/
utils.py
219 lines (191 loc) · 7.37 KB
/
utils.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
# -*- coding: utf-8 -*-
# @Author: twankim
# @Date: 2017-05-05 20:22:13
# @Last Modified by: twankim
# @Last Modified time: 2017-10-26 03:25:34
import numpy as np
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import pandas as pd
def accuracy(y_true,y_pred):
return 100*np.sum(y_true==y_pred)/float(len(y_true))
def mean_accuracy(y_true,y_pred):
labels = np.unique(y_true)
accuracy = np.zeros(len(labels))
hamming = y_true==y_pred
accuracy = [100*np.sum(hamming[y_true==label])/float(np.sum(y_true==label)) \
for label in labels]
return np.mean(accuracy)
def error(y_true,y_pred):
return 100*np.sum(y_true!=y_pred)/float(len(y_true))
def mean_error(y_true,y_pred):
labels = np.unique(y_true)
num_error = np.zeros(len(labels))
hamming = y_true!=y_pred
error = [100*np.sum(hamming[y_true==label])/float(np.sum(y_true==label)) \
for label in labels]
return np.mean(error)
# Find best matching permutation of y_pred clustering
# Also need to change mpp of algorithm
def find_permutation(dataset,algo):
# Calculate centers of original clustering
label_org = list(np.unique(dataset.y))
means_org = [np.mean(dataset.X[dataset.y==label,:],axis=0) for label in label_org]
labels_map = {} # Map from algorithm's label to true label
# Initialize label mapping
for label in xrange(algo.k+1):
labels_map[label] = 0
if len(algo.labels)==0:
return algo.y
for label,mpp in zip(algo.labels,algo.mpps):
# Calculate distance between estimated center and true centers
dist = [np.linalg.norm(mpp-mean_org) for mean_org in means_org]
# Assign true cluster label to the algorithm's label
idx_best = np.argmin(dist)
labels_map[label] = label_org[idx_best]
# Remove assigned label from the list
del means_org[idx_best]
del label_org[idx_best]
return [labels_map[y] for y in algo.y]
# Plot eta v.s. evaluation
# res: rep x len(qs) x len(etas)
def print_eval(eval_metric,res,etas,fname,
is_sum=False,weak='random',params=None):
assert weak in ['random','local','global'], \
"weak must be in ['random','local','global']"
if weak == 'random':
i_name = 'q'
t_name = weak
else:
i_name = 'c_dist'
t_name = weak +' distance'
rep = res.shape[0]
if not is_sum:
df_res = pd.DataFrame(res.mean(axis=0),
columns=etas,
index=params
)
df_res.index.name=i_name
df_res.columns.name='eta'
print "\n<{}. {}-weak (Averaged over {} experiments)>".format(
eval_metric,t_name, rep)
else:
df_res = pd.DataFrame(res.sum(axis=0),
columns=etas,
index=params
)
df_res.index.name=i_name
df_res.columns.name='eta'
print "\n<{}. {}-weak (Total Sum over {} experiments)>".format(
eval_metric,t_name,rep)
print df_res
df_res.to_csv(fname)
# Plot eta v.s. evaluation
# res: rep x len(qs) x len(etas)
def plot_eval(eval_metric,res,etas,fig_name,
is_sum=False,weak='random',params=None,res_org=None):
assert weak in ['random','local','global'], \
"weak must be in ['random','local','global']"
# cmap = plt.cm.get_cmap("jet", len(params)) -> cmap(i_p)
cmap = ['g','r','b','k','y','m','c']
if weak == 'random':
i_name = 'q'
