spatial_stats 0.2.2 → 1.0.4
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/README.md +126 -55
- data/Rakefile +7 -0
- data/ext/spatial_stats/csr_matrix.c +380 -0
- data/ext/spatial_stats/csr_matrix.h +34 -0
- data/ext/spatial_stats/extconf.rb +6 -0
- data/ext/spatial_stats/spatial_stats.c +32 -0
- data/lib/spatial_stats.rb +1 -0
- data/lib/spatial_stats/global/bivariate_moran.rb +60 -22
- data/lib/spatial_stats/global/moran.rb +43 -36
- data/lib/spatial_stats/global/stat.rb +55 -27
- data/lib/spatial_stats/local/bivariate_moran.rb +84 -2
- data/lib/spatial_stats/local/geary.rb +35 -5
- data/lib/spatial_stats/local/getis_ord.rb +45 -17
- data/lib/spatial_stats/local/moran.rb +39 -9
- data/lib/spatial_stats/local/multivariate_geary.rb +45 -22
- data/lib/spatial_stats/local/stat.rb +112 -80
- data/lib/spatial_stats/narray_ext.rb +5 -5
- data/lib/spatial_stats/spatial_stats.so +0 -0
- data/lib/spatial_stats/utils.rb +25 -0
- data/lib/spatial_stats/utils/lag.rb +10 -10
- data/lib/spatial_stats/version.rb +1 -1
- data/lib/spatial_stats/weights/contiguous.rb +20 -10
- data/lib/spatial_stats/weights/distant.rb +38 -20
- data/lib/spatial_stats/weights/weights_matrix.rb +83 -26
- metadata +33 -11
- data/MIT-LICENSE +0 -20
@@ -12,10 +12,25 @@ module SpatialStats
|
|
12
12
|
def initialize(scope, field, weights)
|
13
13
|
@scope = scope
|
14
14
|
@field = field
|
15
|
-
@weights = weights
|
15
|
+
@weights = weights.standardize
|
16
16
|
end
|
17
17
|
attr_accessor :scope, :field, :weights
|
18
18
|
|
19
|
+
##
|
20
|
+
# A new instance of Stat, from vector and weights.
|
21
|
+
#
|
22
|
+
# @param [Array] x observations of dataset
|
23
|
+
# @param [WeightsMatrix] weights to define relationships between observations
|
24
|
+
#
|
25
|
+
# @return [Stat]
|
26
|
+
def self.from_observations(x, weights)
|
27
|
+
raise ArgumentError, 'Data size != weights.n' if x.size != weights.n
|
28
|
+
|
29
|
+
instance = new(nil, nil, weights.standardize)
|
30
|
+
instance.x = x
|
31
|
+
instance
|
32
|
+
end
|
33
|
+
|
19
34
|
def stat
|
20
35
|
raise NotImplementedError, 'method stat not defined'
|
21
36
|
end
|
@@ -28,6 +43,15 @@ module SpatialStats
|
|
28
43
|
raise NotImplementedError, 'method variance not implemented'
|
29
44
|
end
|
30
45
|
|
46
|
+
def x=(values)
|
47
|
+
@x = values.standardize
|
48
|
+
end
|
49
|
+
alias z= x=
|
50
|
+
|
51
|
+
def y=(values)
