(Solved): Data Mining Homework Using Bigrams And N Grams- Prediction

assignment1_part2 September 30, 2021 [ ]: version = "REPLACE_PACKAGE_VERSION" 1 Assignment 1 Part 2: N-gram Language Models (Cont.) (30 pts) In this assignment, we’re going to train an n-gram language model that is able to “imitate” William Shakespeare’s writing. [ ]: # Configure nltk import nltk nltk_data_path = "assets/nltk_data" if nltk_data_path not in nltk.data.path: nltk.data.path.append(nltk_data_path) [1]: # Copy and paste the functions you wrote in Part 1 here and import any␣ ↪→libraries necessary # We have tried a more elegant solution by using # from ipynb.fs.defs.assignment1_part1 import load_data, build_vocab,␣ ↪→build_ngrams # but it doesn't work with the autograder... def load_data(): sentences = [] with open ('assets/gutenberg/THE_SONNETS.txt') as fin: pattern = '[0-9]' list = [re.sub(pattern, '', i) for i in fin] for line in list: line = line.lower() if((line.strip()) and not (line.isnumeric())): sentence_list = line.strip().split(" ") last_char = sentence_list[-1][-1] last_word = sentence_list[-1] 1 last_word = last_word[:-1] sentence_list[-1] = last_word sentence_list.append(last_char) sentences.append(sentence_list) print(len(sentences)) return sentences def build_vocab(sentences): vocab = [] for x in sentences: for y in x: if y not in vocab: vocab.append(y) vocab.append('<s>') vocab.append('</s>') for x in vocab: print(x) return vocab def build_ngrams(n, sentences): all_ngrams = [] padded_zip_codes_pattern = [pad_both_ends(zcp, n=n) for zcp in sentences] all_ngrams = [ngrams(pzcp, n=n) for pzcp in padded_zip_codes_pattern] return all_ngrams 1.1 Question 4: Guess the next token (20 pts) With the help of the three functions you wrote in Part 1, let’s first answer the following question as a review on n-grams. Assume we are now working with bi-grams. What is the most likely token that comes after the sequence <s> <s> <s>, and how likely? Remember that a bi-gram language model is essentially a first-order Markov Chain. So, what determines the next state in a first-order Markov Chain? Complete the function below to return a tuple, where tuple[0] is a str representing the mostly likely token and tuple[1] is a float representing its (conditional) probability of being the next token. 2 [ ]: def bigram_next_token(start_tokens=("<s>", ) * 3): """ Take some starting tokens and produce the most likely token that follows␣ ↪→under a bi-gram model """ next_token, prob = None, None # YOUR CODE HERE raise NotImplementedError() return next_token, prob [ ]: # Autograder tests stu_ans = bigram_next_token(start_tokens=("<s>", ) * 3) assert isinstance(stu_ans, tuple), "Q4: Your function should return a tuple. " assert len(stu_ans) == 2, "Q4: Your tuple should have two elements. " assert isinstance(stu_ans[0], str), "Q4: tuple[0] should be a str. " assert isinstance(stu_ans[1], float), "Q4: tuple[1] should be a float. " # Some hidden tests del stu_ans 1.2 Question 5: Train an n-gram language model (10 pts) Now we are well positioned to start training an n-gram language model. We can fit a language model using the MLE class from nltk.lm. It requires two inputs: a list of all n-grams for each sentence and a vocabulary, both of which you have already written a function to build. Now it’s time to put them together to work. Complete the function below to return an nltk.lm.MLE object representing a trained n-gram language model. [ ]: from nltk.lm import MLE def train_ngram_lm(n): """ Train a n-gram language model as specified by the argument "n" """ lm = MLE(n) # YOUR CODE HERE raise NotImplementedError() 3 return lm [ ]: # Autograder tests stu_n = 4 stu_lm = train_ngram_lm(stu_n) stu_vocab = build_vocab(load_data()) assert isinstance(stu_lm, nltk.lm.MLE), "Q3b: Your function should return an␣ ↪→nltk.lm.MLE object. " assert hasattr(stu_lm, "vocab") and len(stu_lm.vocab) == len(stu_vocab) + 1,␣ ↪→"Q3b: Your language model wasn't trained properly. " del stu_n, stu_lm, stu_vocab FINALLY, are you ready to compose sonnets like the real Shakespeare?! We provide some starter code below, but absolutely feel free to modify any parts of it on your own. It’d be interesting to see how the “authenticity” of the sonnets is related to the parameter n. Do the sonnets feel more Shakespeare when you increase n? [ ]: # Every time it runs, depending on how drunk it is, a different sonnet is␣ ↪→written. n = 3 num_lines = 14 num_words_per_line = 8 text_seed = ["<s>"] * (n - 1) lm = train_ngram_lm(n) sonnet = [] while len(sonnet) < num_lines: while True: # keep generating a line until success try: line = lm.generate(num_words_per_line, text_seed=text_seed) except ValueError: # the generation is not always successful. need to␣ ↪→capture exceptions continue else: line = [x for x in line if x not in ["<s>", "</s>"]] sonnet.append(" ".join(line)) break # pretty-print your sonnet print("\n".join(sonnet)) 4 Assignment 1 Part 2: N-gram Language Models (Cont.) (30 pts) Question 4: Guess the next token (20 pts) Question 5: Train an n-gram language model (10 pts)

Data Mining Homework Using Bigrams And N Grams- Prediction

Related Study Services

Post your Answer - free or at a fee

Answer Review

Data Mining Homework Using Bigrams And N Grams- Prediction

Related Study Services

Post your Answer - free or at a fee

Login to your tutor account to post an answer