Big Data Science

12 Aug 2024 17:59

⚡️📊OpenAI now provides normal structured JSON with data

I would like to remind you that the JSON mode has been working for about a year, but the outputs of the models corresponded to the declared format in less than half of the cases.

However, there is great news for developers who need good data markup. The updated version gpt-4o-2024-08-06 no longer has this problem: 100% of tests have no errors in the format.

The code and tutorial on using the feature are here.

Big Data Science

07 Aug 2024 17:59

💡😎Interesting Caldera Dataset

The Caldera dataset is an open source scene dataset containing much of the geometry found in the game Call of Duty®: Warzone™.

This includes geometry that can be visualized, as well as some alternate, usually unseen representations used in other calculations. For example, the developers have included volumes here to aid in lighting calculations or simple shapes for collision detection. Excluded are many single-point entities, such as character spawn locations or complex script-based models. As the developers note, they decided not to include textures and materials in this release. That would have added complexity and size to an already heavy scene. They focused on the many connections between spatial elements that can be found in this set, rather than an accurate visual representation.

Big Data Science

31 Jul 2024 18:03

🌎TOP DS-events all over the world in August
Aug 2-4 - MLMI 2024 - Osaka, Japan - https://mlmi.net/
Aug 3-9 - International Joint Conference on Artificial Intelligence (IJCAI) - Jeju, South Korea - https://ijcai24.org/
Aug 5-6 - ICASAM 2024 - Vancouver, Canada - https://waset.org/applied-statistics-analysis-and-modeling-conference-in-august-2024-in-vancouver
Aug 7-8 - CDAO Chicago - Chicago, United States - https://da-metro-chicago.coriniumintelligence.com/
Aug 12-14 - AI4 2024 - Las Vegas, United States - https://ai4.io/vegas/
Aug 16-17 - Machine Learning for Healthcare 2024 - Toronto, Canada - https://www.mlforhc.org/
Aug 19-20 - Artificial Intelligence and Machine Learning - Toronto, Canada - https://www.scitechseries.com/artificial-intelligence-machine
Aug 19-22 - The Bioprocessing Summit - Boston, USA - https://www.bioprocessingsummit.com/
Aug 25-29 - ACM KDD 2024 - Barcelona, Spain - https://kdd2024.kdd.org/
Aug 27 - Azure AI Summer Jam -
Aug 27-29 - ITCN Asia 25th - Karachi, Pakistan - https://itcnasia.com/karachi/
Aug 31 - DATA SATURDAY #52 - Oslo, Norway - https://datasaturdays.com/Event/20240831-datasaturday0052

Big Data Science

26 Jul 2024 17:58

😎💡Benchmark for comprehensive assessment of LLM logical thinking

ZebraLogic is a benchmark based on logic puzzles and is a set of 1000 program-generated tasks of varying difficulty - with a grid from 2x2 to 6x6.

Each puzzle consists of N houses (numbered from left to right) and M features for each house. The task is to determine the unique distribution of feature values across the houses based on the provided clues.
Language models are given one example of the puzzle solution with a detailed explanation of the reasoning process and the answer in JSON format. Models must then solve a new problem, providing both the reasoning progress and the final solution in a given format.

Evaluation Metrics:
1. Puzzle-level accuracy (percentage of completely correctly solved puzzles).
2. Cell-level accuracy (percentage of correctly completed cells in the solution matrix).

🟡 Project Page
🟡 Dataset

Local launch of ZebraLogic as part of the ZeroEval framefork:

# Install via conda

conda create -n zeroeval python=3.10
conda activate zeroeval

# pip install vllm -U # pip install -e vllm 

pip install vllm==0.5.1
pip install -r requirements.txt
# export HF_HOME=/path/to/your/custom/cache_dir/

# Run Meta-Llama-3-8B-Instruct via local, with greedy decoding on `zebra-grid`
bash zero_eval_local.sh -d zebra-grid -m meta-llama/Meta-Llama-3-8B-Instruct -p Meta-Llama-3-8B-Instruct -s 4

Big Data Science

22 Jul 2024 17:58

😎Graph database implemented on the Apache Apache TinkerPop3 framework

HugeGraph is an open-source graph database implemented on the Apache TinkerPop3 framework and fully compatible with the Gremlin query language.

HugeGraph supports the import of over 10 billion vertices and edges and can process queries very quickly (at the ms level).

Typical HugeGraph application scenarios include exploring relationships between objects, association analysis, path finding, feature extraction, data clustering, community detection, and graph construction.

