More than just a machine learning library

what is slicematrix-io?

SliceMatrix-IO is an end-to-end (e2e) computational ecosystem which enables users to build machine intelligence systems with speed, agility and scalability on demand. What makes SliceMatrix-IO extremely powerful is that there is MINIMAL CODING required to get STARTED! With a FEW LINES OF CODE, users can build machine intelligent models and systems.

Diagram 1 Note that in the diagram 1 the end-user is referred to as a ‘Quant Conductor’ because IO enables the end-user to orchestrate the design, development and execution of machine intelligent systems with speed and ease that the end-user is able to do the task of a team of data scientist.

Diagram 1 Note that in the diagram 1 the end-user is referred to as a ‘Quant Conductor’ because IO enables the end-user to orchestrate the design, development and execution of machine intelligent systems with speed and ease that the end-user is able to do the task of a team of data scientist.

Real-Time Persistence: Build Once & Repurpose

Once a model or pipeline are built, it is stored in the cloud, meaning it can be re-used multiple times by different processes in parallel. This saves the end-user valuable time and effort when building models & analyzing multiple datasets across many projects. Hence, making the firm agile and flexible.

SliceMatrix-IO is a computational software platform that delivers powerful machine learning algorithms such as manifold learning, neural networks, network graphs, and much more. In addition to a deep and rich suite of machine learning solutions, one of the key benefits of SliceMatrix-IO is it’s ease of use. With minimal code users can build a model and begin processing data immediately.

However, SliceMatrix-IO is much more than a machine learning library and to only focus on IO’s algorithms is to neglect powerful aspects of its design such as built-in communication network and distributed computational architecture.

How does it work?


Users can build end-to-end machine intelligent solutions in Python with a few lines of code!
There are 2 core innovative concepts that allows users to build powerful machine learning with minimal code:

1. The Analytical Pipeline(AP) can be thought of as assembly lines of code which can train machine learning models. AP’s are reusable, meaning you can use the same AP to train multiple machine learning models using different input data.

2. The Model
The model are where the real magic happens. After the Pipeline trains a Model, the Model live’s on SliceMatrix-IO’s machine learning infrastructure in the cloud. That means you can access your model after you create it in a distributed fashion. You can also train models in parallel subject to your usage plan’s request / throttling limits.

Multi Analytical Pipeline
quant_conductor_3_cloud_big2.png

Shrink your API Deployment Time with SliceMatrix-IO Computational API

Move seamlessly from prototyping to
production. Scale up or down on
demand, based on business needs
and requirements.

how do you pay for slicematrix-io?

A platform build for real-time prediction

We built the SliceMatrix-IO Platform as a Service (PaaS) so that real-time prediction just works. Once you train your model its live and ready for production use in your client systems. Suppose you are a pairs trader (stat arb) that needs to generate hedge ratios for trading a list of co-integrated symbols. How many symbols could we monitor in real-time with SliceMatrix-IO?


Real World use case

As any good pairs trader will tell you, hedge ratios are liable to change over time. One way to combat this problem is to use a rolling window of regressions: i.e. constantly updating your hedge ratio with a trailing window of data. The problem is this method is guaranteed to fail over time as you're always looking backwards, averaging what has happened in the past as your best guess of the future hedge ratio.

But a superior method is to employ a dynamic linear model such as the Kalman Filter (KF). The KF is the optimal model for estimating the parameters of linear gaussian models such as pairs trading.

How to Calculate Monthly Cost

Now some back of the envelope calculation, assuming you have the Neo Matrix usage plan i.e. 5k API calls per month

1 initial call per pair to train the model and 1 call to get the initial hedge ratio (using the "getState" method)

the one (1) call to update the model with every day's new closing prices (using the "update" method). This method will retrain the model with new information in the most optimal fashion and generate a new hedge ratio for the pair.
Assuming 20 trading days per month:

5000 / 22 calls per stock = ~227 stocks per month
i.e. you could generate daily hedge ratios for 227 stock symbols for just $25 per month

 
icon1white.png
 

slicematrix-io(beta) + quant trading

 
Code.png
 

analytical pipelines & models up close

The diagram below is a hedge ratio workflow that begins by training an Isomap model, a manifold learning algorithm which learns the hidden connections within the dataset. We can use the methods contained within this model to find the best hedges for a given stock, say AAPL.

Then we use another model called the Kalman Filter. This is an example of a class of algorithms known as Bayesian Filters. These models can learn new information as time passes, and so are ideal for generating  hedge ratios in a pairs trading strategy.

 
 

slicematrix-io(beta) powerful machine intelligence algorithms