This dataset was converted from https://github.com/openai/prm800k using the following script. import json import os from datasets import Dataset, DatasetDict

def generate_data(data_path: str): with open(data_path, "r", encoding="utf-8") as f: for line in f: data = json.loads(line) yield { "problem": data["problem"], "answer": data["answer"], }

def main(): trainset = Dataset.from_generator(generate_data… See the full description on the dataset page: https://huggingface.co/datasets/hiyouga/math12k.

The IBEM dataset consists of 600 documents with a total number of 8272 pages, containing 29603 isolated and 137089 embedded Mathematical Expressions (MEs). The objective of the IBEM dataset is to facilitate the indexing and searching of MEs in massive collections of STEM documents. The dataset was built by parsing the LaTeX source files of documents from the KDD Cup Collection. Several experiments can be carried out with the IBEM dataset ground-truth (GT): ME detection and extraction, ME recognition, etc.

The dataset consists of the following files:

• “IBEM.json”: file containing the IBEM GT information. The data is firstly organized by pages, then by the type of expression (“embedded” or “displayed”), and lastly by the GT of each individual ME. For each ME we provide:
  • xy page-level coordinates, reported as relative (%) to the width/height of the page image.
  • “split” attribute indicating the number of fragments in which the ME has been split. MEs can be split over various lines, columns or pages. The LaTeX transcript of split MEs have been exactly replicated (entire LaTeX definition) for each fragment.
  • “latex” original transcript as extracted from the LaTeX source files of the documents. This definition can contain user-defined macros. In order to be able to compile these expressions, each page includes the preamble of the source files containing the defined macros and the packages used by the authors of the documents.
  • “latex_expand” transcript reconstructed from the output stream of the LuaLaTeX engine in which user-defined macros have been expanded. The transcript has the same visual representation as the original transcript, with the addition that the LaTeX definitions are tokenized, the order of sub/super script elements have been fixed, and matrices have been transformed to arrays.
  • “latex_norm” transcript resulting from applying an extra normalization process to the “latex_expand” expression. This normalization process includes removing font information such as slant, style, and weight.
• “partitions/*.lst”: files containing list of pages forming the partition sets.
• “pages/*.jpg”: individual pages extracted from the documents.

The dataset is partitioned into various sets as provided for the ICDAR 2021 Competition on Mathematical Formula Detection. The ground-truth related to this competition, which is included in this dataset version, can also be found here. More information about the competition can be found in the following paper:

D. Anitei, J.A. Sánchez, J.M. Fuentes, R. Paredes, and J.M. Benedí. ICDAR 2021 Competition on Mathematical Formula Detection. In ICDAR, pages 783–795, 2021.

For ME recognition tasks, we recommend rendering the “latex_expand” version of the formulae in order to create standalone expressions that have the same visual representation as MEs found in the original documents (see attached python script “extract_GT.py”). Extracting MEs from the documents based on coordinates is more complex, as special care is needed to concatenate the fragments of split expressions. Baseline results for ME recognition tasks will soon be made available.

MetaMath QA

Mathematical Questions for Large Language Models

By Huggingface Hub [source]

About this dataset

This dataset contains meta-mathematics questions and answers collected from the Mistral-7B question-answering system. The responses, types, and queries are all provided in order to help boost the performance of MetaMathQA while maintaining high accuracy. With its well-structured design, this dataset provides users with an efficient way to investigate various aspects of question answering models and further understand how they function. Whether you are a professional or beginner, this dataset is sure to offer invaluable insights into the development of more powerful QA systems!

More Datasets

For more datasets, click here.

Featured Notebooks

• 🚨 Your notebook can be here! 🚨!

How to use the dataset

Data Dictionary

The MetaMathQA dataset contains three columns: response, type, and query. - Response: the response to the query given by the question answering system. (String) - Type: the type of query provided as input to the system. (String) - Query:the question posed to the system for which a response is required. (String)

Preparing data for analysis

It’s important that before you dive into analysis, you first familiarize yourself with what kind data values are present in each column and also check if any preprocessing needs to be done on them such as removing unwanted characters or filling in missing values etc., so that it can be used without any issue while training or testing your model further down in your process flow.

