DavidVivancos/NeuraxonLife2.5-100K-TimeSeries

NeuraxonLife2.5-100K-DeepTimeSeries: Artificial Life Neuraxon Neural Network Simulation Deep Time Series Dataset

Dataset Description

The NeuraxonLife 2.5 Deep Time Series Dataset is a massive, comprehensive collection of simulation data from the Neuraxon Game of Life environment. It tracks the evolution of over 100,000 autonomous agents ("NxErs") evolving biologically-plausible neural networks under survival pressures.

This dataset represents a significant expansion over previous versions, featuring "Deep Time Series" exploration with over 279 million plasticity events and high-resolution per-agent time series data. It is designed for validating the Neuraxon paper: 'A New Neural Growth & Computation Blueprint' by David Vivancos https://www.vivancos.com/ & Dr. Jose Sanchez https://josesanchezgarcia.com/ for Qubic Science. https://qubic.org/

Paper Reference: Neuraxon: A New Neural Growth & Computation Blueprint

Dataset Summary

This dataset provides granular insights into emergent neural computation in an artificial life setting. It covers 2,791 distinct simulation games involving complex neural phenomena.

Key Features:

• Massive Scale: Data from 2,791 games, tracking 100,828 unique agents (NxErs), over 10 independent Sessions with about 270+ Rounds each.
• Deep Plasticity: Captures ~279 million synaptic plasticity events (LTP/LTD) and ~278 million associativity events.
• Multi-Scale Time Series:
  • Global: Network-wide metrics per tick.
  • Per-NxEr: Individual agent metrics (energy, neurotransmitters, branching ratios) tracked over time (162 million rows).
• Evolutionary Lineage: New clan_history and hall_of_fame tables tracking lineage merging and ancestral success.
• Neuromodulation: Detailed tracking of Dopamine, Serotonin, Acetylcholine, and Norepinephrine dynamics and threshold modulations.

Processing Stats:

• Games Processed: 2,791
• Total Neurons: ~3.76 Million
• Total Synapses: ~17 Million
• Total Event Rows: >1 Billion combined events

Supported Tasks

• Neural Dynamics Analysis: Study phase coherence, branching ratios, and criticality over millions of ticks.
• Plasticity & Learning: Analyze STDP, cooperative associativity, and synaptic weight evolution across fast, slow, and meta timescales.
• Evolutionary Biology: Track clan formation, lineage survival, and the correlation between neural architecture and fitness.
• Neuromodulation Research: Investigate how neurotransmitter gradients influence global network states and agent behavior.
• Behavioral Correlation: Link input/output patterns (vision/movement) to internal neural states and survival outcomes.

Dataset Structure

The dataset consists of 28 interconnected Parquet tables.

/
├── games.parquet                       # Global game metadata and summary stats
├── nxers.parquet                       # Agent config, stats, and sensory setup
├── network_params.parquet              # Hyperparameters and architecture per agent
├── neurons.parquet                     # Neuron snapshots (membrane potential, health)
├── synapses.parquet                    # Synaptic weights, types, and integrity
├── time_series.parquet                 # Global (Game-level) time series metrics
├── per_nxer_time_series.parquet        # Individual (Agent-level) time series metrics (HUGE)
├── plasticity_events.parquet           # LTP/LTD specific events
├── weight_evolution_events.parquet     # Weight deltas across timescales
├── associativity_events.parquet        # Cooperative LTP events
├── neuromodulator_events.parquet       # Threshold crossing events
├── threshold_modulation_events.parquet # ACh/Autoreceptor threshold impacts
├── subthreshold_events.parquet         # Near-threshold integration data
├── autoreceptor_events.parquet         # Autoreceptor specific data
├── homeostatic_events.parquet          # Homeostatic threshold adjustments
├── adaptive_threshold_events.parquet   # Criticality adjustments
├── dendritic_events.parquet            # Dendritic spikes and plateau potentials
├── phase_events.parquet                # Phase coherence and reset events
├── spontaneous_events.parquet          # Spontaneous firing logs
├── silent_synapse_events.parquet       # Synapse activation/deactivation
├── nxer_events.parquet                 # Life events (Birth, Death, Mating)
├── clan_history.parquet                # Lineage and clan merging history
├── hall_of_fame.parquet                # Top ranked agents
├── io_patterns.parquet                 # Input/Output vectors
├── itu_fitness.parquet                 # ITU/Aigarth fitness tracking
├── synapse_neighbor_ids.parquet        # Synaptic topology
├── neuron_state_history.parquet        # Recent state buffers
├── nxer_visited_history.parquet        # Spatial exploration history
├── foods.parquet                       # Food source locations
├── food_progress.parquet               # Harvest trackers
├── world_grids.parquet                 # Terrain maps
└── manifest.json                       # Dataset metadata

