Random allocation

.github/workflows/coverage.yml

@@ -11,7 +11,7 @@ jobs:
     runs-on: Ubuntu-latest
     strategy:
       matrix:
-        python-version: [3.9]
+        python-version: [3.12]
     steps:
     - uses: actions/checkout@v2
 

.github/workflows/test.yml

@@ -12,7 +12,7 @@ jobs:
     runs-on: Ubuntu-latest
     strategy:
       matrix:
-        python-version: [3.7, 3.8, 3.9]
+        python-version: [3.11, 3.12]
     steps:
     - uses: actions/checkout@v2
     - name: Set up Python ${{ matrix.python-version }}
@@ -41,7 +41,7 @@ jobs:
     runs-on: macOS-latest
     strategy:
       matrix:
-        python-version: [3.7, 3.8, 3.9]
+        python-version: [3.11, 3.12]
     steps:
     - uses: actions/checkout@v2
     - name: Set up Python ${{ matrix.python-version }}
@@ -70,7 +70,7 @@ jobs:
     runs-on: windows-latest
     strategy:
       matrix:
-        python-version: [3.7, 3.8, 3.9]
+        python-version: [3.11, 3.12]
     steps:
     - uses: actions/checkout@v2
     - name: Set up Python ${{ matrix.python-version }}

cryptorandom/sample.py

@@ -179,20 +179,20 @@ def random_allocation(a, sizes, replace=False, fast=False, p=None, method="sampl
         raise ValueError('sample sizes greater than population size')
 
     samples = [0] * len(sizes)
-    # sort sizes from smallest to largest
-    sizes.sort()
-    # get random samples for all the groups except the largest one
-    for i in range(len(sizes) - 1):
+    # get random samples for all the groups except the largest one. Work from smallest to largest for numerical stability
+    size_increasing_inx = np.argsort(sizes)
+    biggest = size_increasing_inx[-1]
+    for i in size_increasing_inx[:-1]:
         sam = random_sample(list(indices), sizes[i], replace, fast, p, method, prng)
         samples[i] = a[sam]
         if not replace:
-            indices = set(indices) - set(sam)
+            indices = list(set(indices) - set(sam))
     # get the sample for the largest group
-    if not replace and N == np.sum(sizes):
+    if (not replace) and N == np.sum(sizes) and fast:
         sam = list(indices)
     else:
-        sam = random_sample(list(indices), sizes[-1], replace, fast, p, method, prng)
-    samples[-1] = a[sam]
+        sam = random_sample(list(indices), sizes[biggest], replace, fast, p, method, prng)
+    samples[biggest] = a[sam]
 
     return samples
 

cryptorandom/tests/test_sample.py

@@ -313,19 +313,21 @@ def test_permute_by_index():
 
 def test_random_allocation():
     # test random allocation without replacement
-    samples = random_allocation(10, [5, 5], replace = False)
-    assert all(x in np.arange(10) for x in samples[0])
-    assert all(x in np.arange(10) for x in samples[1])
-    assert np.sum(samples) == np.sum(np.arange(10))
+    samples = random_allocation(15, [6, 4, 5], replace = False)
+    assert len(samples[0]) == 6
+    assert len(samples[1]) == 4
+    assert len(samples[2]) == 5
+    assert all(x in np.arange(15) for x in samples[0])
+    assert all(x in np.arange(15) for x in samples[1])
+    assert all(x in np.arange(15) for x in samples[2])
+    assert set(samples[0]) | set(samples[1]) | set(samples[2]) == set(np.arange(15))
 
     # test with replacement
     a = [1, 2, 2, 3, 3]
     sizes = [2, 3]
     samples = random_allocation(a, sizes, replace = True)
     assert all(x in a for x in samples[0])
     assert all(x in a for x in samples[1])
-    # test that smallest sample is sample first
-    assert len(samples[0]) == np.min(sizes) 
 
     # test when sum of sample sizes is less than population size
     a = [1, 2, 2, 3, 3]

requirements/default.txt

@@ -1,2 +1,2 @@
-numpy>=1.20
-scipy>=1.6
+numpy>=2.0
+scipy>=1.12

requirements/doc.txt

@@ -1,2 +1,2 @@
-Sphinx~=4.0
+Sphinx>=5.0
 numpydoc>=1.1