##### Training Models using Mistral 7B

Mistral 7B is an open source framework designed for building machine learning models quickly and easily from tabular (csv) datasets such as those found in this dataset 'MetaMathQA ' . After collecting and preprocessing your dataset accordingly Mistral 7B provides with support for various Machine Learning algorithms like Support Vector Machines (SVM), Logistic Regression , Decision trees etc , allowing one to select from various popular libraries these offered algorithms with powerful overall hyperparameter optimization techniques so soon after selecting algorithm configuration its good practice that one use GridSearchCV & RandomSearchCV methods further tune both optimizations during model building stages . Post selection process one can then go ahead validate performances of selected models through metrics like accuracy score , F1 Metric , Precision Score & Recall Scores .

##### Testing phosphors :

After successful completion building phase right way would be robustly testing phosphors on different evaluation metrics mentioned above Model infusion stage helps here immediately make predictions based on earlier trained model OK auto back new test cases presented by domain experts could hey run quality assurance check again base score metrics mentioned above know asses confidence value post execution HHO updating baseline scores running experiments better preferred methodology AI workflows because Core advantage finally being have relevancy inexactness induced errors altogether impact low

Research Ideas

• Generating natural language processing (NLP) models to better identify patterns and connections between questions, answers, and types.
• Developing understandings on the efficiency of certain language features in producing successful question-answering results for different types of queries.
• Optimizing search algorithms that surface relevant answer results based on types of queries

Acknowledgements

If you use this dataset in your research, please credit the original authors. Data Source

License

License: CC0 1.0 Universal (CC0 1.0) - Public Domain Dedication No Copyright - You can copy, modify, distribute and perform the work, even for commercial purposes, all without asking permission. See Other Information.

Columns

File: train.csv | Column name | Description | |:--------------|:------------------------------------| | response | The response to the query. (String) | | type | The type of query. (String) |

Acknowledgements

If you use this dataset in your research, please credit the original authors. If you use this dataset in your research, please credit Huggingface Hub.

The NaturalProofs Dataset is a large-scale dataset for studying mathematical reasoning in natural language. NaturalProofs consists of roughly 20,000 theorem statements and proofs, 12,500 definitions, and 1,000 additional pages (e.g. axioms, corollaries) derived from ProofWiki, an online compendium of mathematical proofs written by a community of contributors.

The MATHWELL Human Annotation Dataset contains 5,084 synthetic word problems and answers generated by MATHWELL, a reference-free educational grade school math word problem generator released in MATHWELL: Generating Educational Math Word Problems Using Teacher Annotations, and comparison models (GPT-4, GPT-3.5, Llama-2, MAmmoTH, and LLEMMA) with expert human annotations for solvability, accuracy, appropriateness, and meets all criteria (MaC). Solvability means the problem is mathematically possible to solve, accuracy means the Program of Thought (PoT) solution arrives at the correct answer, appropriateness means that the mathematical topic is familiar to a grade school student and the question's context is appropriate for a young learner, and MaC denotes questions which are labeled as solvable, accurate, and appropriate. Null values for accuracy and appropriateness indicate a question labeled as unsolvable, which means it cannot have an accurate solution and is automatically inappropriate. Based on our annotations, 82.2% of the question/answer pairs are solvable, 87.3% have accurate solutions, 78.1% are appropriate, and 58.4% meet all criteria.

This dataset is designed to train text classifiers to automatically label word problem generator outputs for solvability, accuracy, and appropriateness. More details about the dataset can be found in our paper.

This is a filtered and metadata enriched version of open-thoughts/OpenThoughts-114k. While the original dataset is a valuable resource containing DeepSeek-R1 outputs, it has very little metadata (only 2 fields: system and conversations). It does not contain, for instance, the original solution label, which means that we can not verify the model answers.

  What we did

filtered the dataset for math content (math questions were prefixed by "Return your final response within… See the full description on the dataset page: https://huggingface.co/datasets/open-r1/OpenThoughts-114k-math.

