1
|
|
|
# -*- coding: utf-8 -*- |
2
|
|
|
# emacs: -*- mode: python; py-indent-offset: 4; indent-tabs-mode: nil -*- |
3
|
|
|
# vi: set ft=python sts=4 ts=4 sw=4 et: |
4
|
|
|
""" |
5
|
|
|
Utilities to make crumbs |
6
|
|
|
""" |
7
|
|
|
import os |
8
|
|
|
import os.path as op |
9
|
|
|
import operator |
10
|
|
|
import fnmatch |
11
|
|
|
|
12
|
|
|
from collections import Mapping |
13
|
|
|
from functools import partial, reduce |
14
|
|
|
from itertools import product |
15
|
|
|
|
16
|
|
|
|
17
|
|
|
def remove_duplicates(lst): |
18
|
|
|
""" Return a sorted lst of non-duplicated elements from `lst`. |
19
|
|
|
|
20
|
|
|
Parameters |
21
|
|
|
---------- |
22
|
|
|
lst: sequence of any |
23
|
|
|
|
24
|
|
|
Returns |
25
|
|
|
------- |
26
|
|
|
fslst: |
27
|
|
|
Filtered and sorted `lst` with non duplicated elements of `lst`. |
28
|
|
|
""" |
29
|
|
|
return sorted(list(set(lst))) |
30
|
|
|
|
31
|
|
|
|
32
|
|
|
def remove_ignored(strs, ignore): |
33
|
|
|
""" Remove from `strs` the matches to the `fnmatch` (glob) patterns and |
34
|
|
|
return the result in a list.""" |
35
|
|
|
nustrs = strs.copy() |
36
|
|
|
for ign in ignore: |
37
|
|
|
nustrs = [item for item in nustrs if not fnmatch.fnmatch(item, ign)] |
38
|
|
|
|
39
|
|
|
return nustrs |
40
|
|
|
|
41
|
|
|
|
42
|
|
|
def list_children(path, just_dirs=False, ignore=[]): |
43
|
|
|
""" Return the immediate elements (files and folders) in `path`. |
44
|
|
|
|
45
|
|
|
Parameters |
46
|
|
|
---------- |
47
|
|
|
path: str |
48
|
|
|
|
49
|
|
|
just_dirs: bool |
50
|
|
|
If True will return only folders. |
51
|
|
|
|
52
|
|
|
ignore: sequence of str |
53
|
|
|
Sequence of glob patterns to ignore from the listing. |
54
|
|
|
|
55
|
|
|
Returns |
56
|
|
|
------- |
57
|
|
|
paths: list of str |
58
|
|
|
""" |
59
|
|
|
if not op.exists(path): |
60
|
|
|
raise IOError("Expected an existing path, but could not" |
61
|
|
|
" find {}.".format(path)) |
62
|
|
|
|
63
|
|
|
if op.isfile(path): |
64
|
|
|
if just_dirs: |
65
|
|
|
vals = [] |
66
|
|
|
else: |
67
|
|
|
vals = [path] |
68
|
|
|
else: |
69
|
|
|
if just_dirs: # this means we have to list only folders |
70
|
|
|
vals = [d for d in os.listdir(path) if op.isdir(op.join(path, d))] |
71
|
|
|
else: # this means we have to list files |
72
|
|
|
vals = os.listdir(path) |
73
|
|
|
|
74
|
|
|
if ignore: |
75
|
|
|
vals = remove_ignored(vals, ignore) |
76
|
|
|
|
77
|
|
|
return vals |
78
|
|
|
|
79
|
|
|
|
80
|
|
|
class ParameterGrid(object): |
81
|
|
|
""" |
82
|
|
|
Picked from sklearn: https://github.com/scikit-learn/scikit-learn |
83
|
|
|
|
84
|
|
|
Grid of parameters with a discrete number of values for each. |
85
|
|
|
Can be used to iterate over parameter value combinations with the |
86
|
|
|
Python built-in function iter. |
87
|
|
|
|
88
|
|
|
Read more in the :ref:`User Guide <grid_search>`. |
89
|
|
|
Parameters |
90
|
|
|
---------- |
91
|
|
|
param_grid : dict of string to sequence, or sequence of such |
92
|
|
|
The parameter grid to explore, as a dictionary mapping estimator |
93
|
|
|
parameters to sequences of allowed values. |
94
|
|
|
An empty dict signifies default parameters. |
95
|
|
|
A sequence of dicts signifies a sequence of grids to search, and is |
96
|
|
|
useful to avoid exploring parameter combinations that make no sense |
97
|
|
|
or have no effect. See the examples below. |
98
|
|
|
Examples |
