
cluster_std = [1.0, 2.5, 0.5, 1.0, 2.5, 0.5, 1.0, 2.5, 0.5, 1.0][:n_clusters]
|
85 |
+
assert len(cluster_std) == n_clusters
|
86 |
+
X, labels = make_blobs(
|
87 |
+
n_samples=N_SAMPLES, centers=n_clusters, cluster_std=cluster_std, random_state=SEED
|
88 |
+
)
|
89 |
+
return normalize(X), labels
|
90 |
+
|
91 |
+
|
92 |
+
def get_spiral(n_clusters):
|
93 |
+
# from https://scikit-learn.org/stable/auto_examples/cluster/plot_agglomerative_clustering.html
|
94 |
+
np.random.seed(SEED)
|
95 |
+
t = 1.5 * np.pi * (1 + 3 * np.random.rand(1, N_SAMPLES))
|
96 |
+
x = t * np.cos(t)
|
97 |
+
y = t * np.sin(t)
|
98 |
+
X = np.concatenate((x, y))
|
99 |
+
X += 0.7 * np.random.randn(2, N_SAMPLES)
|
100 |
+
X = np.ascontiguousarray(X.T)
|
101 |
+
|
102 |
+
labels = np.zeros(N_SAMPLES, dtype=int)
|
103 |
+
return normalize(X), labels
|
104 |
+
|
105 |
+
|
106 |
+
DATA_MAPPING = {
|
107 |
+
'regular': get_regular,
|
108 |
+
'circles': get_circles,
|
109 |
+
'moons': get_moons,
|
110 |
+
'spiral': get_spiral,
|
111 |
+
'noise': get_noise,
|
112 |
+
'anisotropic': get_anisotropic,
|
113 |
+
'varied': get_varied,
|
114 |
+
}
|
115 |
+
|
116 |
+
|
117 |
+
def get_groundtruth_model(X, labels, n_clusters, **kwargs):
|
118 |
+
# dummy model to show true label distribution
|
119 |
+
class Dummy:
|
120 |
+
def __init__(self, y):
|
121 |
+
self.labels_ = labels
|
122 |
+
|
123 |
+
return Dummy(labels)
|
124 |
+
|
125 |
+
|
126 |
+
def get_kmeans(X, labels, n_clusters, **kwargs):
|
127 |
+
model = KMeans(init="k-means++", n_clusters=n_clusters, n_init=10, random_state=SEED)
|
128 |
+
model.set_params(**kwargs)
|
129 |
+
return model.fit(X)
|
130 |
+
|
131 |
+
|
132 |
+
def get_dbscan(X, labels, n_clusters, **kwargs):
|
133 |
+
model = DBSCAN(eps=0.3)
|
134 |
+
model.set_params(**kwargs)
|
135 |
+
return model.fit(X)
|
136 |
+
|
137 |
+
|
138 |
+
def get_agglomerative(X, labels, n_clusters, **kwargs):
|
139 |
+
connectivity = kneighbors_graph(
|
140 |
+
X, n_neighbors=n_clusters, include_self=False
|
141 |
+
)
|
142 |
+
# make connectivity symmetric
|
143 |
+
connectivity = 0.5 * (connectivity + connectivity.T)
|
144 |
+
model = AgglomerativeClustering(
|
145 |
+
n_clusters=n_clusters, linkage="ward", connectivity=connectivity
|
146 |
+
)
|
147 |
+
model.set_params(**kwargs)
|
148 |
+
return model.fit(X)
|
149 |
+
|
150 |
+
|
151 |
+
def get_meanshift(X, labels, n_clusters, **kwargs):
|
152 |
+
bandwidth = estimate_bandwidth(X, quantile=0.25)
|
153 |
+
model = MeanShift(bandwidth=bandwidth, bin_seeding=True)
|
154 |
+
model.set_params(**kwargs)
|
155 |
+
return model.fit(X)
|
156 |
+
|
157 |
+
|
158 |
+
def get_spectral(X, labels, n_clusters, **kwargs):
|
159 |
+
model = SpectralClustering(
|
160 |
+
n_clusters=n_clusters,
|
161 |
+
eigen_solver="arpack",
|
162 |
+
affinity="nearest_neighbors",
|
163 |
+
)
|
164 |
+
model.set_params(**kwargs)
|
165 |
+
return model.fit(X)
|
166 |
+
|
167 |
+
|
168 |
+
def get_optics(X, labels, n_clusters, **kwargs):
|
169 |
+
model = OPTICS(
|
170 |
+
min_samples=7,
|
171 |
+
xi=0.05,
|
172 |
+
min_cluster_size=0.1,
|
173 |
+
)
|
174 |
+
model.set_params(**kwargs)
|
175 |
+
return model.fit(X)
|
176 |
+
|
177 |
+
|
178 |
+
def get_birch(X, labels, n_clusters, **kwargs):
|
179 |
+
model = Birch(n_clusters=n_clusters)
|
180 |
+
model.set_params(**kwargs)
|
181 |
+
return model.fit(X)
|
182 |
+
|
183 |
+
|
184 |
+
def get_gaussianmixture(X, labels, n_clusters, **kwargs):
|
185 |
+
model = GaussianMixture(
|
186 |
+
n_components=n_clusters, covariance_type="full", random_state=SEED,
|
187 |
+
)
|
188 |
+
