If it runs, its fine… right? Not really, because as programmers we should strive to write good quality code in order to make our life easier. In this post, I will first showcase some standard practices, then move on to some debatable points which some might disagree.

Sequence data is everywhere. One example is timestamped transactions, something that almost every company has. Increasingly companies are also collecting unstructured natural language data such as product reviews. While techniques like RNN are widely used for NLP problems, we can actually use it for any form of sequence-like predictions.Therefore, in this post I will explore more on how we can utilise CNN and LSTM for sequence modelling!

Image classification is one of the areas of deep learning that has developed very rapidly over the last decade. However, due to limited computation resources and training data, many companies found it difficult to train a good image classification model. Therefore, one of the emerging techniques that overcomes this barrier is the concept of transfer learning.

Time series forecasting is one of the common problems that we face everyday. Some of these include predicting equity prices, inventory levels, sales quantity, and the list goes on. In R, Auto ARIMA is one of the favourite time-series modelling techniques. However, if you are a Python user, you can implement that too using Pyramid. As Auto ARIMA has many tunable parameters, it is crucial for us to understand the mechanics behind this algorithm!

In June 2016, Google published a paper on wide and deep learning. This neural network architecture is supposedly great for regression and classification problems with sparse inputs, such as recommendation systems or search ranking problems. Since, I am currently working on a search ranking problem, I thought it would be great to familiarise myself with this architecture and see how it works. In this post, I will first go into the intuition behind wide and deep learning. After which, I will showcase the full implementation of this technique.

Many time a company might have many features that it wants to roll out to its customers. However, no one really knows if a new feature is beneficial as it has never been implemented before. Hence, an experiment is usually set up to test its incremental benefit. A properly crafted experiment will allow the experimenter to understand what is the minimum sample size to collect before the experiment. During or after the experiment, the experimenter can also understand what is statistical power of the experiment and determine if he/she should collect more samples.

Decision tree is one of the most interpretable machine learning model. However, it tends to overfit when learning the decision boundaries of the training data. Hence, decision trees are usually used as the base learner of ensembling methods such as Gradient Boosting and Random Forest. These ensemnbling methods have seen extreme success in the data science world. XGBoost, LightGBM, and Catboost are some of the popular gradient boosting methods that are widely used in competitions. But before getting there, we need to first understand how decision tree works as the base learner.

The field of artificial intelligence (AI) was founded in 1956. For the next 56 years till 2012, artificial intelligence was not widely adopted due to reasons such as lack of computing power, fundings, among others. In 2012, the one and only AI breakthrough happened and many AI applications started to build on top of this discovery.

Logistic regression is one of most widely used classification learning algorithms in various fields, including machine learning, most medical fields, and social sciences. Similar to the post on linear regression, I will go into the mechanics behind logistic regression in order for us to gain a deeper understanding of it.

Linear regression is one of the most basic learning algorithms in machine learning. At the same time, it is very important because it introduces concepts such as gradient descent and cost functions, which are used in most machine learning models. Hence, it is very important to understand the mechanics behind linear regression.

Recently, I discovered xLearn which is a high performance, scalable ML package that implements factorization machines (FM) and field-aware factorization machines (FFM). The first version of xLearn was released about 1 year ago as of writing.