klastrovanie.com

Why klastroTest?

While Propensity Score Matching (PSM) is a popular technique for reducing bias in observational studies, it has its own limitations. We do not use Propensity Score Matching. klastroTest observes all features of the input data and matches the control with treatment groups.

A/B Test Results in 3.6 Seconds? The Wait is Over!

In the fast-paced world of retail, timing is everything. Our solution, over thousand times faster than the traditional CPU based solution, delivers A/B test results in just about 3.6 seconds, empowering you to make real-time, data-driven decisions. Optimize campaigns daily and stay ahead of trends—no more waiting.

GPU Accelerated

Our benchmarks show our advanced solution processing matching tasks with tens of thousands of data points in about 3.6 seconds—something that takes traditional methods over 15+hours. Experience thousand times faster speeds with klastroTest*

*Experiment group: 45,192 x 12, Control group: 11,297 x 12 – Traditional solution: 15+ hours vs. klastroTest: 3.6 seconds

When the number of columns exceeds 100, Apple Silicon outperforms server GPUs by more than 2x.**

As dataset size grows, the performance gap only widens. Apple Silicon is so fast that cloud GPUs are simply unnecessary.
From personal to enterprise, we deliver the full speed of MPS acceleration in Python.

**Experiment group: 12,000 x 100, Control group: 12,000 x 100- server GPU took more than 20 secs to complete matching. (excluding network upload & download overheads bottleneck)

For Even Faster Results, Experience the Power of On-Premises Solutions:

On your Apple Silicon Mac, you can experience the blazing-fast speed of just 3.6 seconds for the same dataset-over thousands times faster than traditional CPU solutions! If you have a high-performance GPUs on severs, you can still complete the same task in as little as about 3.7-3.8 seconds. klastroTest is optimized for both cutting-edge Apple Silicon and NVIDIA GPUs, ensuring researchers, analysts, and developers can achieves results in record time with unmatched reliability.*

*This product is independent from and not affiliated with, endorsed by, or sponsored by Apple Inc. or NVIDIA Corporation. Benchmark results are based on internal tests conducted under standard conditions and may vary depending on hardware and software configurations.

Missing Confounders? Not a Problem

PSM relies on observed variables to estimate the propensity score, which represents the probability of an individual receiving a treatment. If crucial confounding variables are missing from the dataset, the propensity score will be inaccurate, leading to imperfect matching and biased treatment effect estimates.
In contrast, klastroTest ensures comprehensive matching without missing confounders, delivering more reliable treatment effect estimates.

Optimized for High-Dimensional Data Matching

As the number of variables increases, accurately estimating the propensity score becomes more complex. In high-dimensional data, the “curse of dimensionality” can hinder PSM’s ability to effectively match individuals, potentially increasing bias and reducing the efficiency of the analysis.
Howerver, klastroTest handles high-dimensional data with ease, avoiding the pitfalls of traditional methods.

Adaptable Matching Without Linear Assumptions

Many PSM implementations utilize logistic regression to predict propensity scores. Logistic regression assumes a linear relationship between the variables and the outcome. However, this assumption might not hold true in all cases, leading to inaccurate propensity score estimation and subsequent matching errors.
Unlike traditional methods, klastroTest requires no linear assumptions, ensuring flexibility and accuracy across diverse datasets.

klastroTest is Simpler

klastroTest is conceptually simpler and computationally less demanding than PSM. It doesn’t require complex modeling or estimation procedures, making it easier for researchers to implement. This can reduce the potential for errors and biases that can arise during the propensity score estimation process in PSM.

klastroTest is not Deep Learning

klastroTest does not require large-scale data training processes and delivers accurate results without the need for pre-existing training data. Unlike deep learning, it eliminates the need for data acquisition and preprocessing, saving both time and cost. Additionally, klastroTest provides more precise matching by analyzing multidimensional data, surpassing simple Euclidean distance or Cosine Similarity-based methods. This enables it to operate effectively even in high-dimensional data environments, delivering both accuracy and efficiency.