|
52
|
+
@y = values.standardize
|
53
|
+
end
|
54
|
+
|
31
55
|
##
|
32
56
|
# Z-score for each observation of the statistic.
|
33
57
|
#
|
@@ -42,12 +66,12 @@ module SpatialStats
|
|
42
66
|
|
43
67
|
##
|
44
68
|
# Conditional randomization algorithm used in permutation testing.
|
45
|
-
#
|
46
|
-
#
|
47
|
-
# num_neighbors is the number of neighbors for that observation.
|
69
|
+
# Returns a matrix with permuted index values that will be used for
|
70
|
+
# selecting values from the original data set.
|
48
71
|
#
|
49
|
-
# The
|
50
|
-
#
|
72
|
+
# The width of the matrix is the max number of neighbors + 1
|
73
|
+
# which is way less than it would be if the original vector
|
74
|
+
# was shuffled in full.
|
51
75
|
#
|
52
76
|
# This is super important because most weight matrices are very
|
53
77
|
# sparse so the amount of shuffling/multiplication that is done
|
@@ -55,42 +79,26 @@ module SpatialStats
|
|
55
79
|
#
|
56
80
|
# @see https://github.com/pysal/esda/blob/master/esda/moran.py#L893
|
57
81
|
#
|
58
|
-
# @return [
|
82
|
+
# @return [Numo::Int32] matrix of shape perms x wc_max + 1
|
59
83
|
#
|
60
|
-
def crand(
|
84
|
+
def crand(permutations, rng)
|
61
85
|
# basing this off the ESDA method
|
62
86
|
# need to get k for max_neighbors
|
63
87
|
# and wc for cardinalities of each item
|
64
88
|
# this returns an array of length n with
|
65
|
-
# (permutations x
|
89
|
+
# (permutations x neighbors) Numo Arrays.
|
66
90
|
# This helps reduce computation time because
|
67
91
|
# we are only dealing with neighbors for each
|
68
92
|
# entry not the entire list of permutations for each entry.
|
69
93
|
n_1 = weights.n - 1
|
70
94
|
|
71
95
|
# weight counts
|
72
|
-
wc =
|
73
|
-
k = 0
|
74
|
-
(0..n_1).each do |idx|
|
75
|
-
wc[idx] = (w[idx, true] > 0).count
|
76
|
-
end
|
77
|
-
|
96
|
+
wc = weights.wc
|
78
97
|
k = wc.max + 1
|
79
98
|
prange = (0..permutations - 1).to_a
|
80
99
|
|
81
|
-
arr = Numo::DFloat.cast(arr)
|
82
|
-
|
83
|
-
ids = (0..n_1).to_a
|
84
100
|
ids_perm = (0..n_1 - 1).to_a
|
85
|
-
|
86
|
-
|
87
|
-
(0..n_1).map do |idx|
|
88
|
-
idsi = ids.dup
|
89
|
-
idsi.delete_at(idx)
|
90
|
-
idsi.shuffle!(random: rng)
|
91
|
-
idsi = Numo::Int32.cast(idsi)
|
92
|
-
arr[idsi[rids[true, 0..wc[idx] - 1]]]
|
93
|
-
end
|
101
|
+
Numo::Int32.cast(prange.map { ids_perm.sample(k, random: rng) })
|
94
102
|
end
|
95
103
|
|
96
104
|
##
|
@@ -111,31 +119,40 @@ module SpatialStats
|
|
111
119
|
# its neighbors. Then we will only test for that item instead
|
112
120
|
# of the entire set. This will be done for each item.
|
113
121
|
rng = gen_rng(seed)
|
114
|
-
|
115
|
-
n = weights.n
|
116
|
-
# r is the number of equal to or more extreme samples
|
117
|
-
stat_orig = stat
|
118
|
-
rs = [0] * n
|
119
|
-
|
120
|
-
ws = neighbor_weights
|
122
|
+
rids = crand(permutations, rng)
|
121
123
|
|
122
|
-
|
123
|
-
|
124
|
-
|
124
|
+
n_1 = weights.n - 1
|
125
|
+
sparse = weights.sparse
|
126
|
+
row_index = sparse.row_index
|
127
|
+
ws = sparse.values
|
128
|
+
wc = weights.wc
|
129
|
+
stat_orig = stat
|
125
130
|
|
126
|
-
|
127
|
-
|
131
|
+
arr = Numo::DFloat.cast(x)
|
132
|
+
ids = (0..n_1).to_a
|
133
|
+
observations = Array.new(weights.n)
|
134
|
+
(0..n_1).each do |idx|
|
135
|
+
idsi = ids.dup
|
136
|
+
idsi.delete_at(idx)
|
137
|
+
idsi.shuffle!(random: rng)
|
138
|
+
idsi = Numo::Int32.cast(idsi)
|
139
|
+
sample = arr[idsi[rids[true, 0..wc[idx] - 1]]]
|
128
140
|
|
129
|
-
|
130
|
-
|
131
|
-
|
132
|
-
|
133
|
-
|
141
|
+
# account for case where there are no neighbors
|
142
|
+
row_range = row_index[idx]..(row_index[idx + 1] - 1)
|
143
|
+
if row_range.size.zero?
