-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathsession_level.py
313 lines (276 loc) · 12.6 KB
/
session_level.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
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
import os
import tqdm
import pandas as pd
import numpy as np
import argparse
import editdistance
from collections import Counter, OrderedDict
import matplotlib.pyplot as plt
from utils import *
def df_mark_session(dataset_name, data_path, threshold, save_root):
df = pd.read_csv(data_path)
df = df.sort_values(['author_id', 'timestamp'])
user_zfill = len(str(len(set(df['author_id'].tolist()))))
length = len(df)
user_list = []
session_list = []
num_queries = []
lengths_per_user = []
time_interval = []
for i, (_, row) in enumerate(df.iterrows()):
tqdm_df(length, i)
author_id = row['author_id']
ts = cut_timestamp_sec(row['timestamp'], dataset_name)
if i == 0:
prev_id = author_id
prev_ts = ts
user_idx = 0
session_idx = 0
session_list.append(str(user_idx).zfill(user_zfill)+'_'+str(session_idx))
user_list.append(str(user_idx).zfill(user_zfill))
interval = []
num_q = 1
else:
if author_id == prev_id:
time_diff = get_time_diff(prev_ts, ts)
if time_diff > threshold:
prev_ts = ts
session_idx += 1
num_queries.append(num_q)
if len(interval) > 0:
time_interval.append((num_q, np.array(interval).mean()))
num_q = 1
interval = []
else:
num_q += 1
interval.append(time_diff)
else:
prev_id = author_id
prev_ts = ts
lengths_per_user.append(session_idx+1)
num_queries.append(num_q)
if len(interval) > 0:
time_interval.append((num_q, np.array(interval).mean()))
user_idx += 1
num_q = 1
interval = []
session_idx = 0
session_list.append(str(user_idx).zfill(user_zfill)+'_'+str(session_idx))
user_list.append(str(user_idx).zfill(user_zfill))
prev_ts = ts
lengths_per_user.append(session_idx+1)
num_queries.append(num_q)
if len(interval) > 0:
time_interval.append((num_q, np.array(interval).mean()))
df['session'] = session_list
df['user'] = user_list
print('statistics of number of queries per session')
_, _, _, _ = data_statistics(num_queries)
print('statistics of number of sessions per user')
_, _, _, _ = data_statistics(lengths_per_user)
fn = os.path.join(save_root, '{}_session_{}min.csv'.format(dataset_name, threshold))
df.to_csv(fn, index=False)
return fn
def get_edit_distance(session_df_path):
df = pd.read_csv(session_df_path)
all_edit_distance = []
for session_idx in tqdm.tqdm(set(df['session'].tolist())):
data = df.loc[df['session'] == session_idx]
tokens = read_tokens(data)
if len(tokens) >= 2:
for i in range(len(tokens) - 1):
all_edit_distance.append(editdistance.eval(tokens[i], tokens[i+1]))
else:
continue
_, _, _, _ = data_statistics(all_edit_distance)
def find_all_repeat(dataset_name, session_df_path, save_root):
session_df = pd.read_csv(session_df_path)
threshold = os.path.basename(session_df_path).split('min')[-2].split('_')[-1]
freq_dict = {}
all_sessions = list(set(session_df['session'].tolist()))
for session_idx in tqdm.tqdm(all_sessions):
data = session_df.loc[session_df['session'] == session_idx]
tokens = read_tokens(data)
for t in tokens:
p = ' '.join(t)
if p not in freq_dict.keys():
freq_dict[p] = [1, [session_idx]]
else:
freq_dict[p][0] += 1
freq_dict[p][1].append(session_idx)
freq_dict = cut_list_dict(dict(OrderedDict(sorted(freq_dict.items(), key=lambda t: t[1][0], reverse=True))), value_index=1)
num_users = []
num_per_user = []
