SVM in Python¶
Let's build a Machine Learning model that can identify whether an email is spam or not, so this is a binary classification problem:
Step 1. Reading the data set¶
import pandas as pd
total_data = pd.read_csv("https://raw.githubusercontent.com/4GeeksAcademy/machine-learning-content/master/assets/spam.csv")
total_data.head()
Step 2: Data processing¶
Categorical to numerical transformation¶
We transform our two categories spam and ham into numerical values (0 and 1) since this, like most models, does not work with categorical class variables:
total_data["Category"] = total_data["Category"].apply(lambda x: 1 if x == "spam" else 0).astype(int)
total_data.head()
Elimination of repeated values¶
We can easily count how many cases of each class we have to analyze whether the data set is balanced or not:
print(total_data.shape)
print(f"Spam: {len(total_data.loc[total_data.Category == 1])}")
print(f"No spam: {len(total_data.loc[total_data.Category == 0])}")
Duplicates, if any, should also be removed:
total_data = total_data.drop_duplicates()
total_data = total_data.reset_index(inplace = False, drop = True)
total_data.shape
In this case we see that more than 400 repeated records have been eliminated.
Text processing¶
In order to train the model it is necessary to first apply a transformation process to the text. We start by transforming the text to lowercase and removing punctuation marks and special characters:
import regex as re
def preprocess_text(text):
# Remove any character that is not a letter (a-z) or white space ( )
text = re.sub(r'[^a-z ]', " ", text)
# Remove white spaces
text = re.sub(r'\s+[a-zA-Z]\s+', " ", text)
text = re.sub(r'\^[a-zA-Z]\s+', " ", text)
# Multiple white spaces into one
text = re.sub(r'\s+', " ", text.lower())
# Remove tags
text = re.sub("</?.*?>"," <> ", text)
return text.split()
total_data["Message"] = total_data["Message"].apply(preprocess_text)
total_data.head()
The next step is lemmatization of the text, which is the process of simplifying words to their base or canonical form, so that words with different forms, but the same semantic core are treated as a single word. For example, the verbs "running", "ran" and "runs" will be lemmatized to "run", just as the words "best" and "better" could be lemmatized to "good".
In addition, taking advantage of lemmatization, we will also eliminate stopwords, which are words that we consider irrelevant for text analysis because they appear very frequently in the language and do not provide meaningful information. There are two ways: either we create our own list of words to eliminate or we use external libraries.
Both tasks will be carried out with the Python library NLTK, which is one of the most important in terms of NLP:
from nltk import download
from nltk.corpus import stopwords
from nltk.stem import WordNetLemmatizer
download("wordnet")
lemmatizer = WordNetLemmatizer()
download("stopwords")
stop_words = stopwords.words("english")
def lemmatize_text(words, lemmatizer = lemmatizer):
tokens = [lemmatizer.lemmatize(word) for word in words]
tokens = [word for word in tokens if word not in stop_words]
tokens = [word for word in tokens if len(word) > 3]
return tokens
total_data["Message"] = total_data["Message"].apply(lemmatize_text)
total_data.head()
Something very common once we have the tokens is to represent them in a word cloud. A word cloud is a visual representation of the words that make up a text, where the size of each word indicates its frequency or importance in the text.
This visual representation allows us to quickly identify the most relevant or repeated terms or concepts in a dataset, since the most frequent or significant words stand out by their size. We can easily implement it using the wordcloud library for Python:
import matplotlib.pyplot as plt
from wordcloud import WordCloud
wordcloud = WordCloud(width = 800, height = 800, background_color = "black", max_words = 1000, min_font_size = 20, random_state = 42)\
.generate(str(total_data["Message"]))
fig = plt.figure(figsize = (8, 8), facecolor = None)
plt.imshow(wordcloud)
plt.axis("off")
plt.show()
The last step before training the model is to convert it into numbers, since models cannot be trained with textual categories. In previous modules we saw how we could transform text into numeric vectors using the CountVectorizer of scikit-learn, but here we will apply a new process to provide more tools to the student:
from sklearn.feature_extraction.text import TfidfVectorizer
tokens_list = total_data["Message"]
tokens_list = [" ".join(tokens) for tokens in tokens_list]
vectorizer = TfidfVectorizer(max_features = 5000, max_df = 0.8, min_df = 5)
X = vectorizer.fit_transform(tokens_list).toarray()
y = total_data["Category"]
X[:5]
TfidfVectorizer converts a collection of raw text documents into a TF-IDF feature matrix. TF-IDF is a measure that quantifies the importance of a word in a document relative to a corpus. It is composed of two terms:
- TF (Term Frequency): it is the frequency of a word in a document.
- IDF (Inverse Document Frequency): Measures the importance of the term in the corpus. A term that appears in many documents may not be as informative.
Therefore, by using TfidfVectorizer, we transform a collection of text documents into a numerical matrix representing the relative importance of each word in each document, relative to the entire corpus. This matrix is commonly used as input for machine learning algorithms, especially in natural language processing tasks such as text classification.
Train test split¶
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 42)
Step 3: Initialization and training of the model¶
from sklearn.svm import SVC
model = SVC(kernel = "linear", random_state = 42)
model.fit(X_train, y_train)
Step 4: Model prediction¶
Once the model has been trained, it can be used to predict with the test data set.
y_pred = model.predict(X_test)
y_pred
With raw data, it is very difficult to know whether the model is getting it right or not. To do this, we must compare it with reality. There are many metrics to measure the effectiveness of a model in predicting, including accuracy, which is the fraction of predictions that the model makes correctly.
from sklearn.metrics import accuracy_score
accuracy_score(y_test, y_pred)
Step 5: Saving the model¶
Once we have the model we were looking for (presumably after hyperparameter optimization), to be able to use it in the future it is necessary to store it in our directory, together with the seed.
from pickle import dump
dump(model, open("svm_classifier_linear_42.sav", "wb"))
Adding an explanatory name to the model is vital, since in the case of losing the code that has generated it we will know, on the one hand, what configuration it has (in this case we say linear because the kernel is linear) and also the seed to replicate the random components of the model, which in this case we do by adding a number to the file name, 42.