Day 4 of the Machine Learning 100 Days Challenges
Machine learning is changing the way we work and think. Experts are in high demand, and companies fight for their attention.
Rule #26: Look at errors to find new features.
Alice works on improving a machine learning model that predicts install rates for apps. She adds a feature that reduces the logistic loss. The model’s metrics rise.
Day 1: What is Machine Learning?
Machine learning is the umbrella term for solving problems that are difficult or impossible to solve with traditional programming. It allows machines to identify patterns and make predictions based on past experience.
For example, your email inbox relies on machine learning to automatically filter out spam. It also powers medical imaging, fraud detection and drug development. The list goes on and on.
Day 2: Neural Networks
Neural networks are at the heart of machine learning algorithms. They are a powerful tool for classifying, clustering and identifying patterns in data that human experts could not easily identify.
Neural networks are able to perform complex tasks, such as pattern recognition and nonlinear mapping. This makes them ideal for tasks like image and speech recognition, financial analysis and predictive modeling.
Day 3: Classification
Day 3 of the machine learning 100 days challenges introduces classifiers, which categorize data based on their output. This is an important first step in building complex ML models.
The next few days focus on data preprocessing and visualization. These are critical steps in the ML process, and they involve handling missing values, outliers, and duplicates; normalizing data; and transforming data through encoding, selecting features, and reducing dimensionality.
Day 4: Regression
Regression is the process of finding a line that fits data points using a mathematical criterion. The most common regression models are linear and logistic regression.
Regression is useful for forecasting quantitative questions, such as predicting someone’s height based on their parents’ heights or predicting customer churn based on past purchasing behavior. It also helps answer qualitative questions like ‘when will this occur’.
Day 5: Clustering
Clustering is a common machine learning technique that finds natural groups or segments within your data. It is used as a preprocessing step before using a classification algorithm.
For example, a bookstore can use clustering to identify customer demographics and create targeted marketing campaigns. Cluster analysis can also be used to analyze sports or crime data. It can also help find outlier data points and identify anomalies.
Day 6: Natural Language Processing
The ability to use natural language in computer programs has led to such innovations as speech-to-text apps and smart speakers like Amazon’s Alexa and Google Home. It’s also behind the best chatbots, which learn to recognize context and provide better responses over time.
This course includes a wide range of Python exercises that explore data preprocessing, modeling, analysis, and visualization using Matplotlib and Seaborn. It also includes supervised and unsupervised learning, regression models and clustering.
Day 7: Linear Regression with Derivation
The goal of machine learning is to predict values for a dependent variable (Y) based on a set of independent features or variables (X). This is called regression.
Finding a linear regression equation is pretty tricky because it requires some serious math. Watch the video below to learn more. It also covers some basic data preprocessing and visualization techniques. It also introduces the concept of regularization, which is used to prevent models from overfitting to the data.
Day 8: Perceptrons
Perceptrons are an algorithm or linear classifier that facilitates supervised learning binary classification. They have four key parameters: input values, weights, bias and activation function.
It takes a set of scalar input values and multiplies them by a weight vector and a constant bias to get a net sum. This net sum is then passed through an activation function to determine the output.
Day 9: McCulloch-Pitts Neurons
Neurons are able to calculate the sum of incoming excitatory inputs. When the AND function is passed through a neuron, the output is 1. If the AND is not passed, it gives output as 0.
The early 20th century brought about cross-disciplinary efforts by mathematicians, computer scientists and neuroscientists to formalize computability theory. This resulted in a slew of characterizations of computational devices including the Turing machine.
Day 10: Deep Learning
Machine learning is a subfield of artificial intelligence, which seeks to create computer programs that mimic intelligent human behavior. It’s used in digital assistants, like Alexa or Google Assistant, in medical imaging and diagnostics, like identifying tuberculosis on x-rays, and self-driving cars.
It also powers fraud detection, by analyzing patterns to identify potentially suspicious credit card transactions or log-in attempts. But it requires large amounts of labeled data and computing power.