Data from a comparative judgement survey consisting of 62 working mathematics educators (ME) at Norwegian universities or city colleges, and 57 working mathematicians at Norwegian universities. A total of 3607 comparisons of which 1780 comparisons by the ME and 1827 ME. The comparative judgement survey consisted of respondents comparing pairs of statements on mathematical definitions compiled from a literature review on mathematical definitions in the mathematics education literature. Each WM was asked to judge 40 pairs of statements with the following question: “As a researcher in mathematics, where your target group is other mathematicians, what is more important about mathematical definitions?” Each ME was asked to judge 41 pairs of statements with the following question: “For a mathematical definition in the context of teaching and learning, what is more important?” The comparative judgement was done with No More Marking software (nomoremarking.com) The data set consists of the following data: comparisons made by ME (ME.csv) comparisons made by WM (WM.csv) Look up table of codes of statements and statement formulations (key.csv) Each line in the comparison represents a comparison, where the "winner" column represents the winner and the "loser" column the loser of the comparison.

This dataset was converted from https://huggingface.co/datasets/BytedTsinghua-SIA/DAPO-Math-17k and https://huggingface.co/datasets/math-ai/aime24 using the following script. from datasets import Dataset, load_dataset,DatasetDict from mathruler.grader import extract_boxed_content

def generate_data_DAPO_Math_17k(data_path): dataset = load_dataset("parquet", data_files=data_path,split="train") dataset=dataset.select(range(17917)) prefix = 'Solve the following math problem step by… See the full description on the dataset page: https://huggingface.co/datasets/Saigyouji-Yuyuko1000/dapo17k.

redefine-math (Xudong Shen)

  General description

In this task, the author tests whether language models are able to work with common symbols when they are redefined to mean something else. The author finds that larger models are more likely to pick the answer corresponding to the original definition rather than the redefined meaning, relative to smaller models. This task demonstrates that it is difficult for language models to work with new information given at inference… See the full description on the dataset page: https://huggingface.co/datasets/inverse-scaling/redefine-math.

Abstract: Graph Neural Networks (GNNs) have recently gained traction in transportation, bioinformatics, language and image processing, but research on their application to supply chain management remains limited. Supply chains are inherently graph-like, making them ideal for GNN methodologies, which can optimize and solve complex problems. The barriers include a lack of proper conceptual foundations, familiarity with graph applications in SCM, and real-world benchmark datasets for GNN-based supply chain research. To address this, we discuss and connect supply chains with graph structures for effective GNN application, providing detailed formulations, examples, mathematical definitions, and task guidelines. Additionally, we present a multi-perspective real-world benchmark dataset from a leading FMCG company in Bangladesh, focusing on supply chain planning. We discuss various supply chain tasks using GNNs and benchmark several state-of-the-art models on homogeneous and heterogeneous graphs across six supply chain analytics tasks. Our analysis shows that GNN-based models consistently outperform statistical ML and other deep learning models by around 10-30% in regression, 10-30% in classification and detection tasks, and 15-40% in anomaly detection tasks on designated metrics. With this work, we lay the groundwork for solving supply chain problems using GNNs, supported by conceptual discussions, methodological insights, and a comprehensive dataset.

import numpy as np import torch from tqdm import tqdm from datasets import load_dataset, DatasetDict, Dataset import datasets import pickle

def get_top_n_docs(scores, n): """Return top-n document indices for a query, ignoring negative scores.""" valid_docs = np.where(scores >= 0)[0] # Filter out negative scores sorted_indices = np.argsort(-scores[valid_docs]) # Descending order top_n_indices = valid_docs[sorted_indices][:n] # Take top n return set(top_n_indices)

def… See the full description on the dataset page: https://huggingface.co/datasets/pxyyy/mix-math-20k-removed-top500-by-mp-for-MATH-Correct-2k.

MathEquiv dataset is accompanied to EquivPruner . It is specifically designed for mathematical statement equivalence , serving as a versatile resource applicable to a variety of mathematical tasks and scenarios. It consists of almost 100k math sentences pair with equivalence result and reasoning step generated by GPT-4O.

The dataset consists of three splits:

train with 77.6k problems for training. test with 9.83k samples for testing. valid with 9.75k samples for validation.