Quick start with Docker:

docker run -itd --name=graph -p 8080:8080 hugegraph/hugegraph
# docker exec -it graph bash

Big Data Science

17 Jul 2024 17:59

⚡️🔎Fully Synthetic Dataset

A huge dataset consisting entirely of synthetic data has appeared on Hugging Face.

The LLM (in this case GPT-4o + VLLM) generates answers by representing itself each time with some character: for example, a chemical scientist or a musician.

Synthetic data can sometimes help a lot (especially when the task is abstract and there is no structured information), but they are still treated with caution. They are not realistic enough, they are not diverse enough, and they potentially harbor hallucinations. It is still unclear whether we will ever be free to use “synthetics”, but it is actively being worked on.

Big Data Science

12 Jul 2024 17:59

💡Another small selection of AI tools for Big Data analytics

KNIME Analytics Platform is a free, open-source platform that allows users to stay at the forefront of data science and has 300+ connectors to various data sources. and integrates with all popular machine learning libraries.

Polymer - artificial intelligence for transforming data into an optimized, flexible and powerful database. All a user needs to do is upload their spreadsheet to the platform to instantly transform it into an optimized database that can then be mined for insights.

IBM Cognos Analytics is a componentized online business intelligence (BI) service that provides access to a wide range of functions for creating business reports, data analysis, event monitoring and metrics to develop effective business decisions.

Akkio is a business intelligence and forecasting tool that allows users to analyze their data and predict potential outcomes. The AI ​​tool allows users to upload their dataset and select the variable they want to predict, which helps Akkio build a neural network around that variable. Like many other tools, Akkio requires no prior programming experience.

Monkeylearn - uses AI data analytics capabilities to help users visualize and reorganize their data. It can also be used to set up text classifiers and text extractors, which help automatically sort data according to topic or intent, and extract product characteristics or user data.

Big Data Science

08 Jul 2024 17:59

⚡️💡💻 MySQL 9.0.0 has been released

Oracle recently released MySQL DBMS 9.0.0. The developers of the project have prepared and made publicly available MySQL Community Server 9.0.0 builds for major Linux, FreeBSD, macOS and Windows distributions.

In 2023, the company announced a change in the MySQL DBMS release formation model. Developers began releasing two types of MySQL branches: Innovation (new features, frequent updates, three months of support) and LTS (with extended support time and unchanged behavior).

As the developers note, the MySQL 9.0 project is assigned to the Innovation branch, which will also include the next major releases of MySQL 9.1 and 9.2.

Distributions based on Innovation branches are recommended for those users who want to get access to new functionality earlier. They are published every 3 months and are supported only until the next major release is published (for example, after the 9.1 branch is released, support for the 9.0 branch will be discontinued).

Big Data Science

03 Jul 2024 17:59

🎼Datasets and projects for music generation and analysis tasks

MAESTRO - (MIDI and Audio Edited for Synchronous Tracks and Organization) contains over 200 hours of annotated recordings of international piano competitions over the past ten years.

NSynth - the dataset consists of 305,979 musical notes and includes recordings of 1006 different instruments, such as flute, guitar, piano and organ. The dataset is annotated by instrument type (acoustic, electronic or synthetic) and other sound parameters.

Lakh MIDI v0.1 - There are 176,581 MIDI files in the dataset, of which 45,129 are associated with samples from the Million Song Dataset. This dataset is designed to simplify the search for music information based on text and audio content on a large scale.

Music21 - contains musical performances from 21 categories and is aimed at solving research problems (for example, finding an answer to the question: “Which group used these chords for the first time ?)

Big Data Science

28 Jun 2024 17:59

⚡️Hyperconverged cloud open-source database

MatrixOne is a hyperconverged cloud distributed database with a structure that separates storage, compute and transactions into a single HSTAP data engine.
This mechanism allows a single database system to handle a variety of business workloads such as OLTP, OLAP, and stream computing.

MatrixOne supports deployment and use in public and private clouds, providing compatibility with a variety of infrastructures.

🖥 GitHub
🟡 Documentation

Big Data Science

24 Jun 2024 17:59

📊A huge dataset of images and their captions

Pixel Prose is a dataset that contains over 16 million diverse images from three different web databases (commonPool, CC12M, RedCaps) with captions created using Google Gemini 1.0 Pro Vision.

