SearchEngine101Final.py

import numpy as np
import os
import re
from nltk.corpus import stopwords
from scipy.spatial.distance import cosine
from sklearn.feature_extraction.text import TfidfVectorizer

"""
Notes about set, list, dictionary, tuple:
    
    1.  list = [1, 2, 3]
        dictionary = {1: "one", 2: "two", 3: "three"}
        tuple = (1, 2, 3)
        set = {1, 2, 3}
    2.
        functions
        print(list((1, 2, 3)))
        print(dict(one = 1, two = 2, three = 3))
        print(tuple((1, 2, 3)))
        print(set((1, 2, 3)))
    3.  
        mutability:
            list: mutable
            dictionary: mutable
            tuple: immutable
            set: immutable
            
    4.Order:
        Lists and tuples maintain order. 
        Dictionaries and sets are unordered     

    5.Duplication:
        Lists and tuples allow duplication. 
        Dictionary and set items are unique.
    6.Index
        List and tuple items may be accessed by index. 
        Dictionary items are accessed by key. 
        Set items cannot be accessed by index
    7.Set Operations
            subset, superset, union (|), intersection (&), and difference (-).

"""

class SearchEngine101:

    #test_content store a list of content of all files, each file is represented by str.
    def __init__(self, dir):
        all_files = os.listdir(dir)
        self.text_content = [open(dir+"/"+file_name).read() for file_name in all_files]
        self.preprocess()
        self.index()
        print "Preprocessing and Indexing done..."

    #remove non-letter occurences and convert into lowercase, store them in list called clean_texts.
    def preprocess(self):
        """
        Cleans self.texts and sets the cleaned_texts attribute.
        """
        self.cleaned_texts = []

        # remove non-letter occurrences from the texts
        self.cleaned_texts = [re.sub(r"[^A-Za-z]"," ",text) for text in self.text_content]

        # convert into lowercase
        self.cleaned_texts = [text.lower() for text in self.cleaned_texts]

    # remove stop words return a list contained each words.
    def tokenize(self, text):
        """
        Will split a text into words and remove stop words
        :param text: string - can be a document or a query
        :return: list - a list of words
        """
        words = None

        # Split the text into a sequence of words and store it in words
        # Use .split()
        words = text.split()

        # remove stopwords
        stopwords_ = set(stopwords.words('english'))
        words = [word for word in words if word not in stopwords_]

        return words


    def index(self):
        split_texts = [self.tokenize(text) for text in self.cleaned_texts]

        # use a dictionary (HashMap) for storing the inverted index
        # Key : word
        # Value : List of document indices
        self.inverted_index = {}

        for index,split_text in enumerate(split_texts):
            split_text = set(split_text) #beauty of using set right here.
            for word in split_text:
                self.inverted_index[word] = self.inverted_index.get(word, []) # beauty of of using get(x,y) funciotn
                self.inverted_index[word].append(index)

        # sort all the values of the dictionary
        self.inverted_index = {key: sorted(self.inverted_index[key]) for key in self.inverted_index}

    def intersection(self, list1, list2):
        """
        Should return intersection of list1 and list2
        :param list1: list of integers
        :param list2: list of integers
        :return:
        """
        intersection = []
        #populate the intersection list and return. practice the two-fingure algorithm.
        ptr1 = 0
        ptr2 = 0
        while ptr1<len(list1) and ptr2<len(list2):
            if list1[ptr1] == list2[ptr2]:
                intersection.append(list1[ptr1])
                ptr1 += 1
                ptr2 += 1
            elif list1[ptr1]<list2[ptr2]:
                ptr1 += 1
            else:
                ptr2 += 1

        return intersection


    # return a founded documents list which contains two components, one for the [cleaned tests], another for [text_content].
    def filter(self, query):
        """
        Returns the filtered list of texts [both cleaned and original] which contain the query terms
        :param query: string - user query
        :return: filterd_list: list - list of documents that contain the query terms
        """
        query_terms = self.tokenize(query)

        # Retrieve document list for each of the terms in the query
        document_lists = [self.inverted_index[term] for term in query_terms]

        # !!! Optimise the document lists for faster intersection
        document_lists = sorted(document_lists, key = lambda x: len(x))

        # Now, iteratively take intersection
        document_indices = document_lists[0]
        for document_list in document_lists[1:]:
            document_indices = self.intersection(document_indices, document_list)


        return [self.cleaned_texts[index] for index in document_indices], [self.text_content[index] for index in document_indices]


    def vectorize(self, filtered_texts, query):
        """
        Store the vectors and vectorizer.
        """
        self.vectors = None
        self.vectorizer = None

        # !!! Use TfIdfVectorizer. It automatically converts into a matrix.
        self.vectorizer = TfidfVectorizer(lowercase=True, stop_words="english", norm=None, use_idf=True, smooth_idf=False,
                                     sublinear_tf=False, decode_error="ignore")
        self.vectors = self.vectorizer.fit_transform(filtered_texts+[query])

    def retrieve_ranked_list(self):
        """
        Return the indices of text_vectors in decreasing order of cosine similarity and the scores
        :param text_vectors: the vectors of the text
        :param query_vector: the vector of the query
        :return: indices, scores: indices of top 10 documents and scores of all documents
        """
        similarities = []

        # !!! Populate the similarities array with cosine similarities between text_vectors and query_vector
        query_vector = self.vectors.getrow(self.vectors.shape[0] - 1)
        for i in xrange(0,self.vectors.shape[0] - 1):
            similarities.append(cosine(self.vectors.getrow(i).todense(),query_vector.todense()))

        # Return the top 10 indices of the similarities array
        return np.argsort(similarities)[::-1][:10], similarities

    def print_list(self,text_content, scores, text_indices):
        print len(text_indices), "Results Found!\n"
        print "*******************************\n"
        for index in text_indices:
            print scores[index]
            print text_content[index]
            print "*******************************\n"

    def search(self, query):
        filtered_clean, filtered_orig = self.filter(query)
        self.vectorize(filtered_texts=filtered_clean, query=query)
        text_indices, scores = self.retrieve_ranked_list()
        self.print_list(text_content=filtered_orig, scores= scores, text_indices=text_indices)


if __name__ == "__main__":
    # Write the driver code here
    engine = SearchEngine101(dir = "data")
    engine.search("dalhousie university")