# Data Science

**URL:** https://forum.numer.ai/c/data-science/5.md

[Latest](https://forum.numer.ai/latest.md) · [Categories](https://forum.numer.ai/categories.md)

---

## [About the Data Science category](https://forum.numer.ai/t/about-the-data-science-category/19)

<div class="topic-metadata">

**Author:** [@pschork](https://forum.numer.ai/u/pschork)\
**Replies:** 6\
**Last updated:** [January 11, 2025, 9:56am UTC](https://forum.numer.ai/t/about-the-data-science-category/19 "2025-01-11T09:56:37Z")

</div>

Discussions on data science

---

## [Xgboost vector leaf output - two targets experiment](https://forum.numer.ai/t/xgboost-vector-leaf-output-two-targets-experiment/8355)

<div class="topic-metadata">

**Author:** [@aininja](https://forum.numer.ai/u/aininja)\
**Replies:** 0\
**Last updated:** [September 2, 2026, 1:37pm UTC](https://forum.numer.ai/t/xgboost-vector-leaf-output-two-targets-experiment/8355 "2026-09-02T13:37:04Z")

</div>

Recently @jrb posted on Discord link to Xgboost article about Vector-Leaf model and I thought it was worth trying. What is Vector-Leaf model? So with Xgboost 3.4.1 you have an option to have one tree with multiple output…

---

## [I built a small DL helper package](https://forum.numer.ai/t/i-built-a-small-dl-helper-package/8282)

<div class="topic-metadata">

**Author:** [@klaes\_ashford](https://forum.numer.ai/u/klaes_ashford)\
**Replies:** 0\
**Last updated:** [April 24, 2026, 4:51pm UTC](https://forum.numer.ai/t/i-built-a-small-dl-helper-package/8282 "2026-04-24T16:51:03Z")

</div>

Built a tiny utility from my own deep learning experimentation workflow and figured others here might find it useful. I kept getting messy training logs during iterative experiments, so I made epoch-printer — a lightwei…

---

## [Distribution shifts](https://forum.numer.ai/t/distribution-shifts/8270)

<div class="topic-metadata">

**Author:** [@anonemaus](https://forum.numer.ai/u/anonemaus)\
**Replies:** 0\
**Last updated:** [April 3, 2026, 7:31pm UTC](https://forum.numer.ai/t/distribution-shifts/8270 "2026-04-03T19:31:03Z")

</div>

The discussion about old models no longer working led me to thinking about a quick analysis of distribution shifts. There are libraries out there for this kind of thing, so maybe there are better ways of doing this. As y…

---

## [Foundation Models on Numerai Data: TabPFN v2.5 and TabICL v2](https://forum.numer.ai/t/foundation-models-on-numerai-data-tabpfn-v2-5-and-tabicl-v2/8269)

<div class="topic-metadata">

**Author:** [@degerhan](https://forum.numer.ai/u/degerhan)\
**Replies:** 0\
**Last updated:** [March 30, 2026, 3:07pm UTC](https://forum.numer.ai/t/foundation-models-on-numerai-data-tabpfn-v2-5-and-tabicl-v2/8269 "2026-03-30T15:07:57Z")

</div>

In the quest for mmc, I wanted to experiment with TabPFN v2.5 and TabICL v2, foundation models for tabular data and about as different from gradient-boosted trees as you can get. Neither works at Numerai scale, so the i…

---

## [How often do you re-train your model?](https://forum.numer.ai/t/how-often-do-you-re-train-your-model/8164)

<div class="topic-metadata">

**Author:** [@spn30n](https://forum.numer.ai/u/spn30n)\
**Replies:** 8\
**Last updated:** [March 21, 2026, 9:40pm UTC](https://forum.numer.ai/t/how-often-do-you-re-train-your-model/8164 "2026-03-21T21:40:28Z")

</div>

My model performed well on the first few weeks, and then the performance degraded. I wonder how often should we re-train our model to incorporate the new data.