The following Python script can be used to load a dataset using the API:
from datasets import load_dataset
# for downloading the whole data
ds = load_dataset("tomg-group-umd/pixelprose")

Big Data Science

19 Jun 2024 17:59

📊💡Dataset for video analysis

CinePile is a question and answer based video understanding dataset. It was created using advanced large language models (LLMs). The dataset includes approximately 300,000 data points for training and 5,000 data points for testing.

Each row in the dataset consists of a question (dtype: string), five answer choices (dtype: list), and an answer_key (dtype: string). The auxiliary columns store the movie title, movie genre, video clip titles, etc.
To load a dataset via Python script, you can use the following command:
from datasets import load_dataset
dataset = load_dataset("tomg-group-umd/cinepile")

Big Data Science

14 Jun 2024 17:59

💡🔎Interesting and useful repository
Jailbreak - a repository that contains a dataset consisting of 15,140 ChatGPT queries from Reddit, Discord, hacking websites, and open source datasets (including 1,405 jailbreak queries gpt answers).
According to the developers, they collected 15,140 messages from these four platforms between December 2022 and December 2023.

Big Data Science

10 Jun 2024 17:59

💡🔎Platform Extension Framework (PXF): advantages and disadvantages
The Platform Extension Framework (PXF) is a powerful tool provided by many modern platforms to extend their functionality. PXF allows developers to create plugins and add-ons that integrate into the core platform, providing system flexibility and extensibility.
Its advantages include:
1. Time-tested solution based on open source code with the possibility of modification to suit your needs
2. Modularity: PXF allows functionality to be divided into independent modules. This makes it easier to develop, test, and maintain code.
3. Extensibility: With PXF, you can easily add new capabilities or integrate with external services and tools, allowing the platform to evolve with your business needs.
4. Speed ​​up development: PXF-enabled platforms often provide ready-made tools and APIs that speed up the development process and make it easier to implement new features.
5. A set of connectors to popular data sources available out of the box (Hadoop stack, data sources available via JDBC, cloud storage).
But there are also a number of disadvantages:
1. The need to support a separate solution based on your own stack.
2. Allocation of resources, as a rule, on the same servers where the DBMS itself is deployed.
3. Multiple transformations and transfer of the same data on the way from representation in the DBMS to the types that PXF itself operates on.
4. Security: Since extensions may have access to sensitive data and platform functions, it is important to ensure their security and prevent possible vulnerabilities.
5. Compatibility: Platform updates may introduce compatibility issues with existing extensions, requiring additional testing and adaptation.

The Platform Extension Framework provides powerful capabilities for extending and adapting platforms, allowing developers to create custom solutions and improve system functionality. However, it is important to consider the potential challenges and risks associated with integrating and supporting extensions in order to maximize the potential of PXF.

Big Data Science

05 Jun 2024 17:59

💡🔎Not very well known, but very useful ETL services
Astera Centerprise is an enterprise-grade, ready-to-use ETL solution that offers data integration and transformation capabilities for raw data of any complexity and size in a variety of formats: from complex hierarchical files and unstructured documents to industry formats such as EDI, and even legacy data such as COBOL.
Talend is an open source software platform that offers data integration and management solutions. Talend specializes in big data integration. This tool provides features such as cloud, big data, enterprise application integration, data quality, and master data management. It also provides a single repository for storing and reusing metadata.
Skyvia is a web service for cloud data integration and backup. It offers ETL tools to integrate cloud CRM with other data sources and allows users to control all their business data. Data can be viewed and manipulated using SQL. Skyvia provides easy data integration without programming skills.
Pentaho is a business intelligence tool that provides clients with a wide range of business intelligence solutions. It is capable of reporting, data analysis, data integration, data extraction, etc. Pentaho also offers a complete set of BI features that can improve business productivity and efficiency.
Hevo Data is an ETL platform that supports data integration, movement, and processing. It supports a wide range of data sources and offers real-time data replication. This tool facilitates the extraction, transformation and loading of data to the designated target destinations.

Big Data Science

09 Aug 2024 17:59

💡😎The book "PostgreSQL 16 from the inside" is now freely available

The Postgres Professional DBMS developer has released a new book "PostgreSQL 16 from the inside". The electronic version of the textbook is freely available. The author of the book is Egor Rogov, Director of Educational Program Development at Postgres Professional.

The first edition of this textbook, based on version 14 of PostgreSQL, was released in March 2022 and updated to version 15. Due to great reader interest, the company translated the book into English. It later became the most popular thematic publication of 2023 according to Postgres Weekly and was included in the list of professional literature on the official website of the PostgreSQL community.