|
144
|
+
observations[idx] = permutations
|
145
|
+
next
|
146
|
+
end
|
134
147
|
|
135
|
-
|
148
|
+
wi = Numo::DFloat.cast(ws[row_range])
|
149
|
+
stat_i_new = mc_i(wi, sample, idx)
|
150
|
+
stat_i_orig = stat_orig[idx]
|
151
|
+
observations[idx] = mc_observation_calc(stat_i_orig, stat_i_new,
|
152
|
+
permutations)
|
136
153
|
end
|
137
154
|
|
138
|
-
|
155
|
+
observations.map do |ri|
|
139
156
|
(ri + 1.0) / (permutations + 1.0)
|
140
157
|
end
|
141
158
|
end
|
@@ -154,30 +171,40 @@ module SpatialStats
|
|
154
171
|
# @return [Array] of p-values
|
155
172
|
def mc_bv(permutations, seed)
|
156
173
|
rng = gen_rng(seed)
|
157
|
-
|
158
|
-
n = weights.n
|
174
|
+
rids = crand(permutations, rng)
|
159
175
|
|
176
|
+
n_1 = weights.n - 1
|
177
|
+
sparse = weights.sparse
|
178
|
+
row_index = sparse.row_index
|
179
|
+
ws = sparse.values
|
180
|
+
wc = weights.wc
|
160
181
|
stat_orig = stat
|
161
|
-
rs = [0] * n
|
162
|
-
|
163
|
-
ws = neighbor_weights
|
164
182
|
|
165
|
-
|
166
|
-
|
167
|
-
|
168
|
-
|
169
|
-
|
183
|
+
arr = Numo::DFloat.cast(y)
|
184
|
+
ids = (0..n_1).to_a
|
185
|
+
observations = Array.new(weights.n)
|
186
|
+
(0..n_1).each do |idx|
|
187
|
+
idsi = ids.dup
|
188
|
+
idsi.delete_at(idx)
|
189
|
+
idsi.shuffle!(random: rng)
|
190
|
+
idsi = Numo::Int32.cast(idsi)
|
191
|
+
sample = arr[idsi[rids[true, 0..wc[idx] - 1]]]
|
170
192
|
|
171
|
-
|
172
|
-
|
173
|
-
|
174
|
-
|
175
|
-
|
193
|
+
# account for case where there are no neighbors
|
194
|
+
row_range = row_index[idx]..(row_index[idx + 1] - 1)
|
195
|
+
if row_range.size.zero?
|
196
|
+
observations[idx] = permutations
|
197
|
+
next
|
198
|
+
end
|
176
199
|
|
177
|
-
|
200
|
+
wi = Numo::DFloat.cast(ws[row_range])
|
201
|
+
stat_i_new = mc_i(wi, sample, idx)
|
202
|
+
stat_i_orig = stat_orig[idx]
|
203
|
+
observations[idx] = mc_observation_calc(stat_i_orig, stat_i_new,
|
204
|
+
permutations)
|
178
205
|
end
|
179
206
|
|
180
|
-
|
207
|
+
observations.map do |ri|
|
181
208
|
(ri + 1.0) / (permutations + 1.0)
|
182
209
|
end
|
183
210
|
end
|
@@ -200,8 +227,7 @@ module SpatialStats
|
|
200
227
|
# @return [Array] of labels
|
201
228
|
def quads
|
202
229
|
# https://github.com/pysal/esda/blob/master/esda/moran.py#L925
|
203
|
-
|
204
|
-
z_lag = SpatialStats::Utils::Lag.neighbor_average(w, z)
|
230
|
+
z_lag = SpatialStats::Utils::Lag.neighbor_average(weights, z)
|
205
231
|
zp = z.map(&:positive?)
|
206
232
|
lp = z_lag.map(&:positive?)