---

## [AI for ML by jefferythewind](https://forum.numer.ai/t/ai-for-ml-by-jefferythewind/8245)

<div class="topic-metadata">

**Author:** [@unsentient](https://forum.numer.ai/u/unsentient)\
**Replies:** 0\
**Last updated:** [February 14, 2026, 9:04pm UTC](https://forum.numer.ai/t/ai-for-ml-by-jefferythewind/8245 "2026-02-14T21:04:04Z")

</div>

Based on a presentation given at the January 2026 Numerai Symposium by @jefferythewind Restated and posted with permission. The Problem: Convergence in Quantitative Finance If you are a data scientist or a participant i…

---

## [What World Are Our Models Building?](https://forum.numer.ai/t/what-world-are-our-models-building/8240)

<div class="topic-metadata">

**Author:** [@perseverance](https://forum.numer.ai/u/perseverance)\
**Replies:** 2\
**Last updated:** [February 11, 2026, 2:24pm UTC](https://forum.numer.ai/t/what-world-are-our-models-building/8240 "2026-02-11T14:24:04Z")

</div>

Hey Numerai community, I’ve been loving the tournament, it’s super fun and mathematically challenging. But as we pour our expertise into these micro-predictions that power the hedge fund’s meta-model, I’ve started wonde…

---

## [Vibesciencing my way through v5.2 data (Faith II)](https://forum.numer.ai/t/vibesciencing-my-way-through-v5-2-data-faith-ii/8214)

<div class="topic-metadata">

**Author:** [@degerhan](https://forum.numer.ai/u/degerhan)\
**Replies:** 1\
**Last updated:** [February 4, 2026, 2:31pm UTC](https://forum.numer.ai/t/vibesciencing-my-way-through-v5-2-data-faith-ii/8214 "2026-02-04T14:31:09Z")

</div>

Four things collided during the long Christmas weekend: My building anxiety over not sciencing the numerai data for a year Faith II v5.2 changes everything (yet again) Codex CLI with gpt-52-xhigh is a beast. It will no…

---

## [Fine-Tuning LLMs with RL for ML Code Generation: Post-Mortem](https://forum.numer.ai/t/fine-tuning-llms-with-rl-for-ml-code-generation-post-mortem/8235)

<div class="topic-metadata">

**Author:** [@jefferythewind](https://forum.numer.ai/u/jefferythewind)\
**Replies:** 0\
**Last updated:** [January 29, 2026, 4:44am UTC](https://forum.numer.ai/t/fine-tuning-llms-with-rl-for-ml-code-generation-post-mortem/8235 "2026-01-29T04:44:39Z")

</div>

Following up from my talk at the Numerai Symposium today - thanks to everyone who attended and especially those who asked great questions afterward. The feedback inspired me to dig into what the LLM was actually writing …

---

## [200 Trees Per Second with Cuda](https://forum.numer.ai/t/200-trees-per-second-with-cuda/7823)

<div class="topic-metadata">

**Author:** [@murkyautomata](https://forum.numer.ai/u/murkyautomata)\
**Replies:** 12\
**Last updated:** [December 30, 2025, 4:35pm UTC](https://forum.numer.ai/t/200-trees-per-second-with-cuda/7823 "2025-12-30T16:35:44Z")

</div>

You can build 200 trees per second on v5-all with my cuda extra tree booster (on an A100,) and it can fit a dataset several times larger. https://drive.google.com/file/d/1EU0a4mMLyUBCBLdjIS3\_qEM85bHERtKi/view?usp=sharin…

---

## [What can explain so many models getting same score?](https://forum.numer.ai/t/what-can-explain-so-many-models-getting-same-score/7094)

<div class="topic-metadata">

**Author:** [@coffee\_ai](https://forum.numer.ai/u/coffee_ai)\
**Replies:** 9\
**Last updated:** [November 14, 2025, 5:37am UTC](https://forum.numer.ai/t/what-can-explain-so-many-models-getting-same-score/7094 "2025-11-14T05:37:22Z")

</div>

Are there example models that many people are being submitted on? Very high frequency of exact same score.