We implemented a five-tiered classification system. This granular approach was adopted to enhance the stability of the GPT model's outputs, as preliminary experiments with binary classification (equivalent/non-equivalent) revealed inconsistencies in judgments. The five-tiered system yielded significantly more consistent and reliable assessments:

Level 4 (Exactly Equivalent): The statements are mathematically interchangeable in all respects, exhibiting identical meaning and form. Level 3 (Likely Equivalent): Minor syntactic differences may be present, but the core mathematical content and logic align. Level 2 (Indeterminable): Insufficient information is available to make a definitive judgment regarding equivalence. Level 1 (Unlikely Equivalent): While some partial agreement may exist, critical discrepancies in logic, definition, or mathematical structure are observed. Level 0 (Not Equivalent): The statements are fundamentally distinct in their mathematical meaning, derivation, or resultant outcomes.

This dataset tracks annual math proficiency from 2015 to 2022 for Sadler Means Ywla vs. Texas and Austin Independent School District

CONTEXT

Practice Scenario: The UIW School of Engineering wants to recruit more students into their program. They will recruit students with great math scores. Also, to increase the chances of recruitment,​ the department will look for students who qualify for financial aid. Students who qualify for financial aid more than likely come from low socio-economic backgrounds. One way to indicate this is to view how much federal revenue a school district receives through its state. High federal revenue for a school indicates that a large portion of the student base comes from low incomes families.

The question we wish to ask is as follows: Name the school districts across the nation where their Child Nutrition Programs(c25) are federally funded between the amounts $30,000 and $50,000. And where the average math score for the school districts corresponding state is greater than or equal to the nations average score of 282.

The SQL query below in 'Top5MathTarget.sql' can be used to answer this question in MySQL. To execute this process, one would need to install MySQL to their local system and load the attached datasets below from Kaggle into their MySQL schema. The SQL query below will then join the separate tables on various key identifiers.

DATA SOURCE Data is sourced from The U.S Census Bureau and The Nations Report Card (using the NAEP Data Explorer).

Finance: https://www.census.gov/programs-surveys/school-finances/data/tables.html

Math Scores: https://www.nationsreportcard.gov/ndecore/xplore/NDE

COLUMN NOTES

All data comes from the school year 2017. Individual schools are not represented, only school districts within each state.

FEDERAL FINANCE DATA DEFINITIONS

t_fed_rev: Total federal revenue through the state to each school district.

C14- Federal revenue through the state- Title 1 (no child left behind act).

C25- Federal revenue through the state- Child Nutrition Act.

Title 1 is a program implemented in schools to help raise academic achievement ​for all students. The program is available to schools where at least 40% of the students come from low inccom​​e families.

Child Nutrition Programs ensure the children are getting the food they need to grow and learn. Schools with high federal revenue to these programs indicate students that also come from low income​ families.

MATH SCORES DATA DEFINITIONS

Note: Mathematics, Grade 8, 2017, All Students (Total)

average_scale_score - The state's average score for eighth graders taking the NAEP math exam.

Confusion matrix of K-Means clustering results on dataset 6.

Coq-HoTT Q&A Dataset

  Dataset Description

The Coq-HoTT Q&A Dataset is a conversational extension of the Coq-HoTT Dataset, derived directly from the Coq-HoTT GitHub repository (https://github.com/HoTT/Coq-HoTT). This dataset transforms Homotopy Type Theory (HoTT) content into structured Q&A pairs, bridging the gap between formal mathematics and conversational AI. Each entry in the dataset represents a mathematical statement, such as a definition or theorem, converted into a… See the full description on the dataset page: https://huggingface.co/datasets/phanerozoic/Coq-HoTT-QA.

UniMath Q&A Dataset

  Dataset Description

The UniMath Q&A Dataset is a conversational extension of the UniMath Dataset, derived from the UniMath formalization of mathematics (https://github.com/UniMath/UniMath). This dataset transforms Univalent Mathematics content into structured Q&A pairs, making formal mathematical content more accessible through natural language interactions. Each entry represents a mathematical statement from UniMath (definition, theorem, lemma, etc.)… See the full description on the dataset page: https://huggingface.co/datasets/phanerozoic/Coq-UniMath-QA.

