def get_code_trie(self): """ Returns a Trie data structure from the words in the code """ if self.__code_trie == None: self.__code_trie = Trie(self.code_alphabet) for word in self.code_words: self.__code_trie.insert(word) return self.__code_trie
def is_uniquely_decodable(self): """Returns boolean specifying whether or not is uniquely decodable For the code to be uniquely decodable means that the function code that takes a string from a the source and codifies it is inyective. Returns: True if the code is uniquely decodable, False otherwise """ if self.__uniquely_decodable == None: self.__C_n = [self.code_words] trie_0 = Trie(self.code_alphabet, words=self.code_words) while True: new_words = set() trie_n = Trie(self.code_alphabet, words=self.__C_n[-1]) for word_i in self.__C_n[-1]: for new_word in trie_0.complete_word(word_i): new_words.add(new_word) for word_i in self.__C_n[0]: for new_word in trie_n.complete_word(word_i): new_words.add(new_word) if new_words == set() or self.__repeated_C_n(): break self.__C_n.append(new_words) self.__C_infty = set() for i in range(1, len(self.__C_n)): for word in self.__C_n[i]: self.__C_infty.add(word) self.is_uniquely_decodable = len(self.__C_infty & self.__C_n[0]) == 0 return self.is_uniquely_decodable
class Code: """ Core class for Coding and Decoding. It has integrated core metrics to it. To create a new Code you need to pass the rule for the code as a dictionary. Attributes: arity: An integer representing the code arity code_alphabet: A set containing the code alphabet source_alphabet: A set containing the source alphabet """ __code_trie = None __source_trie = None __uniquely_decodable = None def __init__(self, code_dictionary): """" Initiates the class using a code dictionary """ self.code = validate_dictionary(code_dictionary) self.code_alphabet = set() self.source_words = set() self.code_words = set() self.source_alphabet = set() self.__reverse_code = {} coding_output = '' source_input = '' for key, value in self.code.items(): self.source_words.add(key) coding_output += value source_input += key self.code_words.add(value) self.__reverse_code[value] = key for simbol in coding_output: self.code_alphabet.add(simbol) for simbol in source_input: self.source_alphabet.add(simbol) self.arity = len(self.code_alphabet) if self.is_prefix(): self.__decoder = Encoder(self.code_alphabet, self.__reverse_code) self.__encoder = Encoder(self.source_alphabet, self.code) def get_arity(self): """Returns arity of the code Return: Integer value of the arity """ return self.arity def is_prefix(self): """ Uses a Trie data structure to determine if the code is prefix Returns: A boolean True if the code is prefix, False otherwise """ return self.get_code_trie().is_prefix() @staticmethod def kraft_inequality_metric(words_length, arity): """ Calculates Kraft Inequality for given word lengths Measures Kraft Inequality Metric defined as .. math:: \sum_{i} r^{-l_i} where r means arity, and l sub i the length of the ith word Also returns if an instantaneous code can be made with said words Typical Usage: .. highlight:: python .. code-block:: python >>> metric, can_be_inst = kraft_inequality_metric(words_length=[4,4,3,6],arity=2) >>> print(metric, can_be_inst) 0.265625 True Args: words_length: An list with the length of the words arity: Arity of the code Returns: metric , can_be_instantaneous. Kraft Inequality metric. A boolean refering to whether or not a code with this word lenghts and arity exists. This is dependant on the metric being less than 1 """ if arity == None: raise Exception('arity can not be null') if type(arity) != int: raise Exception('arity must be an int') if words_length == None: raise Exception('words_length can not be null') metric = 0.0 epsilon = 0.00001 for l in words_length: metric += float(arity)**float(-l) return metric, metric - epsilon <= 1 def get_code_trie(self): """ Returns a Trie data structure from the words in the code """ if self.__code_trie == None: self.__code_trie = Trie(self.code_alphabet) for word in self.code_words: self.__code_trie.insert(word) return self.__code_trie def __repeated_C_n(self): """True if C_n has began a cycle, False otherwise""" if len(self.__C_n) <= 3: return False for i in range(1, len(self.__C_n) - 1): if self.__C_n[-1] == self.__C_n[i] and self.__C_n[ -2] == self.__C_n[i - 1]: return True return False def is_uniquely_decodable(self): """Returns boolean specifying whether or not is uniquely decodable For the code to be uniquely decodable means that the function code that takes a string from a the source and codifies it is inyective. Returns: True if the code is uniquely decodable, False otherwise """ if self.