Automatic Hypothesis Discovery
The Core of Daisho: Accelerated Problem Solving
“At the end of the day, some machine learning projects succeed and some fail. What makes the difference? Easily the most important factor is the features (hypotheses) used.”
Pedro Domingos
A Few Useful Things About Machine Learning
Accelerated Problem Solving: Feature Discovery in Daisho
An analyst typically starts off with a list of hypotheses - from experience, literature, experts, etc. These are the pale boxes in the picture.
However, the orange box is completely missed - and thats typically where the most interesting insights lie.
The Daisho SignalFactory generates 1000s of sophisticated business understandable hypotheses AUTOMATICALLY so that all four boxes are covered.
The Daisho SignalFilter then AUTOMATICALLY filters the most powerful features!
Optimized compute for BUSINESS context
End-to-End Optimization for Speed and Efficiency
Business data isnt classical Machine Learning data.
While data can be large, it is often very wide but shallow. For example businesses want to analyze weekly sales data, and that by definition has only 52 data-points in a year. But, there typically are thousands of such time-series.
Many kinds of data also tends to be time-series, e.g, SCADA/sensors, sales, etc., while one also sees transaction/event level data which lends itself to profiling, clustering, etc.
Our algorithms are geared towards EXACTLY these kinds of data - they figure out the data types automatically, and analyze them accordingly. All with MINIMAL USER INPUT.
Feature Engineering with the right amount of abstraction, computational robustness and speed.
There are a large number of hypotheses which are mathematically similar. We managed to extract the similarities to be able to build them all. Consider MTBF in manufacturing, and frequency in retail. While the two of them seem and mean very different things, mathematically they are exactly the same.
However, its easy to abstract at the wrong level. Take the example of churn. Churn in e-commerce needs to be treated very differently from churn in a SaaS context.
Our feature engineering strikes the correct balance, and is also designed to ensure robustness. All this while being wicked fast.
Mathematically Correct and Fast Algorithms over "cool" algorithms anyday.
We are fanatical over mathematical correctness and speed: All our algorithm choices are driven by these two factors.
Daisho knows which algorithm to apply in which context - the decision is driven by mathematical correctness. We then optimized these algorithms so that the results are wicked fast. As an example, Daisho is able to analyse 50,000 + individual time-series is less than 5 minutes!
Tailored for Business Consumption
Wide variety of interactive visualizations and simulations
And many many more...
Relationships Strucures in your data
Clusters where an experiment has had the most impact
Anomaly Detection along with Expected Ranges
Scenarios and Simulations
And many many more...
Relationships Strucures in your data
Want to Have a Chat?
We would love to talk.
Or just email us on contact@daisho.ml