Dream EEG and Mentation (DREAM) data set

Data set information

• Common name: TWC_USA
• Full name: Two-way communicaton data from USA team
• Authors: Karen R. Konkoly, Remington Mallett, Ken A. Paller
• Location: Northwestern University
• Year: 2021
• Set ID: 4
• Amendment: 1
• Corresponding author ID: 4

Previous publications:
Konkoly, K. R., Appel, K., Chabani, E., Mangiaruga, A., Gott, J., Mallett, R., ... & Paller, K. A. (2021). Real-time dialogue between experimenters and dreamers during REM sleep. Current Biology, 31(7), 1417-1427.

Correspondence:
karenkonkoly2023@u.northwestern.edu

Metadata

• Key ID: 5
• Date entered: 2023-02-08T03:09:10+00:00
• Number of samples: 33
• Number of subjects: 19
• Proportion REM: 61%
• Proportion N1: 18%
• Proportion N2: 18%
• Proportion W: 0%
• Proportion experience: 82%
• Proportion no-experience: 15%
• Proportion healthy: 100%
• Provoked awakening: Some
• Time of awakening: Mixed
• Form of response: Structured
• Date approved: 2023-02-09T05:50:12+00:00

How to decode data files

• L-MSTD is an electrode on the left mastoid for if a back-up reference channel was needed. The EMG channel is on the chin, and channels 26 and 27 are back-up EMG electrodes located nearby on the chin. On some recordings EMG-2 is the back-up EMG channel instead, which is also located nearby on the chin.
• The "status" channel was created when converting the data to EDF, and contains information about port codes in the data set. Disregard that the port codes are expressed in microvolts. More information about the meanings of the port codes below.
• It may be that there is a second of flat EEG at the end of each recording which appears to be an artifact of converting the file type to .EDF and should be disregarded

• The time of awakening column contains only approximated times based on experimenters' notes and the duration of files

• There are port codes in the data that have slightly meanings for some different participants (in the "status" channel). Here is a guide for their meanings:

  • 32 just indicates that a new script was started (no sounds played)
  • 64 means the volume was turned down (no sounds played)
  • 65 means the volume was turned up (no sounds played)
  • 29 means a TLR auditory cue was presented
  • 23 means a TLR light cue was presented (but code and light cue are triggered manually, so time-locked analyses is not possible here)
  • Codes 1-20 correspond to math problems that were presented during sleep, and were changed after case 08. See below for guide

  Cases 01-08

  • 1: 9-7
  • 2: 3+2
  • 3: 14-13
  • 4: 6+1
  • 5: 19-16
  • 6: 1+1
  • 7: 5-2
  • 8: 1+4
  • 9: 15-10
  • 10: 3+3
  • 11: 8-4
  • 12: 2+2
  • 13: 8-0
  • 14: 4+1
  • 15: 14-13
  • 16: 2+4
  • 17: 16-13
  • 18: 3+1
  • 19: 10-8
  • 20: 1+0

  Cases 09-33

  • 1: 9-7
  • 2: 3+1
  • 3: 8-7
  • 4: 1+2
  • 5: 9-6
  • 6: 1+1
  • 7: 5-2
  • 8: 4-1
  • 9: 8-6
  • 10: 8-5
  • 11: 2+2
  • 12: 2+1
  • 13: 3+0
  • 14: 1+0
  • 15: 7-4
  • 16: 2+0
  • 17: 6-3
  • 18: 3-1
  • 19: 5-4
  • 20: 1+0

Treatment group codes

N/A

Experimental description

Methods:

Twenty-two participants (15 female, age range 18-33 years, M = 21.1 ± 4.3 years) who claimed to remember at least one dream per week were recruited by word of mouth, online forum, and the Northwestern University Psychology Department participant pool. They each participated in one or more nap sessions, which amounted to 27 nap sessions in total.

Procedure:

Participants visited the laboratory at Northwestern University at approximately their normal wake time and received guidance on identifying lucid dreams and instructions for the experiment for about 40 min during preparations for polysomnographic recordings, including EEG, EMG, and EOG, using a Neuroscan SynAmps system. Participants were instructed to signal with a prearranged number of LR eye movements if they became lucid in a dream.

