{"id":2501,"date":"2026-02-02T09:23:12","date_gmt":"2026-02-02T09:23:12","guid":{"rendered":"https:\/\/demo.materiamedica.net\/demo6\/?p=2501"},"modified":"2026-02-02T09:23:51","modified_gmt":"2026-02-02T09:23:51","slug":"chapter-2-random-permutations","status":"publish","type":"post","link":"https:\/\/demo.materiamedica.net\/demo6\/chapter-2-random-permutations\/","title":{"rendered":"Chapter 3: Random Permutations"},"content":{"rendered":"What is a permutation? (quick honest explanation)<\/h3>\n
A permutation<\/strong> is simply a rearrangement<\/strong> of the elements of a sequence.<\/p>\nExamples:<\/p>\n
\n- Original: [A, B, C, D]<\/li>\n
- One permutation: [B, D, A, C]<\/li>\n
- Another: [D, A, C, B]<\/li>\n
- etc.<\/li>\n<\/ul>\n
There are n!<\/strong> possible permutations of n distinct items.<\/p>\nNumPy gives us two main ways to create random permutations:<\/p>\n
\n
\n
\n\n\n| Method<\/th>\n | What it does<\/th>\n | Modifies original?<\/th>\n | Returns<\/th>\n<\/tr>\n<\/thead>\n |
\n\n| np.random.permutation()<\/td>\n | Returns a new<\/strong> randomly shuffled copy<\/td>\nNo<\/strong><\/td>\n| new array<\/td>\n<\/tr>\n | \n| np.random.shuffle()<\/td>\n | Shuffles in place<\/strong> (modifies original)<\/td>\nYes<\/strong><\/td>\nNone<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<\/div>\n<\/div>\n 1. np.random.permutation() \u2014 most commonly used<\/h3>\nCreates a new shuffled copy<\/strong> \u2014 original stays unchanged.<\/p>\n\n \n \n Python<\/div>\n \n # Simple 1D example\r\nnumbers = np.arange(10) # [0 1 2 3 4 5 6 7 8 9]\r\n\r\nshuffled = np.random.permutation(numbers)\r\nprint(\"Original:\", numbers)\r\nprint(\"Shuffled :\", shuffled)<\/code><\/pre>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\nImportant observation<\/strong> \u2014 every time you run it \u2192 different order (unless seeded)<\/p>\n\n \n \n Python<\/div>\n \n np.random.seed(42)\r\nprint(np.random.permutation(numbers))\r\n# [6 3 7 2 9 1 8 4 0 5]<\/code><\/pre>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\nVery common pattern \u2014 permuting indices<\/strong><\/p>\n\n \n \n Python<\/div>\n \n # 1000 samples \u2014 we want to shuffle the order randomly\r\nindices = np.arange(1000)\r\n\r\nrandom_order = np.random.permutation(indices)\r\n\r\n# Now we can use this to shuffle data\r\nX = np.random.randn(1000, 20) # features\r\ny = np.random.randint(0, 2, 1000) # labels\r\n\r\nX_shuffled = X[random_order]\r\ny_shuffled = y[random_order]<\/code><\/pre>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\nThis is exactly what train_test_split does behind the scenes.<\/strong><\/p>\n2. np.random.shuffle() \u2014 shuffles in place<\/h3>\nModifies the array directly<\/strong> \u2014 does not<\/strong> return anything.<\/p>\n\n \n \n Python<\/div>\n \n deck = np.arange(1, 53) # cards 1 to 52\r\nprint(\"Before:\", deck[:10])\r\n\r\nnp.random.shuffle(deck)\r\n\r\nprint(\"After :\", deck[:10])<\/code><\/pre>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\nVery common mistake students make:<\/strong><\/p>\n\n \n \n Python<\/div>\n \n # WRONG \u2014 shuffle returns None!\r\nwrong = np.random.shuffle(deck) # wrong = None\r\nprint(wrong) # None<\/code><\/pre>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\nCorrect usage:<\/strong><\/p>\n\n \n \n Python<\/div>\n \n np.random.shuffle(deck) # modifies deck directly<\/code><\/pre>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n3. Quick comparison table (very useful to remember)<\/h3>\n\n \n \n\n\n| Property<\/th>\n | np.random.permutation()<\/th>\n | np.random.shuffle()<\/th>\n<\/tr>\n<\/thead>\n | \n\n| Returns<\/td>\n | new shuffled array<\/td>\n | None<\/td>\n<\/tr>\n | \n| Modifies original?<\/td>\n | No<\/td>\n | Yes<\/strong><\/td>\n<\/tr>\n\n| Can take integer N?<\/td>\n | Yes<\/strong> \u2014 creates 0..N-1 shuffled<\/td>\nNo<\/strong><\/td>\n<\/tr>\n\n| Memory usage<\/td>\n | creates copy<\/td>\n | no extra memory<\/td>\n<\/tr>\n | \n| Most common use case<\/td>\n | creating shuffled indices, new copy<\/td>\n | shuffling existing dataset in place<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<\/div>\n<\/div>\n 4. Special & very useful feature of permutation()<\/h3>\nYou can pass an integer<\/strong> instead of an array!<\/p>\n\n \n \n Python<\/div>\n \n # Create a random permutation of 0, 1, 2, ..., 999\r\nidx = np.random.permutation(1000)\r\n\r\n# Same as:\r\nidx = np.arange(1000)\r\nnp.random.shuffle(idx) # but this modifies in place<\/code><\/pre>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\nThis is extremely common<\/strong> when you want to shuffle indices without creating the full arange first.