---

## [Multi-Factor Risk Modeling for Cryptocurrency Price Prediction](https://forum.numer.ai/t/multi-factor-risk-modeling-for-cryptocurrency-price-prediction/7685)

<div class="topic-metadata">

**Author:** [@marianacu](https://forum.numer.ai/u/marianacu)\
**Replies:** 5\
**Last updated:** [August 27, 2024, 2:36pm UTC](https://forum.numer.ai/t/multi-factor-risk-modeling-for-cryptocurrency-price-prediction/7685 "2024-08-27T14:36:17Z")

</div>

In this analysis, I developed and evaluated a multi-factor risk model to predict cryptocurrency price movements. The model incorporates a wide range of features sourced from diverse datasets, including a 100-token datase…

---

## [Implications of heavily correlated MMC and CORR?](https://forum.numer.ai/t/implications-of-heavily-correlated-mmc-and-corr/8072)

<div class="topic-metadata">

**Author:** [@pharus](https://forum.numer.ai/u/pharus)\
**Replies:** 3\
**Last updated:** [June 5, 2025, 4:07pm UTC](https://forum.numer.ai/t/implications-of-heavily-correlated-mmc-and-corr/8072 "2025-06-05T16:07:54Z")

</div>

Does this imply the prediction is ‘unique’? Maybe even random/irrelevant? Is this model worth examining more? Love to hear any thoughts about this.

---

## [DANCR R&D AI agents, running locally](https://forum.numer.ai/t/dancr-r-d-ai-agents-running-locally/8030)

<div class="topic-metadata">

**Author:** [@noptus](https://forum.numer.ai/u/noptus)\
**Replies:** 1\
**Last updated:** [May 6, 2025, 4:02pm UTC](https://forum.numer.ai/t/dancr-r-d-ai-agents-running-locally/8030 "2025-05-06T16:02:53Z")

</div>

I introduce DANCR, a self-driving research pipeline that continuously generates, prioritizes, and executes modeling strategies for Numerai no manual intervention required. I built DANCR to leverage LM Studio’s local L…

---

## [New DataScientist on board - Where do I start?](https://forum.numer.ai/t/new-datascientist-on-board-where-do-i-start/6762)

<div class="topic-metadata">

**Author:** [@nolyzlel](https://forum.numer.ai/u/nolyzlel)\
**Replies:** 3\
**Last updated:** [April 15, 2025, 5:15am UTC](https://forum.numer.ai/t/new-datascientist-on-board-where-do-i-start/6762 "2025-04-15T05:15:56Z")

</div>

Hello dear community. I am completely new to the topic of stock market prediction. I understand the concept of Numerai and have worked as a Data Scientist for a tech company for multiple years. My specialties are RNNs,…

---

## [GFlowNets for Signal Miner: A New Way to Find Diverse, High-Performing Models](https://forum.numer.ai/t/gflownets-for-signal-miner-a-new-way-to-find-diverse-high-performing-models/7966)

<div class="topic-metadata">

**Author:** [@jefferythewind](https://forum.numer.ai/u/jefferythewind)\
**Replies:** 1\
**Last updated:** [March 20, 2025, 3:52pm UTC](https://forum.numer.ai/t/gflownets-for-signal-miner-a-new-way-to-find-diverse-high-performing-models/7966 "2025-03-20T15:52:37Z")

</div>

Hey everyone, I wanted to share something exciting I’ve been working on that could significantly improve the search for diverse, high-performing models in the Numerai ecosystem. If you’ve been using Signal Miner (GitHub)…

---

## [Modular Defi Crypto Portfolio Management System](https://forum.numer.ai/t/modular-defi-crypto-portfolio-management-system/7887)

<div class="topic-metadata">

**Author:** [@jefferythewind](https://forum.numer.ai/u/jefferythewind)\
**Replies:** 2\
**Last updated:** [December 18, 2024, 10:23pm UTC](https://forum.numer.ai/t/modular-defi-crypto-portfolio-management-system/7887 "2024-12-18T22:23:57Z")

</div>

OK Party People, here it is. An open source, modular system design, to implement decentralized portfolios based on the Numerai crypto meta model, but the generic oracles could really be based on anything. For all the mo…

---

## [Super Massive Data Release: Deep Dive](https://forum.numer.ai/t/super-massive-data-release-deep-dive/4053)

<div class="topic-metadata">

**Author:** [@master\_key](https://forum.numer.ai/u/master_key)\
**Replies:** 81\
**Last updated:** [November 22, 2021, 3:13pm UTC](https://forum.numer.ai/t/super-massive-data-release-deep-dive/4053 "2021-11-22T15:13:03Z")