Next, participants practiced making ocular signals and responding to questions using combinations of LR signals. Subsequently, participants completed the Targeted Lucidity Reactivation (TLR) procedure while lying in bed. This procedure was derived from the procedure developed by Carr and collegues. A method of reality checking to induce lucid dreaming was paired with sensory stimulation and accelerated in a single session immediately before sleep, and then cues were presented again during REM sleep. In this procedure, participants were trained to associate a novel cue sound with a lucid state of mind during wake. The sound consisted of three pure-tone beeps increasing in pitch (400, 600, and 800 Hz) at approximately 40-45 dB SPL and lasting approximately 650 ms. For one participant (ppt. 121) the pure-tone beeps had previously been associated with a different task in an unrelated study. Thus, for this participant, a 1000-ms violin sound and low-intensity flashing-red LED lights were used as cues. All participants were informed that this cue would be given during sleep to help promote a lucid dream. Over the next 15 min, the TLR sound was played up to 15 times. The first 4 times, it was followed by verbal guidance to enter a lucid state as follows. ‘‘As you notice the signal, you become lucid. Bring your attention to your thoughts and notice where your mind has wandered.[pause] Now observe your body, sensations, and feelings.[pause] Observe your breathing. [pause] Remain lucid, critically aware, and notice how aspects of this experience are in any way different from your normal waking experience.’’

Participants often fell asleep before all 15 TLR cue presentations were completed. Standard polysomnographic methods were used to determine sleep state. Once participants entered REM sleep, TLR cues were presented again, at about 30-s intervals, as long as REM sleep remained stable. After participants responded to a cue with a lucid eye signal, or after approximately 10 cues were presented without response, we began the math problem portion of the experiment.

We devised the following task to engage auditory perception of math problems, working memory, and the ability to express the correct answer. We used simple addition and subtraction problems that could each be answered by a number between 1 and 4 (LR = 1, LRLR = 2, LRLRLR = 3, LRLRLRLR = 4), or between 1 and 6 for the first 5 participants.

From the above dataset, data was included in DREAM if there was a period of sleep on the EEG followed by a report of a dream (or a lack of dream). The EEG data includes the last period of continuous sleep before the dream report was collected, starting with the first epoch scored as wake, and ending at the last second before clear movement/alpha activity indicating wake. Also, there are a few instances, noted in the “Remarks” column in the “Records” file, where I included epochs that were scored as wake, when the wake seemed due to alpha from participants attempting to answer questions with eye movements (only one of these included wake in the last 20 seconds of the recording, case21_sub111).

EEG sleep data was NOT included if it was not followed by a verbal/written dream report or clear note on the experimenter’s log that there was no recall. Also not included is data where participants completed eye signals or answered questions, but it was not part of the continuous period of sleep before a dream report was given. Also excluded was a case in which a dream report was collected at the end of the nap but the participant had been in and out of sleep beforehand, so it was unclear which sleep period the report referred to.

DREAM categorization procedure

Karen Konkoly rated reports according to the DREAM categorization. If the participant reported remembering any sort of mental content from sleep, it was rated “2”. If the participant reported remembering a dream but none of its content, it was rated “1”. If the participant reported not remembering anything, or not falling asleep, it was rated “0”.

Zhang et al. (https://link.springer.com/article/10.1140/epjb/e2017-80122-8) suggest a temporal random network with changing dynamics that follow a Markov process, allowing for a continuous-time network history moving from a static definition of a random graph with a fixed number of nodes n and edge probability p to a temporal one. Defining lambda = probability per time granule of a new edge to appear and mu = probability per time granule of an existing edge to disappear, Zhang et al. show that the equilibrium probability of an edge is p=lambda/(lambda+mu) Our implementation, a Python package that we refer to as RandomDynamicGraph https://github.com/ScanLab-ossi/DynamicRandomGraphs, generates large-scale dynamic random graphs according to the defined density. The package focuses on massive data generation; it uses efficient math calculations, writes to file instead of in-memory when datasets are too large, and supports multi-processing. Please note the datetime is arbitrary.

Análisis de los niveles de motivación en estudiantes de secundaria hacia el aprendizaje de las matemáticas.