|
207
233
|
|
@@ -221,6 +247,22 @@ module SpatialStats
|
|
221
247
|
end
|
222
248
|
end
|
223
249
|
|
250
|
+
##
|
251
|
+
# Summary of the statistic. Computes +stat+, +mc+, and +groups+ then returns the values
|
252
|
+
# in a hash array.
|
253
|
+
#
|
254
|
+
# @param [Integer] permutations to run. Last digit should be 9 to produce round numbers.
|
255
|
+
# @param [Integer] seed used in random number generator for shuffles.
|
256
|
+
#
|
257
|
+
# @return [Array]
|
258
|
+
def summary(permutations = 99, seed = nil)
|
259
|
+
p_vals = mc(permutations, seed)
|
260
|
+
data = weights.keys.zip(stat, p_vals, groups)
|
261
|
+
data.map do |row|
|
262
|
+
{ key: row[0], stat: row[1], p: row[2], group: row[3] }
|
263
|
+
end
|
264
|
+
end
|
265
|
+
|
224
266
|
private
|
225
267
|
|
226
268
|
def stat_i
|
@@ -231,8 +273,12 @@ module SpatialStats
|
|
231
273
|
raise NotImplementedError, 'method mc_i not defined'
|
232
274
|
end
|
233
275
|
|
276
|
+
def mc_observation_calc(_stat_i_orig, _stat_i_new, _permutations)
|
277
|
+
raise NotImplementedError, 'method mc_observation_calc not defined'
|
278
|
+
end
|
279
|
+
|
234
280
|
def w
|
235
|
-
weights.
|
281
|
+
@w ||= weights.dense
|
236
282
|
end
|
237
283
|
|
238
284
|
def gen_rng(seed = nil)
|
@@ -242,20 +288,6 @@ module SpatialStats
|
|
242
288
|
Random.new
|
243
289
|
end
|
244
290
|
end
|
245
|
-
|
246
|
-
def neighbor_weights
|
247
|
-
# record the non-zero weights in variable length arrays for each
|
248
|
-
# row in the weights table
|
249
|
-
ws = [[]] * weights.n
|
250
|
-
(0..weights.n - 1).each do |idx|
|
251
|
-
neighbors = []
|
252
|
-
w[idx, true].each do |wij|
|
253
|
-
neighbors << wij if wij != 0
|
254
|
-
end
|
255
|
-
ws[idx] = neighbors
|
256
|
-
end
|
257
|
-
ws
|
258
|
-
end
|
259
291
|
end
|
260
292
|
end
|
261
293
|
end
|
@@ -12,11 +12,11 @@ module Numo
|
|
12
12
|
#
|
13
13
|
# @ example
|
14
14
|
#
|
15
|
-
# Numo::DFloat [[0, 1, 1], [1, 1, 1]].
|
15
|
+
# Numo::DFloat [[0, 1, 1], [1, 1, 1]].row_standardize
|
16
16
|
# Numo::DFloat [[0, 0.5, 0.5], [0.33333, 0.33333, 0.33333]]
|
17
17
|
#
|
18
18
|
# @return [Numo::NArray]
|
19
|
-
def
|
19
|
+
def row_standardize
|
20
20
|
# every row will sum up to 1, or if they are all 0, do nothing
|
21
21
|
standardized = each_over_axis.map do |row|
|
22
22
|
sum = row.sum
|
@@ -38,16 +38,16 @@ module Numo
|
|
38
38
|
#
|
39
39
|
# @ example
|
40
40
|
#
|
41
|
-
# Numo::DFloat [[0, 1, 0], [1, 0, 1], [0, 1, 0]].
|
41
|
+
# Numo::DFloat [[0, 1, 0], [1, 0, 1], [0, 1, 0]].window
|
42
42
|
# Numo::DFloat [[1, 1, 0], [1, 1, 1], [0, 1, 1]]
|
43
43
|
#
|
44
44
|
# @ example
|
45
45
|
# # Input will be equivalent to output in this case
|
46
|
-
# Numo::DFloat [[1, 1, 0], [1, 0, 1], [0, 1, 0]].
|
46
|
+
# Numo::DFloat [[1, 1, 0], [1, 0, 1], [0, 1, 0]].window
|
47
47
|
# Numo::DFloat [[1, 1, 0], [1, 0, 1], [0, 1, 0]]
|
48
48
|
#
|
49
49
|
# @return [Numo::NArray]
|
50
|
-
def
|
50
|
+
def window
|
51
51
|
# in windowed calculations, the diagonal is set to 1
|
52
52
|
# if trace (sum of diag) is 0, add it, else return input
|
53
53
|
if trace.zero?
