-
Notifications
You must be signed in to change notification settings - Fork 12
/
utils.py
124 lines (104 loc) · 4.93 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
import os
import numpy as np
import glob
from numba import jit
class Logger:
def __init__(self, path):
self.path = path
if path != '':
folder = '/'.join(path.split('/')[:-1])
if not os.path.exists(folder):
os.makedirs(folder)
def print(self, message):
print(message)
if self.path != '':
with open(self.path, 'a') as f:
f.write(message '\n')
f.flush()
def get_contiguous_indices(indices_opt_init):
# needed when indices_opt_init are not contiguous (e.g. mnist_2_6: coord=613 has [12, 13, 15, 16, 18])
running_diffs = indices_opt_init[1:] - indices_opt_init[:-1] # [1, 2, 1, 2]
where_change_contiguous_regions = np.where(running_diffs != 1)[0] # find the last contiguous index - 1
if len(where_change_contiguous_regions) > 0:
last_el_first_contiguous_region = where_change_contiguous_regions[0]
elif np.sum(running_diffs != np.ones(len(running_diffs))) == 0: # if all are optimal (mnist_2_6: coord=72)
last_el_first_contiguous_region = len(running_diffs)
elif len(indices_opt_init) == 1: # the easiest and most common situation - just 1 optimal index
last_el_first_contiguous_region = 0
else:
raise Exception('this case has not been handled')
indices_opt = indices_opt_init[:last_el_first_contiguous_region 1] # [12, 13]
return indices_opt
@jit(nopython=True)
def minimum(arr1, arr2):
# take element-wise minimum of 2 arrays compatible with numba (instead of np.minimum(arr1, arr2))
return arr1 * (arr1 < arr2) arr2 * (arr1 >= arr2)
@jit(nopython=True)
def clip(val, val_min, val_max):
# identical to np.clip
return min(max(val, val_min), val_max)
def print_arr(arr):
""" Pretty printing of a 2D numpy array. """
for i, row in enumerate(arr):
string = ''
for el in row:
string = '{:.3f} '.format(el)
print(i 1, string)
def extract_hyperparam(model_name, substr):
return model_name.split(substr)[1].split(' ')[0]
def finalize_curr_row(latex_str, weak_learner, flag_n_trees_latex):
# finalizing the current row: apply boldfacing and add \\
# (relies on the fact that we have only 3 metrics, i.e. TE,RTE,URTE or TE,LRTE,URTE or 4 metrics if flag_n_trees_latex is on)
# result: 'breast-cancer & 0.3 & 0.7 & 85.4 & 85.4 & 5.1 & 11.7 & 11.7 & 5.1 & 11.7 & 11.7'
curr_row = latex_str.split(r'\\')[-1]
curr_str_bf = ' & '.join(curr_row.split(' & ')[:2]) ' & '
metrics_str = ' & '.join(curr_row.split(' & ')[2:])
n_metrics = 4 if flag_n_trees_latex else 3
metrics_curr_row = dict([(i, []) for i in range(n_metrics)])
# result: {0: [0.7, 5.1, 5.1], 1: [85.4, 11.7, 11.7], 2: [85.4, 11.7, 11.7]}
for i_val, val_str in enumerate(metrics_str.split(' & ')):
# for n_trees we need int, for the rest float
val = int(val_str) if flag_n_trees_latex and i_val % n_metrics == n_metrics - 1 else float(val_str)
metrics_curr_row[i_val % n_metrics].append(val)
# form the boldfaced str that corresponds to the current row
for tup in zip(*metrics_curr_row.values()):
for i_m, m in enumerate(tup):
# boldfacing condition: if minimum and it's not the number of trees (if the flag is turned on)
if (m == min(metrics_curr_row[i_m]) and not (flag_n_trees_latex and i_m == 3) and
not (weak_learner == 'stump' and i_m == 2)): # if URTE for stumps, don't boldface
curr_str_bf = '\\textbf{' str(m) '} & '
else:
curr_str_bf = '{} & '.format(m)
curr_str_bf = ' ' # just a margin for better latex code quality
curr_str_bf = curr_str_bf.strip()[:-1] # get rid of the last ' & '
curr_row_final = curr_str_bf r'\\' '\n' # new table line
return curr_row_final
def get_model_names(datasets, models, exp_folder, weak_learner, tree_depth):
model_names = []
for dataset in datasets:
for model in models:
depth_str = 'max_depth=' str(tree_depth) if weak_learner == 'tree' else ''
search_str = '{}/*dataset={} weak_learner={} model={}*{}*.metrics'.format(
exp_folder, dataset, weak_learner, model, depth_str)
model_names_curr = glob.glob(search_str)
model_names_curr.sort(key=lambda x: os.path.getmtime(x))
if model_names_curr != []:
# model_name_final = model_names_curr[-1]
for model_name_final in model_names_curr:
model_name_final = model_name_final.split('.metrics')[0].split(exp_folder '/')[1]
model_names.append(model_name_final)
return model_names
def get_n_proc(n_ex):
if n_ex > 40000:
n_proc = 50
elif n_ex > 20000:
n_proc = 40
elif n_ex > 2500:
n_proc = 25
elif n_ex > 1000:
n_proc = 10
elif n_ex > 200:
n_proc = 5
else:
n_proc = 1
return n_proc