The current edition of the book "PostgreSQL 16 from the Inside" takes into account readers' comments, corrects typos, and reflects changes that occurred in the PostgreSQL 16 version. Postgres Professional has also updated the localized documentation for PostgreSQL 16.

Big Data Science

05 Aug 2024 14:04

💡😎A startup that revolutionized the way we process data

CRAM is a new memory technology that can reduce energy consumption when processing AI data by 1000 times.

Researchers from the University of Minnesota have developed a new technology, Computational Random-Access Memory (CRAM), that can reduce energy consumption when processing data. Unlike traditional solutions, where data moves between memory and the processor, CRAM allows data to be processed directly in memory cells.

This is achieved through the use of a high-density and reconfigurable spintronic structure embedded in memory cells. Thus, the data does not leave the memory, which minimizes response delays and energy consumption associated with the transfer of information.

With CRAM, data never leaves memory, but is instead processed entirely within the computer’s memory array. This allows a system running an AI computing application to reduce power consumption by “about 1,000 times compared to a state-of-the-art solution,” according to the research team.

Big Data Science

29 Jul 2024 17:59

💡Datasets used to build various ML bases

Iphone dataset - a set of datasets on the basis of which more than 40 thousand dynamic and more than 100 thousand static Gaussians, 20 SE(3) bases were built using Shape of Motion

The training time on 1xGPU A100 using the Adam optimizer with a resolution of 960x720 was just over 2 hours at a rendering speed of 40 frames per second.

According to the results of tests during the training process, Shape of Motion showed good results in the quality and consistency of scene construction.
However, the method still requires optimization for each specific scene and cannot handle significant changes in camera angle. There is also a critical dependence on precise camera parameters and user input to create a moving object mask.

Big Data Science

24 Jul 2024 17:58

⚡️The largest collection of datasets of ~ 1 million pairs of problems and solutions for mathematical competitions

NuminaMath - datasets consisting of 1 million pairs of problems and solutions for various mathematical problems.

🔎Chain of Reasoning (CoT): 860 thousand pairs of problems and solutions created using CoT.

🛠 Tool-Integrated Reasoning (TIR): 73K synthetic solutions derived from GPT-4 with code execution feedback to break complex problems into simpler subproblems that can be solved using Python.

According to the researchers, models trained on NuminaMath achieve best-in-class performance among open-weight models and approach or beat their own models in math competition scores.

Big Data Science

19 Jul 2024 17:59

💡 Large video dataset with long duration and structured annotations

Tencent's MiraData is an off-the-shelf dataset with a total video duration of 16 thousand hours, designed to train text-to-video generation models. It includes long videos (average 72.1 seconds) with high motion intensity and detailed structured annotations (average 318 words per video).

To evaluate the quality of the dataset, a MiraBench benchmark system of 17 metrics assessing temporal consistency, motion in the frame, video quality, and other parameters was even specially created. According to their results, MiroData outperforms other known datasets available in open sources, which mostly consist of short videos with floating quality and short descriptions.

Big Data Science

15 Jul 2024 17:59

🔎Lakehouse architecture: advantages and disadvantages

Lakehouse architecture is designed to provide more flexible and efficient data processing, including data storage, processing and analytics. It is a hybrid approach that combines elements of a traditional Data Warehouse and a Data Lake.

Lakehouse advantages:

1. Data unification: Lakehouse architecture allows you to store structured and unstructured data in one place. This simplifies data access and analysis, eliminating the need for separate systems for each type of data.
2. Cost-effective: By using low-cost data storage solutions such as cloud storage objects, Lakehouse architecture can be more cost-effective compared to traditional data warehouses.
3. Flexibility and Scalability: Lakehouse supports scalability, making it easy to increase data storage and processing power as needed. This is especially important for companies working with large volumes of data and requiring high performance.
4. Compatibility with modern analytical tools: Many modern analytical tools and platforms, such as Apache Spark, Delta Lake and others, integrate with the Lakehouse architecture, providing high performance and reliability of data analysis.

Disadvantages of Lakehouse

1. Implementation Difficulty: Implementing Lakehouse architecture can require significant effort and expense in planning, designing, and configuring the system. This may include training staff and adapting existing processes and tools.
2. Data Quality Management: Merging data from different sources can lead to data quality issues, especially if there are no rigorous data cleaning and validation processes in place.
3. Security and Privacy: Consolidating large amounts of data in one place increases the risks associated with data security and privacy. Additional measures are required to protect data from unauthorized access and leaks.
4. Potential Data Access Latency: In some cases, the Lakehouse architecture may experience latency in data access, especially when processing large volumes of unstructured data.