|
Binary file
|
data/lib/spatial_stats/utils.rb
CHANGED
@@ -6,5 +6,30 @@ module SpatialStats
|
|
6
6
|
##
|
7
7
|
# The Utils module contains various utilities used in the gem.
|
8
8
|
module Utils
|
9
|
+
##
|
10
|
+
# Compute the false discovery rate (FDR) of a set of p-values given
|
11
|
+
# an alpha value.
|
12
|
+
#
|
13
|
+
# If there is no FDR available in the dataset, the Bonferroni Bound is
|
14
|
+
# returned instead.
|
15
|
+
#
|
16
|
+
# @param [Array] pvals from an mc test
|
17
|
+
# @param [Float] alpha value for the fdr
|
18
|
+
#
|
19
|
+
# @returns [Float] either the FDR or Bonferroni Bound
|
20
|
+
def self.fdr(pvals, alpha)
|
21
|
+
n = pvals.size
|
22
|
+
b_bound = alpha / n
|
23
|
+
pvals.sort!
|
24
|
+
|
25
|
+
p_val = b_bound
|
26
|
+
(0..n - 1).each do |i|
|
27
|
+
p_fdr = (i + 1) * b_bound
|
28
|
+
break unless pvals[i] <= p_fdr
|
29
|
+
|
30
|
+
p_val = p_fdr
|
31
|
+
end
|
32
|
+
p_val
|
33
|
+
end
|
9
34
|
end
|
10
35
|
end
|
@@ -11,36 +11,36 @@ module SpatialStats
|
|
11
11
|
# Dot product of the row_standardized input matrix
|
12
12
|
# by the input vector, variables.
|
13
13
|
#
|
14
|
-
# @param [
|
14
|
+
# @param [WeightsMatrix] matrix holding target weights.
|
15
15
|
# @param [Array] variables vector multiplying the matrix
|
16
16
|
#
|
17
17
|
# @return [Array] resultant vector
|
18
18
|
def self.neighbor_average(matrix, variables)
|
19
|
-
matrix = matrix.
|
19
|
+
matrix = matrix.standardize
|
20
20
|
neighbor_sum(matrix, variables)
|
21
21
|
end
|
22
22
|
|
23
23
|
##
|
24
24
|
# Dot product of the input matrix by the input vector, variables.
|
25
25
|
#
|
26
|
-
# @param [
|
26
|
+
# @param [WeightsMatrix] matrix holding target weights.
|
27
27
|
# @param [Array] variables vector multiplying the matrix
|
28
28
|
#
|
29
29
|
# @return [Array] resultant vector
|
30
30
|
def self.neighbor_sum(matrix, variables)
|
31
|
-
matrix.
|
31
|
+
matrix.sparse.mulvec(variables)
|
32
32
|
end
|
33
33
|
|
34
34
|
##
|
35
|
-
# Dot product of the input windowed, row
|
35
|
+
# Dot product of the input windowed, row standardized matrix by
|
36
36
|
# the input vector, variables.
|
37
37
|
#
|
38
|
-
# @param [
|
38
|
+
# @param [WeightsMatrix] matrix holding target weights.
|
39
39
|
# @param [Array] variables vector multiplying the matrix
|
40
40
|
#
|
41
41
|
# @return [Array] resultant vector
|
42
42
|
def self.window_average(matrix, variables)
|
43
|
-
matrix = matrix.