<\/p>\n5. Realistic patterns you will use again and again<\/h3>\nPattern 1 \u2013 Shuffle dataset before training<\/strong><\/p>\n\n \n \n Python<\/div>\n \n # Full dataset\r\nX = np.random.randn(15000, 35)\r\ny = np.random.randint(0, 3, 15000)\r\n\r\n# Shuffle\r\nperm = np.random.permutation(len(X))\r\n\r\nX = X[perm]\r\ny = y[perm]<\/code><\/pre>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\nPattern 2 \u2013 Create k-fold cross-validation indices<\/strong><\/p>\n\n \n \n Python<\/div>\n \n n = 12000\r\nindices = np.random.permutation(n)\r\n\r\nfold_size = n \/\/ 5\r\n\r\nfor i in range(5):\r\n val_start = i * fold_size\r\n val_end = (i+1) * fold_size\r\n \r\n val_idx = indices[val_start:val_end]\r\n train_idx = np.concatenate([indices[:val_start], indices[val_end:]])\r\n \r\n print(f\"Fold {i+1}: train={len(train_idx)}, val={len(val_idx)}\")<\/code><\/pre>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\nPattern 3 \u2013 Random sampling without replacement<\/strong><\/p>\n\n \n \n Python<\/div>\n \n all_customers = np.arange(50000)\r\nselected = np.random.permutation(all_customers)[:500] # first 500 after shuffle<\/code><\/pre>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\nPattern 4 \u2013 Shuffle rows of a matrix<\/strong><\/p>\n\n \n \n Python<\/div>\n \n matrix = np.random.randint(0, 100, size=(1000, 6))\r\nnp.random.shuffle(matrix) # rows are shuffled in place\r\n# or\r\nmatrix = matrix[np.random.permutation(len(matrix))]<\/code><\/pre>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\nSummary \u2013 Quick Decision Guide<\/h3>\n\n \n \n\n\n| You want to…<\/th>\n | Best choice<\/th>\n<\/tr>\n<\/thead>\n | \n\n| Get a new shuffled version (keep original)<\/td>\n | np.random.permutation(arr)<\/td>\n<\/tr>\n | \n| Shuffle existing array in place (save memory)<\/td>\n | np.random.shuffle(arr)<\/td>\n<\/tr>\n | \n| Create random order of 0..n-1<\/td>\n | np.random.permutation(n)<\/td>\n<\/tr>\n | \n| Shuffle rows of a 2D array<\/td>\n | either \u2014 shuffle() or permutation + indexing<\/td>\n<\/tr>\n | \n| Need shuffled indices for splitting<\/td>\n | np.random.permutation(len(data))<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<\/div>\n<\/div>\n Final teacher advice<\/h3>\nAlways<\/strong> think about whether you need the original order preserved:<\/p>\n\n- Need original later \u2192 use permutation()<\/li>\n
- Don\u2019t need original + want to save memory \u2192 use shuffle()<\/li>\n
- Working with indices \u2192 permutation(n) is usually cleanest<\/li>\n<\/ul>\n
Always<\/strong> set a seed when you want reproducible shuffling:<\/p>\n\n \n \n Python<\/div>\n \n np.random.seed(42)\r\nperm = np.random.permutation(1000)<\/code><\/pre>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\nWhere would you like to go next?<\/p>\n \n- Difference between permutation and choice (sampling)<\/li>\n
- Shuffling multi-dimensional arrays correctly<\/li>\n
- Random permutations in machine learning pipelines<\/li>\n
- Common bugs when shuffling labels\/data separately<\/li>\n
- Mini-exercise: shuffle a dataset and create train\/val splits<\/li>\n<\/ul>\n
Just tell me what feels most useful or interesting right now! \ud83d\ude0a<\/p>\n","protected":false},"excerpt":{"rendered":" What is a permutation? (quick honest explanation) A permutation is simply a rearrangement of the elements of a sequence. Examples: Original: [A, B, C, D] One permutation: [B, D, A, C] Another: [D, A,...<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[75],"tags":[],"class_list":["post-2501","post","type-post","status-publish","format-standard","hentry","category-numpy"],"_links":{"self":[{"href":"https:\/\/demo.materiamedica.net\/demo6\/wp-json\/wp\/v2\/posts\/2501","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/demo.materiamedica.net\/demo6\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/demo.materiamedica.net\/demo6\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/demo.materiamedica.net\/demo6\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/demo.materiamedica.net\/demo6\/wp-json\/wp\/v2\/comments?post=2501"}],"version-history":[{"count":2,"href":"https:\/\/demo.materiamedica.net\/demo6\/wp-json\/wp\/v2\/posts\/2501\/revisions"}],"predecessor-version":[{"id":2503,"href":"https:\/\/demo.materiamedica.net\/demo6\/wp-json\/wp\/v2\/posts\/2501\/revisions\/2503"}],"wp:attachment":[{"href":"https:\/\/demo.materiamedica.net\/demo6\/wp-json\/wp\/v2\/media?parent=2501"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/demo.materiamedica.net\/demo6\/wp-json\/wp\/v2\/categories?post=2501"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/demo.materiamedica.net\/demo6\/wp-json\/wp\/v2\/tags?post=2501"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
|
| | | |
| | | | |