Day 3 of making AI make sense Topic: What Kind of Data Can We Feed a ML Model? For a model to learn, it needs data. Data comes in different forms, and every learning approach works better with a specific data form.
 But first, what is data in machine learning? Data, simply put, is the set of examples that an algorithm studies to find structural patterns or mathematical relationship in order to make predictions or perform tasks. 
 Typically, the data fed into the traditional machine learning algorithms is of two parts. Inputs (called features) → facts, attributes, or measurements given to the model. For example, if you want to predict a car's fuel efficiency, its weight, engine size, and horsepower are the features. 
 Outputs (called Labels) → the target outcomes that you want the model to predict for every input given. In the car example, the actual fuel rating (miles per gallon) is the label. Think of it this way, Features are basically just the attributes of the case or object the algorithm studies (just like your physical features are just the visible attributes of your anatomy). And Labels are what a group of these features put together determine (just like how if you're vertebrate, warm-blooded, and hairy, you fall under the "mammal" label). Now that we've established what data is, let's look at the types of data a model could learn from: → Structured data: information that is completely organized into rows and columns, like a database or a spreadsheet. Each row represents an individual record while each column represents a specific attribute. Structured data works very well with traditional machine learning models because the inputs are neatly arranged into columns which human engineers can select and evaluate directly during training. → Unstructured data: raw, messy, and unorganized information that do not come in tables or spreadsheet. These are mostly images, audios, videos and natural language text. Deep learning models work better with unstructured data and instead of relying on manual feature engineering like in the above ML example, they use neural networks (idk…human brain type of stuff) to analyze the raw data and extract features automatically. → Semi-structured data: This is data that sits somewhere between structured and unstructured because it contains "markers" that allow for it to be organized and grouped, but it doesn't follow a rigid relational database (tables with rows and columns) form. The data here is mostly contained in JSON, HTML, IoT sensor streams, etc formats. Engineers still have to extract, "flatten" or  convert these data forms into structured tables before feeding them to classical machine learning models.