t_name = weak
else:
i_name = 'c_{dist}'
t_name = weak + ' distance'
rep = res.shape[0]
if not is_sum:
res_plt = res.mean(axis=0)
res_org_plt = res_org.mean(axis=0)
f = plt.figure()
plt.title(r"{}. {}-weak (Averaged over {} experiments)".format(
eval_metric,t_name,rep))
for i_p,param in enumerate(params):
plt.plot(etas,res_plt[i_p,:],
'x-',c=cmap[i_p],
label=r'SSAC(ours) ${}={}$'.format(i_name,param))
if res_org is not None:
plt.plot(etas,res_org_plt[i_p,:],
'o--',c=cmap[i_p],
label=r'SSAC(original) ${}={}$'.format(i_name,param))
plt.xlabel(r"$\eta$ (Number of samples per cluster)")
plt.ylabel(eval_metric)
else:
res_plt = res.sum(axis=0)
res_org_plt = res_org.sum(axis=0)
f = plt.figure()
plt.title(r"{}. {}-weak (Total sum over {} experiments)".format(
eval_metric,t_name,rep))
for i_p,param in enumerate(params):
plt.plot(etas,res_plt[i_p,:],
'x-',c=cmap[i_p],
label=r'SSAC(ours) ${}={}$'.format(i_name,param))
if res_org is not None:
plt.plot(etas,res_org_plt[i_p,:],
'o--',c=cmap[i_p],
label=r'SSAC(oroginal) ${}={}$'.format(i_name,param))
plt.xlabel(r"$\eta$ (Number of samples per cluster)")
plt.ylabel(eval_metric)
if "accuracy" in eval_metric.lower():
plt.legend(loc=4)
min_val = min(res_plt.min(),res_org_plt.min())
max_val = max(res_plt.max(),res_org_plt.max())
ylim_min = min_val-(max_val-min_val)*0.55
ylim_max = max_val+(max_val-min_val)*0.05
elif ("error" in eval_metric.lower()) or ("fail" in eval_metric.lower()):
plt.legend(loc=1)
max_val = max(res_plt.max(),res_org_plt.max())
ylim_min = 0 - max_val*0.1
ylim_max = max_val*1.35
else:
plt.legend(loc=4)
plt.ylim([ylim_min,ylim_max])
plt.xlim([0,np.round(1.2*max(etas))])
f.savefig(fig_name,bbox_inches='tight')
def plot_hist(gammas,min_gamma,max_gamma,fig_name):
rep = len(gammas)
if rep>40:
n_bins = int(rep/20)
else:
n_bins = 10
f = plt.figure()
plt.hist(gammas,normed=False,bins=n_bins)
plt.title(r"Histogram of $\gamma$. min={}, max={} ({} generation)".format(min_gamma,max_gamma,rep))
plt.xlabel(r"$\gamma$")
plt.ylabel("Number of data generations")
f.savefig(fig_name,bbox_inches='tight')
def plot_cluster(X,y_true,y_pred,k,mpps,gamma,title,f_name,verbose,classes=None):
if classes is not None:
classes = classes
else:
classes = range(k+1)
cmap = plt.cm.get_cmap("jet", k+1)
if verbose:
print " ... Plotting"
f = plt.figure(figsize=(14,7))
plt.suptitle(title)
# Plot original clustering (k-means)
plt.subplot(121)
for i in xrange(1,k+1):
idx = y_true==i
plt.scatter(X[idx,0],X[idx,1],c=cmap(i),label=classes[i],alpha=0.7)
# plt.scatter(X[:,0],X[:,1],c=y_true,label=classes)
plt.title("True dataset ($\gamma$={:.2f})".format(gamma))
plt.legend()
# Plot SSAC result
plt.subplot(122)
for i in xrange(0,k+1):
idx = np.array(y_pred)==i
if sum(idx)>0:
plt.scatter(X[idx,0],X[idx,1],c=cmap(i),label=classes[i],alpha=0.7)
# plt.scatter(X[:,0],X[:,1],c=y_pred,label=classes)
plt.title("SSAC result ($\gamma$={:.2f})".format(gamma))
plt.legend()
# Plot estimated cluster centers
for t in xrange(k):
mpp = mpps[t]
plt.plot(mpp[0],mpp[1],'k^',ms=15,alpha=0.7)
f.savefig(f_name,bbox_inches='tight')
plt.close()