warsztaty-prefect/main.py

175 lines
4.7 KiB
Python
Raw Permalink Normal View History

2020-06-13 23:03:53 +02:00
import pandas as pd
import string
import re
import nltk
from nltk.tokenize import word_tokenize
2020-06-14 14:23:49 +02:00
from nltk.corpus import stopwords
2020-06-13 23:03:53 +02:00
from nltk.stem import WordNetLemmatizer
from sklearn.feature_extraction.text import TfidfVectorizer
2020-06-14 14:23:49 +02:00
from sklearn.linear_model import LogisticRegression
2020-06-13 23:03:53 +02:00
from sklearn.model_selection import train_test_split
2020-06-14 16:41:43 +02:00
from prefect import task, Flow, context
2020-06-14 14:23:49 +02:00
from pandas import DataFrame
2020-06-13 23:03:53 +02:00
nltk.download('stopwords')
nltk.download('wordnet')
nltk.download('punkt')
# remove urls, handles, and the hashtag from hashtags (taken from https://stackoverflow.com/questions/8376691/how-to-remove-hashtag-user-link-of-a-tweet-using-regular-expression)
def remove_urls(text):
new_text = ' '.join(re.sub("(@[A-Za-z0-9]+)|([^0-9A-Za-z \t])|(\w+:\/\/\S+)"," ",text).split())
return new_text
2020-06-14 14:23:49 +02:00
2020-06-13 23:03:53 +02:00
# make all text lowercase
def text_lowercase(text):
return text.lower()
2020-06-14 14:23:49 +02:00
2020-06-13 23:03:53 +02:00
# remove numbers
def remove_numbers(text):
result = re.sub(r'\d+', '', text)
return result
2020-06-14 14:23:49 +02:00
2020-06-13 23:03:53 +02:00
# remove punctuation
def remove_punctuation(text):
translator = str.maketrans('', '', string.punctuation)
return text.translate(translator)
2020-06-14 14:23:49 +02:00
2020-06-13 23:03:53 +02:00
# tokenize
def tokenize(text):
text = word_tokenize(text)
return text
2020-06-14 14:23:49 +02:00
2020-06-13 23:03:53 +02:00
# remove stopwords
stop_words = set(stopwords.words('english'))
def remove_stopwords(text):
text = [i for i in text if not i in stop_words]
return text
2020-06-14 14:23:49 +02:00
2020-06-13 23:03:53 +02:00
# lemmatize
lemmatizer = WordNetLemmatizer()
def lemmatize(text):
text = [lemmatizer.lemmatize(token) for token in text]
return text
2020-06-14 14:23:49 +02:00
2020-06-13 23:03:53 +02:00
def preprocessing(text):
text = text_lowercase(text)
text = remove_urls(text)
text = remove_numbers(text)
text = remove_punctuation(text)
text = tokenize(text)
text = remove_stopwords(text)
text = lemmatize(text)
text = ' '.join(text)
return text
2020-06-14 14:23:49 +02:00
@task
def get_train_set() -> DataFrame:
2020-06-14 16:41:43 +02:00
logger = context.get("logger")
2020-06-14 14:23:49 +02:00
train = pd.read_csv('train.csv')
train = train.drop(['keyword', 'location'], axis=1)
2020-06-14 16:41:43 +02:00
logger.info(f"Train set: {len(train)} elements")
2020-06-14 14:23:49 +02:00
return train
@task
def get_test_set() -> DataFrame:
2020-06-14 16:41:43 +02:00
logger = context.get("logger")
test = pd.read_csv('test.csv')
logger.info(f"Test set: {len(test)} elements")
return test
2020-06-14 14:23:49 +02:00
@task
def preprocess_train(train: DataFrame) -> DataFrame:
pp_text_train = []
for text_data in train['text']:
pp_text_data = preprocessing(text_data)
pp_text_train.append(pp_text_data)
train['pp_text'] = pp_text_train
return train
@task
def preprocess_test(test: DataFrame) -> DataFrame:
pp_text_test = []
for text_data in test['text']:
pp_text_data = preprocessing(text_data)
pp_text_test.append(pp_text_data)
test['pp_text'] = pp_text_test
return test
@task
def prepare_vectorizer(train_data: DataFrame, test_data: DataFrame) -> TfidfVectorizer:
train_text_data = list(train_data['pp_text'])
test_text_data = list(test_data['pp_text'])
corpus = train_text_data + test_text_data
tf = TfidfVectorizer()
fitted_vectorizer = tf.fit(corpus)
return fitted_vectorizer
@task
def transform_train(vectorizer: TfidfVectorizer, train_set: DataFrame) -> DataFrame:
return vectorizer.transform(train_set)
@task
def transform_test(vectorizer: TfidfVectorizer, test_set: DataFrame) -> DataFrame:
return vectorizer.transform(test_set)
@task
def split_test_set(X: DataFrame, Y: DataFrame) -> dict:
X_train, X_test, y_train, y_test = train_test_split(X, Y)
return {'X_train': X_train, 'X_test': X_test, 'y_train': y_train, 'y_test': y_test}
@task
def train_model(X: DataFrame, Y: DataFrame) -> LogisticRegression:
scikit_log_reg = LogisticRegression()
model = scikit_log_reg.fit(X, Y)
return model
@task
def evaluate(model: LogisticRegression, X: DataFrame, Y: DataFrame) -> None:
2020-06-14 16:41:43 +02:00
logger = context.get("logger")
2020-06-14 14:23:49 +02:00
predictions = model.predict(X)
count = 0
for guess, answer in zip(predictions, Y):
if guess == answer:
count += 1
2020-06-14 16:41:43 +02:00
score = count/len(Y)
logger.info(f"model score: {count/len(Y)}")
2020-06-14 14:23:49 +02:00
if __name__ == "__main__":
2020-06-14 16:41:43 +02:00
with Flow("My First Prefect Flow!") as flow:
2020-06-14 14:23:49 +02:00
train_data = get_train_set()
test_data = get_test_set()
train_data = preprocess_train(train_data)
test_data = preprocess_test(test_data)
vectorizer = prepare_vectorizer(train_data, test_data)
vectorized_train_data = transform_train(vectorizer, train_data['pp_text'])
vectorized_test_data = transform_test(vectorizer, train_data['pp_text'])
splitted_data = split_test_set(vectorized_train_data, train_data['target'])
model = train_model(splitted_data['X_train'], splitted_data['y_train'])
evaluate(model, splitted_data['X_test'], splitted_data['y_test'])
flow.validate()
# flow.visualize()
flow.run()