99
|
|
|
-------- |
100
|
|
|
>>> from sklearn.grid_search import ParameterGrid |
101
|
|
|
>>> param_grid = {'a': [1, 2], 'b': [True, False]} |
102
|
|
|
>>> list(ParameterGrid(param_grid)) == ( |
103
|
|
|
... [{'a': 1, 'b': True}, {'a': 1, 'b': False}, |
104
|
|
|
... {'a': 2, 'b': True}, {'a': 2, 'b': False}]) |
105
|
|
|
True |
106
|
|
|
>>> grid = [{'kernel': ['linear']}, {'kernel': ['rbf'], 'gamma': [1, 10]}] |
107
|
|
|
>>> list(ParameterGrid(grid)) == [{'kernel': 'linear'}, |
108
|
|
|
... {'kernel': 'rbf', 'gamma': 1}, |
109
|
|
|
... {'kernel': 'rbf', 'gamma': 10}] |
110
|
|
|
True |
111
|
|
|
>>> ParameterGrid(grid)[1] == {'kernel': 'rbf', 'gamma': 1} |
112
|
|
|
True |
113
|
|
|
""" |
114
|
|
|
|
115
|
|
|
def __init__(self, param_grid): |
116
|
|
|
if isinstance(param_grid, Mapping): |
117
|
|
|
# wrap dictionary in a singleton list to support either dict |
118
|
|
|
# or list of dicts |
119
|
|
|
param_grid = [param_grid] |
120
|
|
|
self.param_grid = param_grid |
121
|
|
|
|
122
|
|
|
def __iter__(self): |
123
|
|
|
"""Iterate over the points in the grid. |
124
|
|
|
Returns |
125
|
|
|
------- |
126
|
|
|
params : iterator over dict of string to any |
127
|
|
|
Yields dictionaries mapping each estimator parameter to one of its |
128
|
|
|
allowed values. |
129
|
|
|
""" |
130
|
|
|
for p in self.param_grid: |
131
|
|
|
# Always sort the keys of a dictionary, for reproducibility |
132
|
|
|
items = sorted(p.items()) |
133
|
|
|
if not items: |
134
|
|
|
yield {} |
135
|
|
|
else: |
136
|
|
|
keys, values = zip(*items) |
137
|
|
|
for v in product(*values): |
138
|
|
|
params = dict(zip(keys, v)) |
139
|
|
|
yield params |
140
|
|
|
|
141
|
|
|
def __len__(self): |
142
|
|
|
"""Number of points on the grid.""" |
143
|
|
|
# Product function that can handle iterables (np.product can't). |
144
|
|
|
product = partial(reduce, operator.mul) |
145
|
|
|
return sum(product(len(v) for v in p.values()) if p else 1 |
146
|
|
|
for p in self.param_grid) |
147
|
|
|
|
148
|
|
|
def __getitem__(self, ind): |
149
|
|
|
"""Get the parameters that would be ``ind``th in iteration |
150
|
|
|
Parameters |
151
|
|
|
---------- |
152
|
|
|
ind : int |
153
|
|
|
The iteration index |
154
|
|
|
Returns |
155
|
|
|
------- |
156
|
|
|
params : dict of string to any |
157
|
|
|
Equal to list(self)[ind] |
158
|
|
|
""" |
159
|
|
|
# This is used to make discrete sampling without replacement memory |
160
|
|
|
# efficient. |
161
|
|
|
for sub_grid in self.param_grid: |
162
|
|
|
# XXX: could memoize information used here |
163
|
|
|
if not sub_grid: |
164
|
|
|
if ind == 0: |
165
|
|
|
return {} |
166
|
|
|
else: |
167
|
|
|
ind -= 1 |
168
|
|
|
continue |
169
|
|
|
|
170
|
|
|
# Reverse so most frequent cycling parameter comes first |
171
|
|
|
keys, values_lists = zip(*sorted(sub_grid.items())[::-1]) |
172
|
|
|
sizes = [len(v_list) for v_list in values_lists] |
173
|
|
|
product = partial(reduce, operator.mul) |
174
|
|
|
total = product(sizes) |
175
|
|
|
|
176
|
|
|
if ind >= total: |
177
|
|
|
# Try the next grid |
178
|
|
|
ind -= total |
179
|
|
|
else: |
180
|
|
|
out = {} |
181
|
|
|
for key, v_list, n in zip(keys, values_lists, sizes): |
182
|
|
|
ind, offset = divmod(ind, n) |
183
|
|
|
out[key] = v_list[offset] |
184
|
|
|
return out |
185
|
|
|
|
186
|
|
|
raise IndexError('ParameterGrid index out of range') |
187
|
|
|
|