Add hmm.py

custom-packages/acoustic_keylogger/hmm.py

@@ -0,0 +1,161 @@
+from hmmlearn import hmm
+import numpy as np
+
+import os
+
+
+def read_encode():
+    path = os.environ['ENCODE_PATH']
+    with open(path, 'r') as f:
+        encode = f.readlines()
+    return np.array([int(v) for v in encode[0].split()])
+
+
+def create_transmat(corpus, keys='abcdefghijklmnopqrstuvwxyz .,'):
+    """Return a transition matrix based on the sequence of letters in corpus.
+
+    Assume that each word in the corpus is delimited by a space.
+
+    Args:
+    corpus -- an iterable containing words as elements
+    keys   -- all letters that are accounted for in the transition matrix
+    """
+    mat = np.zeros((len(keys), len(keys)), dtype=np.int8)
+    key_id_map = id_map(keys)
+
+    # TODO
+
+    return mat
+
+
+# Tests
+
+def test_create_transmat():
+        corpora = [
+        ['This', 'is', 'a', 'sentence'],
+        ['contains', '``', 'unrecognized', "''", 'characters'],
+        [''],
+    ]
+
+    keys = 'abcdefghijklmnopqrstuvwxyz .,'
+    key_map = id_map(keys)
+    reverse_map = dict(enumerate(keys))
+    base_mat = np.zeros((len(keys), len(keys)), dtype=int)
+
+    transmats = [
+        base_mat.copy(),
+        base_mat.copy(),
+        base_mat.copy(),
+    ]
+
+    # Create correct transition matrix for corpora[0]
+    transmats[0][key_map['t']][key_map['h']] = 1
+    transmats[0][key_map['h']][key_map['i']] = 1
+    transmats[0][key_map['i']][key_map['s']] = 2
+    transmats[0][key_map['s']][key_map[' ']] = 2
+    transmats[0][key_map[' ']][key_map['i']] = 1
+    transmats[0][key_map[' ']][key_map['a']] = 1
+    transmats[0][key_map['a']][key_map[' ']] = 1
+    transmats[0][key_map[' ']][key_map['s']] = 1
+    transmats[0][key_map['s']][key_map['e']] = 1
+    transmats[0][key_map['e']][key_map['n']] = 2
+    transmats[0][key_map['n']][key_map['t']] = 1
+    transmats[0][key_map['t']][key_map['e']] = 1
+    transmats[0][key_map['n']][key_map['c']] = 1
+    transmats[0][key_map['c']][key_map['e']] = 1
+
+    # Create correct transition matrix for corpora[1]
+    transmats[1][key_map['c']][key_map['o']] = 2
+    transmats[1][key_map['o']][key_map['n']] = 1
+    transmats[1][key_map['n']][key_map['t']] = 1
+    transmats[1][key_map['t']][key_map['a']] = 1
+    transmats[1][key_map['a']][key_map['i']] = 1
+    transmats[1][key_map['n']][key_map['s']] = 1
+    transmats[1][key_map['s']][key_map[' ']] = 1
+    transmats[1][key_map[' ']][key_map['u']] = 1
+    transmats[1][key_map['u']][key_map['n']] = 1
+    transmats[1][key_map['n']][key_map['r']] = 1
+    transmats[1][key_map['r']][key_map['e']] = 1
+    transmats[1][key_map['e']][key_map['c']] = 1
+    transmats[1][key_map['o']][key_map['g']] = 1
+    transmats[1][key_map['g']][key_map['n']] = 1
+    transmats[1][key_map['n']][key_map['i']] = 1
+    transmats[1][key_map['i']][key_map['z']] = 1
+    transmats[1][key_map['z']][key_map['e']] = 1
+    transmats[1][key_map['e']][key_map['d']] = 1
+    transmats[1][key_map['d']][key_map[' ']] = 1
+    transmats[1][key_map[' ']][key_map['c']] = 1
+    transmats[1][key_map['c']][key_map['h']] = 1
+    transmats[1][key_map['h']][key_map['a']] = 1
+    transmats[1][key_map['a']][key_map['r']] = 1
+    transmats[1][key_map['r']][key_map['a']] = 1
+    transmats[1][key_map['a']][key_map['c']] = 1
+    transmats[1][key_map['c']][key_map['t']] = 1
+    transmats[1][key_map['t']][key_map['e']] = 1
+    transmats[1][key_map['e']][key_map['r']] = 1
+    transmats[1][key_map['r']][key_map['s']] = 1
+
+    for corpus, expected in zip(corpora, transmats):
+        print('--- Running test ---')
+        actual, _ = create_transmat(corpus, keys=keys)
+        diff = actual == expected
+        if not diff.all():
+            for i in range(diff.shape[0]):
+                for j in range(diff.shape[1]):
+                    if not diff[i][j]:
+                        print(f"Failed: '{reverse_map[i]}' -> '{reverse_map[j]}'\n",
+                              f"Expected {expected[i][j]} but got {actual[i][j]}")
+        else:
+            print('Passed')
+        print()
+
+
+# Utilities
+
+def id_map(keys):
+    """Return a dict mapping each char in string to a unique integer.
+    
+    >>> id_map('abcd')
+    {'a': 0,
+     'b': 1,
+     'c': 2,
+     'd': 3}
+    """
+    return {value: key for key, value in dict(enumerate(keys)).items()}
+
+
+def reverse_id_map(keys):
+    """Return a dict mapping a unique integer to each char in string.
+
+    >>> reverse_id_map('abcd')
+    {0: 'a',
+     1: 'b',
+     2: 'c',
+     3: 'd'}
+    """
+    return dict(enumerate(keys))
+
+
+def pprint_transmat(transmat, keys):
+    """Pretty-print the transition matrix."""
+    h, w = transmat.shape
+    print('  ', end='')
+    for key in keys:
+        print(f'{key} ', end='')
+    print()
+    for i in range(h):
+        print(f'{keys[i]} ', end='')
+        for j in range(w):
+            print(f'{transmat[i][j]} ', end='')
+        print()
+
+
+def main():
+    encode = read_encode()
+
+    # TODO
+    model = hmm.GaussianHMM(n_components=2)
+
+
+if __name__ == '__main__':
+    main()