Coverage for application/workflow.py : 19%

import json 

import logging 

import xml 

import cophi 

import lda 

import numpy as np 

import pandas as pd 

from application import database 

from application import utils 

def wrapper(): 

"""Wrapper for the topic modeling workflow. 

""" 

try: 

logging.info("Just started topic modeling.") 

data = utils.get_data("corpus", 

"topics", 

"iterations", 

"stopwords", 

"mfw") 

logging.info("Fetched user data...") 

database.insert_into("textfiles", 

data["corpus"]) 

logging.info("Inserted data into database.") 

# 1. Preprocess: 

dtm, token_freqs, parameters = preprocess(data) 

logging.info("Successfully preprocessed data.") 

database.insert_into("token_freqs", 

json.dumps(token_freqs)) 

# 2. Create model: 

model = create_model(dtm, data["topics"], data["iterations"]) 

parameters["log_likelihood"] = int(model.loglikelihood()) 

database.insert_into("parameters", json.dumps(parameters)) 

logging.info("Successfully created topic model.") 

# 3. Get model output: 

topics, descriptors, document_topic = get_model_output(model, dtm) 

logging.info("Got model output.") 

# 4. Calculate similarities: 

topic_similarities, document_similarities = get_similarities(document_topic) 

logging.info("Successfully calculated topic and document similarities.") 

data = {"document_topic": document_topic.to_json(orient="index", force_ascii=False), 

"topics": json.dumps(topics, ensure_ascii=False), 

"document_similarities": document_similarities.to_json(force_ascii=False), 

"topic_similarities": topic_similarities.to_json(force_ascii=False)} 

database.insert_into("model", data) 

logging.info("Successfully inserted data into database.") 

logging.info("Very nice, great success!") 

except xml.etree.ElementTree.ParseError as error: 

logging.error("ERROR: There is something wrong with your XML files.") 

logging.error("ERROR: {}".format(error)) 

logging.error("Redirect to error page...") 

except UnicodeDecodeError: 

logging.error("ERROR: There is something wrong with your text files. " 

"Are they UTF-8 encoded?") 

logging.error("Redirect to error page...") 

except Exception as error: 

logging.error("ERROR: {}".format(error)) 

logging.error("Redirect to error page...") 

def preprocess(data): 

"""Preprocess text data. 

""" 

# Constructing corpus: 

textfiles = database.select("textfiles") 

documents = utils.get_documents(textfiles) 

corpus = cophi.model.Corpus(documents) 

num_tokens = corpus.num_tokens 

database.update("textfiles", num_tokens.to_dict()) 

# Get paramter: 

D, W = corpus.dtm.shape 

N = num_tokens.sum() 

# Cleaning corpus: 

stopwords = utils.get_stopwords(data, corpus) 

hapax = corpus.hapax 

features = set(stopwords).union(set(hapax)) 

logging.info("Cleaning corpus...") 

dtm = corpus.drop(corpus.dtm, features) 

# Save stopwords: 

database.insert_into("stopwords", json.dumps(stopwords)) 

# Save parameters: 

parameters = {"n_topics": int(data["topics"]), 

"n_iterations": int(data["iterations"]), 

"n_documents": int(D), 

"n_stopwords": int(len(stopwords)), 

"n_hapax": int(len(hapax)), 

"n_tokens": int(N), 

"n_types": int(W)} 

return dtm, num_tokens.tolist(), parameters 

def create_model(dtm, topics, iterations): 

"""Create a topic model. 

""" 

logging.info("Creating topic model...") 

model = lda.LDA(n_topics=topics, 

n_iter=iterations) 

model.fit(dtm.values) 

return model 

def get_model_output(model, dtm): 

"""Get topics and distributions from topic model. 

""" 

logging.info("Fetching model output...") 

# Topics and their descriptors: 

topics = dict(utils.get_topics(model, dtm.columns)) 

descriptors = list(topics.keys()) 

# Document-topic distribution: 

document_topic = utils.get_document_topic(model, dtm.index, descriptors) 

return topics, descriptors, document_topic 

def get_similarities(document_topic): 

"""Calculate similarities between vectors. 

""" 

logging.info("Calculating topic similarities...") 

topics = utils.get_cosine(document_topic.values, document_topic.columns) 

logging.info("Calculating document similarites...") 

documents = utils.get_cosine(document_topic.T.values, document_topic.index) 

return topics, documents