|
43
|
+
matrix = matrix.window.standardize
|
44
44
|
window_sum(matrix, variables)
|
45
45
|
end
|
46
46
|
|
@@ -48,13 +48,13 @@ module SpatialStats
|
|
48
48
|
# Dot product of the input windowed matrix by
|
49
49
|
# the input vector, variables.
|
50
50
|
#
|
51
|
-
# @param [
|
51
|
+
# @param [WeightsMatrix] matrix holding target weights.
|
52
52
|
# @param [Array] variables vector multiplying the matrix
|
53
53
|
#
|
54
54
|
# @return [Array] resultant vector
|
55
55
|
def self.window_sum(matrix, variables)
|
56
|
-
matrix = matrix.
|
57
|
-
matrix.
|
56
|
+
matrix = matrix.window
|
57
|
+
matrix.sparse.mulvec(variables)
|
58
58
|
end
|
59
59
|
end
|
60
60
|
end
|
@@ -15,21 +15,26 @@ module SpatialStats
|
|
15
15
|
#
|
16
16
|
# @return [WeightsMatrix]
|
17
17
|
def self.rook(scope, field)
|
18
|
-
p_key = scope.primary_key
|
19
|
-
keys = scope.pluck(p_key).sort
|
20
|
-
|
21
18
|
neighbors = SpatialStats::Queries::Weights
|
22
19
|
.rook_contiguity_neighbors(scope, field)
|
23
20
|
|
21
|
+
# get keys to make sure we have consistent dimensions when
|
22
|
+
# some entries don't have neighbors.
|
23
|
+
# define a new hash that has all the keys from scope
|
24
|
+
keys = SpatialStats::Queries::Variables.query_field(scope, scope.klass.primary_key)
|
25
|
+
|
24
26
|
neighbors = neighbors.group_by(&:i_id)
|
27
|
+
missing_neighbors = Hash[(keys - neighbors.keys).map { |key| [key, []] }]
|
28
|
+
neighbors = neighbors.merge(missing_neighbors)
|
29
|
+
|
25
30
|
weights = neighbors.transform_values do |value|
|
26
31
|
value.map do |neighbor|
|
27
|
-
hash =
|
32
|
+
hash = { id: neighbor[:j_id] }
|
28
33
|
hash[:weight] = 1
|
29
34
|
hash
|
30
35
|
end
|
31
36
|
end
|
32
|
-
SpatialStats::Weights::WeightsMatrix.new(
|
37
|
+
SpatialStats::Weights::WeightsMatrix.new(weights)
|
33
38
|
end
|
34
39
|
|
35
40
|
##
|
@@ -40,21 +45,26 @@ module SpatialStats
|
|
40
45
|
#
|
41
46
|
# @return [WeightsMatrix]
|
42
47
|
def self.queen(scope, field)
|
43
|
-
p_key = scope.primary_key
|
44
|
-
keys = scope.pluck(p_key).sort
|
45
|
-
|
46
48
|
neighbors = SpatialStats::Queries::Weights
|
47
49
|
.queen_contiguity_neighbors(scope, field)
|
48
50
|
|
51
|
+
# get keys to make sure we have consistent dimensions when
|
52
|
+
# some entries don't have neighbors.
|
53
|
+
# define a new hash that has all the keys from scope
|
54
|
+
keys = SpatialStats::Queries::Variables.query_field(scope, scope.klass.primary_key)
|
55
|
+
|
49
56
|
neighbors = neighbors.group_by(&:i_id)
|
57
|
+
missing_neighbors = Hash[(keys - neighbors.keys).map { |key| [key, []] }]
|
58
|
+
neighbors = neighbors.merge(missing_neighbors)
|
59
|
+
|
50
60
|
weights = neighbors.transform_values do |value|
|
51
61
|
value.map do |neighbor|
|
52
|
-
hash =
|
62
|
+
hash = { id: neighbor[:j_id] }
|
53
63
|
hash[:weight] = 1
|
54
64
|
hash
|
55
65
|
end
|
56
66
|
end
|
57
|
-
SpatialStats::Weights::WeightsMatrix.new(
|
67
|
+
SpatialStats::Weights::WeightsMatrix.new(weights)
|
58
68
|
end
|
59
69
|
end
|
60
70
|
end
|