model.set_params(**kwargs)
|
189 |
+
return model.fit(X)
|
190 |
+
|
191 |
+
|
192 |
+
MODEL_MAPPING = {
|
193 |
+
'True labels': get_groundtruth_model,
|
194 |
+
'KMeans': get_kmeans,
|
195 |
+
'DBSCAN': get_dbscan,
|
196 |
+
'MeanShift': get_meanshift,
|
197 |
+
'SpectralClustering': get_spectral,
|
198 |
+
'OPTICS': get_optics,
|
199 |
+
'Birch': get_birch,
|
200 |
+
'GaussianMixture': get_gaussianmixture,
|
201 |
+
'AgglomerativeClustering': get_agglomerative,
|
202 |
+
}
|
203 |
+
|
204 |
+
|
205 |
+
def plot_clusters(ax, X, labels):
|
206 |
+
set_clusters = set(labels)
|
207 |
+
set_clusters.discard(-1) # -1 signifiies outliers, which we plot separately
|
208 |
+
for label, color in zip(sorted(set_clusters), COLORS):
|
209 |
+
idx = labels == label
|
210 |
+
if not sum(idx):
|
211 |
+
continue
|
212 |
+
ax.scatter(X[idx, 0], X[idx, 1], color=color)
|
213 |
+
|
214 |
+
# show outliers (if any)
|
215 |
+
idx = labels == -1
|
216 |
+
if sum(idx):
|
217 |
+
ax.scatter(X[idx, 0], X[idx, 1], c='k', marker='x')
|
218 |
+
|
219 |
+
ax.grid(None)
|
220 |
+
ax.set_xticks([])
|
221 |
+
ax.set_yticks([])
|
222 |
+
return ax
|
223 |
+
|
224 |
+
|
225 |
+
def cluster(dataset: str, n_clusters: int, clustering_algorithm: str):
|
226 |
+
if isinstance(n_clusters, dict):
|
227 |
+
n_clusters = n_clusters['value']
|
228 |
+
else:
|
229 |
+
n_clusters = int(n_clusters)
|
230 |
+
|
231 |
+
X, labels = DATA_MAPPING[dataset](n_clusters)
|
232 |
+
model = MODEL_MAPPING[clustering_algorithm](X, labels, n_clusters=n_clusters)
|
233 |
+
if hasattr(model, "labels_"):
|
234 |
+
y_pred = model.labels_.astype(int)
|
235 |
+
else:
|
236 |
+
y_pred = model.predict(X)
|
237 |
+
|
238 |
+
fig, ax = plt.subplots(figsize=FIGSIZE)
|
239 |
+
|
240 |
+
plot_clusters(ax, X, y_pred)
|
241 |
+
ax.set_title(clustering_algorithm, fontsize=16)
|
242 |
+
|
243 |
+
return fig
|
244 |
+
|
245 |
+
|
246 |
+
title = "Clustering with Scikit-learn"
|
247 |
+
description = (
|
248 |
+
"This example shows how different clustering algorithms work. Simply pick "
|
249 |
+
"the dataset and the number of clusters to see how the clustering algorithms work. "
|
250 |
+
"Colored cirles are (predicted) labels and black x are outliers."
|
251 |
+
)
|
252 |
+
|
253 |
+
|
254 |
+
def iter_grid(n_rows, n_cols):
|
255 |
+
# create a grid using gradio Block
|
256 |
+
for _ in range(n_rows):
|
257 |
+
with gr.Row():
|
258 |
+
for _ in range(n_cols):
|
259 |
+
with gr.Column():
|
260 |
+
yield
|
261 |
+
|
262 |
+
|
263 |
+
with gr.Blocks(title=title) as demo:
|
264 |
+
gr.HTML(f"<b>{title}</b>")
|
265 |
+
gr.Markdown(description)
|
266 |
+
|
267 |
+
input_models = list(MODEL_MAPPING)
|
268 |
+
input_data = gr.Radio(
|
269 |
+
list(DATA_MAPPING),
|
270 |
+
value="regular",
|
271 |
+
label="dataset"
|
272 |
+
)
|
273 |
+
input_n_clusters = gr.Slider(
|
274 |
+
minimum=1,
|
275 |
+
maximum=MAX_CLUSTERS,
|
276 |
+
value=4,
|
277 |
+
step=1,
|
278 |
+
label='Number of clusters'
|
279 |
+
)
|
280 |
+
n_rows = int(math.ceil(len(input_models) / N_COLS))
|
281 |
+
counter = 0
|
282 |
+
for _ in iter_grid(n_rows, N_COLS):
|
283 |
+
if counter >= len(input_models):
|
284 |
+
break
|
285 |
+
|
286 |
+
input_model = input_models[counter]
|
287 |
+
plot = gr.Plot(label=input_model)
|
288 |
+
fn = partial(cluster, clustering_algorithm=input_model)
|
289 |
+
input_data.change(fn=fn, inputs=[input_data, input_n_clusters], outputs=plot)
|
290 |
+
input_n_clusters.change(fn=fn, inputs=[input_data, input_n_clusters], outputs=plot)
|
291 |
+
counter += 1
|
292 |
+
|
293 |
+
|
294 |
+
demo.launch()
|