Just like Maslow’s hierarchy of needs is a motivational theory which categorises the needs of humans in terms of importance using a bottom-up approach, a similar data hierarchy can be created in our existing data-driven world.

Building on Part 1 of an introduction to reinforcement learning, we looked at the underlying logic of reinforcement learning for the algorithm to learn an optimal policy and to take better actions that lead to good outcomes. We also looked at the Q-function, which helps to adjust the action value…

Quantifying P(s, a, s′) where the probability of each state > action > new state. The goal of reinforcement learning is for the algorithm to learn an optimal policy and take an optimal action when presented with state s. Over time, we reinforce actions that lead to good outcomes and…

Take in a sequence of words that correspond to the n-words and map each of these words to a vector (word embeddings), where word embeddings = meaning of the words. Positional embeddings (d dimensions) are applied to account for word order as word embeddings in their basic form do not…

NLP models will map each word in the vocabulary to a word vector (or word embeddings). The aim is to predict the next or surrounding words for each word in a given document (given word A, it might indicate that another word might be present in the document with a…

Earlier this week, I had the opportunity to help a friend working with an exercise regarding the subtraction of dates between two columns in a data frame. Fortunately, after much trial and error, I found a quick and simple solution for this issue with just three lines of code. Here's…

The softmax function is an activation function in the final layer of a neural network. It is a multi-category equivalent of a sigmoid function and is used whenever there are more than two outcomes (e.g. non-binary). The probabilities of the categories must sum up to 1. Allows for predicting probabilities…

One of the ways deep learning networks learn and improve is via the Gradient Descent (SGD) optimisation algorithm. The algorithm works by seeking to calculate the lowest loss between the actual and predicted value at every step along a curve. A typical SGD process for deep learning algorithms consists of…

In the previous post, we looked at one of the common text-generating techniques, Greedy Search decoding, which aims to select the word with the highest probability at each timestep. Greedy Search Decoding For Text Generation In Python Most of us would likely have a simple conversation with a chatbot in this technological age. But, from automated…medium.com However, instead of taking the absolute probabilities of each token itself, Beam Search decoding considers all possible extensions of each token…