</div>

Highlights We have just released the biggest upgrade to Numerai’s dataset ever. The new dataset has 4x the number of rows, more than 3x the number of features, and 20 optional targets. The fastest way to get started wit…

---

## [Crypto Price Dynamics](https://forum.numer.ai/t/crypto-price-dynamics/7677)

<div class="topic-metadata">

**Author:** [@whiterider](https://forum.numer.ai/u/whiterider)\
**Replies:** 6\
**Last updated:** [October 18, 2024, 1:30pm UTC](https://forum.numer.ai/t/crypto-price-dynamics/7677 "2024-10-18T13:30:11Z")

</div>

The report below outlines a multi-factor model developed to predict cryptocurrency prices by integrating market data, sentiment analysis, and macroeconomic indicators. The work is part of the Ocean Protocol competition h…

---

## [Jane Street Real-Time Market Data Forecasting Kaggle — Really similar format to Numerai’s](https://forum.numer.ai/t/jane-street-real-time-market-data-forecasting-kaggle-really-similar-format-to-numerai-s/7841)

<div class="topic-metadata">

**Author:** [@linkus](https://forum.numer.ai/u/linkus)\
**Replies:** 0\
**Last updated:** [November 11, 2024, 8:31pm UTC](https://forum.numer.ai/t/jane-street-real-time-market-data-forecasting-kaggle-really-similar-format-to-numerai-s/7841 "2024-11-11T20:31:05Z")

</div>

Maybe can even try using your model from Numerai / Signals on this :stuck\_out\_tongue:

---

## [Numerai Crypto TB100 and Statistical Risk Model](https://forum.numer.ai/t/numerai-crypto-tb100-and-statistical-risk-model/7835)

<div class="topic-metadata">

**Author:** [@jefferythewind](https://forum.numer.ai/u/jefferythewind)\
**Replies:** 7\
**Last updated:** [November 8, 2024, 8:53pm UTC](https://forum.numer.ai/t/numerai-crypto-tb100-and-statistical-risk-model/7835 "2024-11-08T20:53:47Z")

</div>

Introduction Following up on some exploratory work on simulating portfolios built from Numerai’s crypto meta model, I wanted to investigate a statistical risk factor model for the crypto market and apply the model to mo…

---

## [Suggestions for undergraduate of quantitative finance](https://forum.numer.ai/t/suggestions-for-undergraduate-of-quantitative-finance/7070)

<div class="topic-metadata">

**Author:** [@autratec](https://forum.numer.ai/u/autratec)\
**Replies:** 4\
**Last updated:** [September 17, 2024, 8:48am UTC](https://forum.numer.ai/t/suggestions-for-undergraduate-of-quantitative-finance/7070 "2024-09-17T08:48:03Z")

</div>

Hi all, I have a friend who recently enrolled in an undergraduate quantitative finance program. In order to secure a better job in the current market, he needs some suggestions for additional free online courses he can t…

---

## [Modeling Cryptocurrency Factors For Price Returns](https://forum.numer.ai/t/modeling-cryptocurrency-factors-for-price-returns/7701)

<div class="topic-metadata">

**Author:** [@okpo](https://forum.numer.ai/u/okpo)\
**Replies:** 2\
**Last updated:** [September 10, 2024, 10:05am UTC](https://forum.numer.ai/t/modeling-cryptocurrency-factors-for-price-returns/7701 "2024-09-10T10:05:50Z")

</div>

Hi everyone, this is my report for the Crypto Factor Modeling Data Challenge. Please read and share your opinion. Thanks. Crypto Factor Modeling Report by Okpo

---

## [Crypto Factor model - Prediction the return of Bitcoin and Crypto with High Trading Volume](https://forum.numer.ai/t/crypto-factor-model-prediction-the-return-of-bitcoin-and-crypto-with-high-trading-volume/7672)

<div class="topic-metadata">

**Author:** [@nishimoto](https://forum.numer.ai/u/nishimoto)\
**Replies:** 6\
**Last updated:** [August 27, 2024, 11:46am UTC](https://forum.numer.ai/t/crypto-factor-model-prediction-the-return-of-bitcoin-and-crypto-with-high-trading-volume/7672 "2024-08-27T11:46:42Z")