Thus, Lakehouse architecture offers many benefits such as data unification, cost efficiency and flexibility, making it attractive to many organizations. However, its implementation is associated with certain challenges, including complexity of integration, data quality management and security issues.

Big Data Science

10 Jul 2024 17:59

⚡️Tool to significantly enhance the database

WrenAI is an open-source tool that makes your existing database RAG-ready.

It allows you to convert text to SQL, explore data from the database without writing SQL, and do many other things

🖥 GitHub
🟡 Documentation

Big Data Science

05 Jul 2024 17:59

💻High-performance distributed database

YugabyteDB is a high-performance distributed database that supports all PostgreSQL features.

YugabyteDB is well suited for cloud-based OLTP applications (i.e. real-time and business-critical) that require absolute data correctness and require scalability or high fault tolerance.

🖥 GitHub
🟡 Documentation

Creating a local YugabyteDB cluster with Docker:

docker run -d --name yugabyte -p7000:7000 -p9000:9000 -p15433:15433 -p5433:5433 -p9042:9042 \
 yugabytedb/yugabyte:2.21.1.0-b271 bin/yugabyted start \
 --background=false

Big Data Science

30 Jun 2024 17:59

🌎TOP DS-events all over the world in July
Jul 9 - The Martech Summit - Hong Kong, China - https://themartechsummit.com/hongkong
Jul 9-11 - DATA 2024 – Dijon, France - https://data.scitevents.org/
Jul 9-11 - Transform 2024 - San Francisco, USA - https://transform24.venturebeat.com/
Jul 11-12 - DataConnect Conference – Ohio, United States - https://www.dataconnectconf.com/
Jul 17 - Data & Analytics Live - Online - https://data-analytics-live.coriniumintelligence.com/
Jul 23 - CDAO Indonesia - Indonesia - https://cdao-id.coriniumintelligence.com/
Jul 26 - The MachineCon 2024 - New York, USA - https://machinecon.aimresearch.co/
Jul 29-30 - Gartner Data Analytics Summit - Sydney, Australia - https://www.gartner.com/en/conferences/apac/data-analytics-australia

Big Data Science

26 Jun 2024 17:59

⚔️🔎ACID in Kafka vs ACID in Airflow when processing Big data: advantages and disadvantages

When considering two popular data science tools such as Apache Kafka and Apache Airflow, it is important to understand how they deal with ACID principles (Atomicity, Consistency, Isolation, Durability). These principles are critical to ensuring reliable and predictable data processing.

Benefits of Kafka ACID:
1. Durability: Kafka stores data in disk memory, which ensures its safety even in the event of a system failure.
2. Consistency: When configured correctly, Kafka ensures that all consumers receive the same data in the same order.
3. Isolation: Messages in Kafka are divided into topics and sections, which helps isolate data processing between different threads.

Disadvantages of Kafka ACID:
1. Atomicity: Kafka does not always guarantee atomicity at the message level. In some cases, duplicate messages or omissions may occur if additional tools such as Kafka Transactions are not used.
2. Complexity of Configuration: Achieving ACID properties in Kafka requires complex configuration and management, including replication and transaction configuration.

Advantages of Airflow ACID:
1. Atomicity: Airflow provides atomicity at the task level. If a task fails, the entire DAG (Directed Acyclic Graph) can be re-run or restored from the point of failure.
2. Consistency: Airflow maintains a strict sequence of tasks, ensuring a consistent state of data.
3. Dependency Management: Airflow allows you to manage dependencies between tasks, making it easier to ensure data isolation and consistency.

Disadvantages of Airflow ACID:
1. Performance: Unlike Kafka, Airflow is not designed for real-time data processing. Its main purpose is to manage long-term and complex work processes.
2. Durability: Although Airflow maintains the state of tasks and DAGs, it relies on external data stores (such as databases) for long-term data storage, which may require additional effort to ensure durability.

Thus, Apache Kafka is better suited for real-time data processing with high performance and durability, but may require complex tuning to achieve atomicity and consistency. Apache Airflow, in turn, is great at managing and orchestrating complex workflows, providing atomicity and consistency at the task level, but is not designed for real-time streaming data processing.