__uniquely_decodable == None: self.__C_n = [self.code_words] trie_0 = Trie(self.code_alphabet, words=self.code_words) while True: new_words = set() trie_n = Trie(self.code_alphabet, words=self.__C_n[-1]) for word_i in self.__C_n[-1]: for new_word in trie_0.complete_word(word_i): new_words.add(new_word) for word_i in self.__C_n[0]: for new_word in trie_n.complete_word(word_i): new_words.add(new_word) if new_words == set() or self.__repeated_C_n(): break self.__C_n.append(new_words) self.__C_infty = set() for i in range(1, len(self.__C_n)): for word in self.__C_n[i]: self.__C_infty.add(word) self.is_uniquely_decodable = len(self.__C_infty & self.__C_n[0]) == 0 return self.is_uniquely_decodable def get_C_n(self): """Returns C_n related to the uniquely decodable algorithm. Returns: List of set of words. """ self.is_uniquely_decodable() return self.__C_n def encode(self, sentence): """"Encodes a given sentence """ return self.__encoder.encode_sentence(sentence) def __encode_brute_force(self, sentence): """Encodes a given sentence using bruteforce """ encoding = '' curr_idx = 0 sentence_length = len(sentence) while curr_idx < sentence_length: advance = False for word in self.source_words: if re.match('^' + word, sentence[curr_idx:]): curr_idx += len(word) encoding += self.code[word] advance = True break if not advance: raise Exception('No match found to encode {} '.format( sentence[curr_idx:])) return encoding def decode(self, sentence, get_any=True): """Decodes a given sentence, in case multiple outputs are possible it returns any of them if the get_any flag is True. Args: sentence: Sentence to be decoded. It must compel with the source alphabet. get_any: True by default. It gives a string when multiple answers are possible. When False it gives the answer in a set of possible answers. Returns: A string with the decoded sentence """ if self.is_prefix(): return self.__decoder.encode_sentence(sentence) result = self.__decode_brute_force(sentence) if len(result) == 0: raise Exception('No decoding possible') if get_any: return result.pop() return result def __decode_brute_force(self, sentence, curr_idx=0): results = set() # Base Case if curr_idx >= len(sentence): return [''] for word in self.code_words: if re.match('^' + word, sentence[curr_idx:]): for result in self.__decode_brute_force( sentence, curr_idx=curr_idx + len(word), ): results.add(self.__reverse_code[word] + result) return results def get_word_length(self, source): """Gets the average word lenght of the code for a given Source Args: source: A source of type Source Returns: A float with the value of the word lenght of the code for the given source. """ if type(source) != Source: raise Exception('source must be of type source') word_length = 0.0 for word, probability in source.dictionary.items(): if word not in source.dictionary: raise Exception('{} not part of the code source '.format(word)) word_length += probability * float(len(self.code[word])) return word_length @staticmethod def make_huffman_code(source, code_alphabet={'0', '1'}): """Generates an optimal Huffman Code Args: source: A source of type Source code_alphabet: Iterable with the alphabet. The items should be of type str Returns: Optimal huffman code for the given source and with arity of the given code_alphabet """ if type(source) != Source: raise Exception('source must be of type source') base_nodes = [(probability, word) for word, probability in source.dictionary.items()] arity = len(code_alphabet) if arity != 2: while len(base_nodes) % (arity - 1) != 1: base_nodes.append((float(0), '')) huffman = Huffman_Generator(simbols=code_alphabet, base_nodes=base_nodes, arity=len(code_alphabet)) return Code(huffman.make_huffman_code())
def test_complete_word(): simbols = {'0' , '2' , '1'} t = Trie(simbols, words = {'01','21','0112','212','011'}) assert t.complete_word('01') == {'1','12'} assert t.complete_word('21') == {'2'} assert t.complete_word('1') == set()
def test_inserting_and_checking(): simbols = {'0' , '1' , '2'} t = Trie(simbols) t.insert('00') t.insert('101') t.insert('2') assert not t.search('0') assert t.search('00') assert not t.search('10') assert t.search('101') assert t.search('2')
def test_words_init(): simbols = {'0' , '1' , '2'} t = Trie(simbols, words = ['120','000','101']) assert t.search('120') assert not t.search('100')
def test_prefix(): simbols = {'0' , '1' , '2'} t = Trie(simbols) t.insert('0') t.insert('1') t.insert('20') t.insert('21') t.insert('22') assert t.is_prefix() t.insert('01') assert not t.is_prefix()