Blog

Today, an online shopper digests information about a product before he/she makes a decision to buy. A chat bot guides shoppers through multiple channels such as social, search, mobile, and video all the way to checkout process.

Inside a business, an AI powered virtual agent (or chat bot) guides employees through procedures according to knowledge that those bots learned from previous best practices.

In both cases, a chat bot needs to  collect data from journeys that people have taken, then attributes to touch points that lead to conversions. For this purpose, algorithmic attribution models is one of approaches to accomplish the goal.

One popular algorithm in this domain is Markov Model. Markov Model formulates a process into states and movements chains, then a movement is the probability of transitioning to a next state from current state. To apply Markov Model in cases explained above, we use states to represent touch points, and the movements are calculated as probabilities that users transitioning from one touch point to a next one.

Attribution analysis is extremely valuable for uncovering the insights of conversion funnels.

In this post, we are going to demonstrate that how to detect objects on still images with Python and OpenCV. This post will focus on feature extraction, then following posts will demonstrate how to use features extracted to train models with samples, and detect objects against test images.

In real estate industry, swimming pool is an important factor in pricing a house. One popular approach to price a house is using a clustering algorithm to figure out the mean value in a certain area, then factor in other important auxiliaries to estimate a final price.

Once swimming pools can be detected from pictures by a machine learning algorithm, the process of price estimation may be automated.

To train such a supervised learning model, we need to extract features from image data. Here we demonstrate how to use OpenCV and Python to implement feature extraction. Note that this step maybe accomplished by a Deep Learning technique such as a R-CNN architecture during convolution.

The idea is that swimming pools are bluish, so we construct HSV masks in certain ranges and apply them to image data. We then use the filtered images as features to train a supervised learning model. This model may be used to detect swimming pools in unseen images.

Online finical service vendors will most likely face a challenge to detect malicious behaviors quick enough, then disable accounts and/or users associated with the malicious behaviors.

To achieve this goal, they need to collect user behavior data and apply online machine learning based algorithms to classify accounts at risk.

We introduce a simple real time fraud detection system. It implements a naive rule engine based on Flink’s Streaming API. Code can be located here.

Today, typical IoT applications collect millions or even billions data records via field-deployed sensors. Those data are time series by nature and may be used for various data driven business intelligence such as analytical insights, predictive monitoring, and prescriptive maintenance.

However, before Data Scientists or Machine Learning Practitioners can apply models on top of these data, one of challenges during pre-processing is that data sources may have been aggregated. Therefore, it is essential to decompose one time series curve into several components.

We show that how to decompose multiple seasonality in time series data with Python and Pandas on Jupyter notebook.

Here is the code to demonstrate how to do some basic exploratory analysis on sensor data collected by Perception System.

Use TensorFlow to build a Faster R-CNN model. See how it helps millions of mobile users to classify print texts. Download on App Store