Big Data Science

21 Jun 2024 17:59

⚡️💡Open-source data container orchestration system for running AI systems

dstack is an open-source container orchestration engine designed for AI workloads in any cloud or data center.

Cloud providers supported by this technology include AWS, GCP, Azure, OCI, Lambda, TensorDock, Vast.ai, RunPod, and CUDO.

If you have standard AWS, GCP, Azure or OCI credentials on your device, the dstack server will pick them up automatically.

🖥GitHub
🟡 Documentation

Big Data Science

17 Jun 2024 17:59

📊Dataset of characters from real to fictional characters

Character Codex - a dataset that contains data on 15,939 characters from a wide variety of sources, from anime to historical figures, scholars, and popular characters, both fictional and non-fictional!

Potential uses include use for generating synthetic data, analyzing RPG data, and more.
You can use a Python script to load a dataset:
from datasets import load_dataset
dataset = load_dataset("NousResearch/CharacterCodex")

Big Data Science

12 Jun 2024 17:59

💡🔎What is NoORM: advantages and disadvantages
NoORM (No Object-Relational Mapping) is an approach to working with databases that rejects the use of traditional ORM (Object-Relational Mapping) frameworks. Instead, developers interact directly with the database using native SQL queries or other specialized data manipulation techniques.
Advantages of NoORM:
1. Query Optimization: Because developers write SQL queries by hand, they can optimize them down to the last detail, often resulting in significant performance improvements over ORM-generated queries.
2. Minimize overhead: Using an ORM adds additional layers of abstraction that can slow down operations. NoORM eliminates these layers, which can also improve performance.
3. Support for complex data structures: NoORM allows you to work with non-standard data structures and relationships that may be difficult to implement through ORM.
4. Process Understanding: Developers have a thorough understanding of how data is accessed and modified, making debugging and optimization easier.
Disadvantages of NoORM:
1. Code Maintenance: Changing the database schema can require updating a lot of code, making the system difficult to maintain and develop.
2. Reduced portability: Code written for one DBMS may require significant changes to work with another DBMS, which reduces the portability of the application.
3. Repetitive code: Without an ORM, developers may find themselves writing the same type of database operations over and over again, which increases code size and reduces readability.
4. Risk of SQL Injection: When writing manual SQL queries, there is a higher risk of errors leading to vulnerabilities such as SQL injection. Developers must be especially careful about validating and escaping input data.

Thus, NoORM is a powerful approach for those who want complete control over database interactions and optimize the performance of their applications. However, it requires a greater level of knowledge and care on the part of developers.

Big Data Science

07 Jun 2024 17:59

📊💡Dataset of interactions with ChatGPT
Wild Chat is a dataset of 1 million real user interactions with ChatGPT, characterized by a wide range of languages and a variety of prompts.
It was collected by providing free access to everyone to ChatGPT and GPT-4 in exchange for collecting chat history.
Using this dataset, the developers created the Llama-2-based WildLlama-7b-user-assistant bot WildLlama-7b-user-assistant, which is capable of predicting both the user's prompts and the responses that ChatGPT might choose.
The following script can also be used to load a dataset:
from datasets import load_dataset
dataset = load_dataset(“allenai/WildChat-1M”)

Big Data Science

04 Jun 2024 17:59

💡🔎📉Adversarial verification: advantages and disadvantages
Adversarial Verification (AV) is a technique that evaluates a modern test data format based on operational data. This is especially useful in machine learning tasks, where the quality of predictions can matter due to the fact that the data relationship between the strategic and test samples is now clearly visible. Let's look at the main advantages and disadvantages of this situation.
Advantages of adversarial verification:
1. Detection of data inconsistencies:
AV helps identify if production and test data have very different distributions. This may signal dangerous problems with generalization models.
2. Improving the quality of models: By eliminating differences between process and test data, the quality of predictive models in test selection can be significantly improved.
3. Optimization of data selection: by using AV, organizational and validation data sets can be used more accurately, which will avoid overfitting and improve the overall quality of the model.
4. Identification of data leaks: AV helps to identify cases where information from the test sample “leaks” into the operational sample, which can lead to biased results of the model.
Disadvantages of the adversarial test:
1. Increased computational cost: Performing AV requires training additional models (usually a classifier), which increases the computational cost and time required for data analysis.
2. Difficulty in Implementation: Setting up and installing AV can require significant knowledge and experience in machine learning, which can be challenging for beginners.
3. Risk of overfitting: Using AV too often to correct data can lead to overtraining of models on operational data and deterioration of their generalization abilities.