stats = []
for value in freq_dict.values():
temp = dict(OrderedDict(sorted(dict(Counter([s_id.split('_')[0] for s_id in value[1]])).items(), key=lambda t: t[0], reverse=True)))
num_users.append(len(temp))
num_per_user.append(list(temp.values()))
stats.append(data_statistics(list(temp.values()), enable_print=False))
df = pd.DataFrame({"prompt": list(freq_dict.keys()),
"overall_freq": [value[0] for value in list(freq_dict.values())],
"num_users": num_users,
"num_per_user": num_per_user,
"stats": stats,
"session_idx": [list(set(value[1])) for value in list(freq_dict.values())],
"all_session_idx": [value[1] for value in list(freq_dict.values())]
})
fn = os.path.join(save_root, '{}_overall_repeats_{}min.csv'.format(dataset_name, threshold))
df.to_csv(fn, index=False)
return fn
def freq_across_user(dataset_name, repeats_df_path, save_root):
repeats_df = pd.read_csv(repeats_df_path)
threshold = os.path.basename(repeats_df_path).split('min')[-2].split('_')[-1]
df = repeats_df[['prompt', 'num_users']]
df.sort_values('num_users', ascending=False)
num_users = df['num_users'].tolist()
plt.figure(figsize=(15,6))
_ = plt.hist(num_users, bins=50, width=5)
plt.yscale('log')
plt.xticks(np.linspace(0, 400, 9))
plt.grid(alpha=0.5)
plt.xlabel('Shared by #users')
plt.ylabel('#Prompts (log-scale)')
fn = os.path.join(save_root, '{}_top100_prompt_across_users_{}min.csv'.format(dataset_name, threshold))
df.head(100).to_csv(fn, index=False)
return fn
def find_repeat_word(tokens1, tokens2):
tokens1_freq = dict(Counter(tokens1))
tokens2_freq = dict(Counter(tokens2))
if abs(len(tokens1) - len(tokens2)) == 1:
diff = set(tokens1_freq.items()) ^ set(tokens2_freq.items())
return list(diff)[0][0]
else:
edit1 = None
edit2 = None
for t, freq1 in tokens1_freq.items():
if t in tokens2_freq.keys():
freq2 = tokens2_freq[t]
if freq1 - freq2 == 1:
edit1 = t
if freq2 - freq1 == 1:
edit2 = t
else:
edit1 = t
if edit1 is not None and edit2 is None:
for t, freq2 in tokens2_freq.items():
if t not in tokens1_freq.keys():
edit2 = t
return [edit1, edit2]
def rank_edits(edit_df, edit_name):
freq = dict(OrderedDict(sorted(dict(Counter(edit_df[edit_name].tolist())).items(), key=lambda t: t[1], reverse=True)))
df = pd.DataFrame({edit_name: list(freq.keys()),
'freq': list(freq.values())})
return df
def rank_replace(replaced_df):
freq = {}
for item in replaced_df['replaced'].tolist():
if tuple(item) not in freq.keys():
freq[tuple(item)] = 1
else:
freq[tuple(item)] += 1
freq = dict(OrderedDict(sorted(freq.items(), key=lambda t: t[1], reverse=True)))
df = pd.DataFrame({'replaced': list(freq.keys()),
'freq': list(freq.values())})
return df
def find_editted_word(dataset_name, data_path, threshold, save_root):
df = pd.read_csv(data_path)
df = df.sort_values(['author_id', 'timestamp'])
user_zfill = len(str(len(set(df['author_id'].tolist()))))
length = len(df)
session_list = []
added = []
deleted = []
replaced = []
all_edits = []
edit_type = []
for i, (_, row) in enumerate(df.iterrows()):
# tqdm_df(length, i)
author_id = row['author_id']
ts = cut_timestamp_sec(row['timestamp'], dataset_name)
tokens = literal_eval(row['tokenized'])
if i == 0:
prev_id = author_id
prev_ts = ts
prev_tokens = tokens
user_idx = 0
session_idx = 0
session_list.append(str(user_idx).zfill(user_zfill)+'_'+str(session_idx))
all_edits.append('nan')
edit_type.append('nan')
else:
if author_id == prev_id:
time_diff = get_time_diff(prev_ts, ts)