</div>

This is my entry for the Numerai/Ocean competition. All pdf: Crypto Factor model - Prediction the return of Bitcoin and Crypto with High Trading Volume.pdf - Google Drive Notebook: desights-ai-crypto-factor-model.ipynb…

---

## [Factor Analysis of Cryptocurrency Returns Using Momentum Indicators](https://forum.numer.ai/t/factor-analysis-of-cryptocurrency-returns-using-momentum-indicators/7667)

<div class="topic-metadata">

**Author:** [@accountnumber1](https://forum.numer.ai/u/accountnumber1)\
**Replies:** 8\
**Last updated:** [August 30, 2024, 5:51am UTC](https://forum.numer.ai/t/factor-analysis-of-cryptocurrency-returns-using-momentum-indicators/7667 "2024-08-30T05:51:07Z")

</div>

This is my entry for the Numerai/Ocean competition. Hope you enjoy. Let me know if there are any questions! Factor Analysis of Cryptocurrency Returns Using Momentum Indicators

---

## [A Detailed Case Study on Crypto Multi-factor Risk Analysis](https://forum.numer.ai/t/a-detailed-case-study-on-crypto-multi-factor-risk-analysis/7682)

<div class="topic-metadata">

**Author:** [@datahunter](https://forum.numer.ai/u/datahunter)\
**Replies:** 3\
**Last updated:** [August 27, 2024, 8:35pm UTC](https://forum.numer.ai/t/a-detailed-case-study-on-crypto-multi-factor-risk-analysis/7682 "2024-08-27T20:35:37Z")

</div>

Introduction In recent years, cryptocurrencies have garnered significant attention from researchers and inventors, among other financial investments. Many investment firms have been investing in and maintaining a strong…

---

## [Crypto Factor Modeling for evaluating crypto return](https://forum.numer.ai/t/crypto-factor-modeling-for-evaluating-crypto-return/7688)

<div class="topic-metadata">

**Author:** [@mlh\_alavi](https://forum.numer.ai/u/mlh_alavi)\
**Replies:** 8\
**Last updated:** [August 27, 2024, 8:28pm UTC](https://forum.numer.ai/t/crypto-factor-modeling-for-evaluating-crypto-return/7688 "2024-08-27T20:28:15Z")

</div>

Introduction Crypto factor modeling is an approach used to analyze and predict the return of cryptocurrencies by identifying and assessing various factors that influence their returns. These factors can include market t…

---

## [Introducing the Crypto Price Variance Model (CPVM)](https://forum.numer.ai/t/introducing-the-crypto-price-variance-model-cpvm/7690)

<div class="topic-metadata">

**Author:** [@polymawutor](https://forum.numer.ai/u/polymawutor)\
**Replies:** 1\
**Last updated:** [August 27, 2024, 2:36pm UTC](https://forum.numer.ai/t/introducing-the-crypto-price-variance-model-cpvm/7690 "2024-08-27T14:36:54Z")

</div>

Abstract We introduce the Crypto Price Variance Model (CPVM), a novel approach to understanding and predicting cryptocurrency price movements. Building upon the foundation of the Fama-French model, the CPVM incorporat…

---

## [Multi-Factor-Risk-Models-for-Cryptocurrency-Price-Variance](https://forum.numer.ai/t/multi-factor-risk-models-for-cryptocurrency-price-variance/7673)

<div class="topic-metadata">

**Author:** [@yunusgumussoy](https://forum.numer.ai/u/yunusgumussoy)\
**Replies:** 0\
**Last updated:** [August 26, 2024, 2:34pm UTC](https://forum.numer.ai/t/multi-factor-risk-models-for-cryptocurrency-price-variance/7673 "2024-08-26T14:34:23Z")

</div>

Hello everyone, I’m excited to share our work on forecasting Bitcoin prices using a variety of advanced techniques. Over the past few weeks, we have been diving deep into time series analysis, model selection, and featu…

[Next page](https://forum.numer.ai/c/data-science/5.md?page=1)