if time_diff > threshold:
session_idx += 1
all_edits.append('nan')
edit_type.append('nan')
else:
if editdistance.eval(prev_tokens, tokens) == 1:
edits = find_repeat_word(prev_tokens, tokens)
if len(tokens) - len(prev_tokens) == 1:
added.append((' '.join(prev_tokens), ' '.join(tokens), edits))
all_edits.append(edits)
edit_type.append('add')
elif len(prev_tokens) - len(tokens) == 1:
deleted.append((' '.join(prev_tokens), ' '.join(tokens), edits))
all_edits.append(edits)
edit_type.append('delete')
else:
if not 'https://s.mj.run/' in edits[0] and not 'https://s.mj.run/' in edits[1]:
replaced.append((' '.join(prev_tokens), ' '.join(tokens), edits))
all_edits.append(edits)
edit_type.append('replace')
else:
all_edits.append('nan')
edit_type.append('nan')
else:
all_edits.append('nan')
edit_type.append('nan')
else:
prev_id = author_id
user_idx += 1
session_idx = 0
all_edits.append('nan')
edit_type.append('nan')
session_list.append(str(user_idx).zfill(user_zfill)+'_'+str(session_idx))
prev_ts = ts
prev_tokens = tokens
add_df = pd.DataFrame({'prompt1': [list(item)[0] for item in added],
'prompt2': [list(item)[1] for item in added],
'added': [list(item)[2] for item in added]})
deleted_df = pd.DataFrame({'prompt1': [list(item)[0] for item in deleted],
'prompt2': [list(item)[1] for item in deleted],
'deleted': [list(item)[2] for item in deleted]})
replaced_df = pd.DataFrame({'prompt1': [list(item)[0] for item in replaced],
'prompt2': [list(item)[1] for item in replaced],
'replaced': [list(item)[2] for item in replaced]})
rank_edits(add_df, 'added').to_csv(os.path.join(save_root, '{}_added_freq.csv'.format(dataset_name)), index=False)
rank_edits(deleted_df, 'deleted').to_csv(os.path.join(save_root, '{}_deleted_freq.csv'.format(dataset_name)), index=False)
fn = os.path.join(save_root, '{}_replaced_freq.csv'.format(dataset_name))
rank_replace(replaced_df).to_csv(fn, index=False)
df['edits'] = all_edits
df['edits_type'] = edit_type
df.to_csv(os.path.join(save_root, '{}_edits.csv'.format(dataset_name)))
return fn
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--dataset_name', required=True, type=str, choices=['midjourney', 'diffusiondb'])
parser.add_argument('--data_path', required=True, type=str)
parser.add_argument('--save_root', required=True, type=str)
parser.add_argument('--threshold', required=True, type=int, default=30)
args = parser.parse_args()
print('Begin to group by sessions with threhold of {}min...'.format(args.threshold))
session_df_path = df_mark_session(args.dataset_name, args.data_path, args.threshold, args.save_root)
print('Results saved to {}.'.format(session_df_path))
print('Begin to find repeats...'.format(args.threshold))
repeats_df_path = find_all_repeat(args.dataset_name, session_df_path, args.save_root)
print('Results saved to {}.'.format(repeats_df_path))
print('Begin to rank most frequent prompts across user...')
top_prompts_across_user_path = freq_across_user(args.dataset_name, repeats_df_path, args.save_root)
print('Results saved to {}.'.format(top_prompts_across_user_path))
print('Statistics of edit distance within sessions...')
get_edit_distance(session_df_path)
print('Begin to find editted word (added/deleted/replaced) within sessions...')
replacement_path = find_editted_word(args.dataset_name, args.data_path, args.threshold, args.save_root)
print('Results of word replacements saved to {}.'.format(replacement_path))