Data Tables Detail

1. Global & Agent Metadata

• games.parquet (2,791 rows)
  • Description: High-level summary of every simulation run.
  • Key Columns: game_id, round_number, total_ticks, peak_network_activity, average_branching_ratio, peak_dopamine, total_plasticity_events, clan_count.
• nxers.parquet (100,828 rows)
  • Description: Static data and final stats for every agent (NxEr) born.
  • Key Columns: nxer_id, clan_id, generation, ancestors_count, stats_fitness_score, stats_temporal_sync_score, born_ts, died_ts.
• network_params.parquet (100,828 rows)
  • Description: The genomic/hyperparameter configuration for each agent's brain.
  • Key Columns: num_hidden_neurons, connection_probability, learning_rate, stdp_window, dopamine_baseline, oscialltor_frequencies, plasticity_energy_cost.
• hall_of_fame.parquet (83,609 rows)
  • Description: Rankings of the top agents by various categories (Survival, Exploration, Efficiency).
  • Key Columns: category, rank, name, ancestors_json, stats_fitness_score.
• clan_history.parquet (14,085,063 rows)
  • Description: Detailed tracking of clan formations, mergers, and active members.
  • Key Columns: clan_id, members_count, merged_from_json, created_at_round.

2. Time Series Data (Temporal)

• time_series.parquet (8.5 Million rows)
  • Description: Global simulation metrics recorded every tick.
  • Key Columns: tick, network_activity, branching_ratio, phase_coherence, total_energy, dopamine, serotonin, acetylcholine, norepinephrine, itu_mean_fitness.
• per_nxer_time_series.parquet (162.9 Million rows | ~20.9 GB)
  • Description: Agent-level metrics recorded every tick. This is the largest table, allowing deep analysis of individual lifespans.
  • Key Columns: nxer_id, tick, alive, network_activity, total_energy, dopamine, mean_w_fast, mean_w_slow, fitness_score.

3. Neural Structure (Snapshots)

• neurons.parquet (3.76 Million rows)
  • Description: State snapshots of neurons (potential, health, adaptation).
• synapses.parquet (17.0 Million rows)
  • Description: Synaptic connections, including weights (w_fast, w_slow, w_meta) and integrity.
• world_grids.parquet (2,791 rows)
  • Description: The 2D terrain map for each game.

4. Event Logs (High Frequency)

• plasticity_events.parquet (278.9 Million rows)
  • Description: Logs every LTP/LTD event.
  • Columns: pre_id, post_id, delta_w, type.
• associativity_events.parquet (278.9 Million rows)
  • Description: Cooperative learning events where neighbor synapses assisted potentiation.
  • Columns: own_delta_w, neighbor_contribution, amplification_factor.
• weight_evolution_events.parquet (279.1 Million rows)
  • Description: Tracking weight shifts across the three timescales (Fast, Slow, Meta).
• neuromodulator_events.parquet (272.0 Million rows)
  • Description: Events where global modulators crossed affinity thresholds.
• subthreshold_events.parquet (277.2 Million rows)
  • Description: Membrane potential integration dynamics before firing.
• threshold_modulation_events.parquet (262.4 Million rows)
  • Description: Adjustments to firing thresholds via Acetylcholine or Autoreceptors.

Paper Section Mapping

The dataset is structured to support specific sections of the Neuraxon research paper. Use these field mappings for validation:

Paper Section	Relevant Columns / Tables
1. Trinary Neuromodulation	dopamine, serotonin, acetylcholine, excitatory_fraction, inhibitory_fraction, neutral_fraction, threshold_modulation_by_ach
2. Temporal Dynamics	oscillator_low/mid/high, temporal_sync, mean_phase_velocity, membrane_potential_mean
3. Synaptic Computation	mean_w_fast, mean_w_slow, mean_w_meta, silent_synapse_count, ionotropic_contribution_mean
4. Plasticity & Adaptation	ltp_rate, ltd_rate, associativity_event_count, mean_learning_rate_mod, plasticity_events (table)
5. Complex Signaling	dendritic_spike_count, mean_plateau_potential, subthreshold_integration_count
6. Self-Generated Activity	spontaneous_firing_count, mean_autocorrelation_window, mean_intrinsic_timescale
7. Synchronization	phase_coherence, branching_ratio, cfc_low_mid, pac_theta_gamma
8. Aigarth/ITU Evolution	itu_mean_fitness, itu_mutation_events, itu_pruning_events, clan_history (table)

Usage

Loading with Hugging Face Datasets

from datasets import load_dataset

# Load specific tables to save memory (e.g., just the game metadata and global time series)
dataset = load_dataset(
    "DavidVivancos/NeuraxonLife2.5-100K-TimeSeries", 
    data_files={
        "games": "games.parquet", 
        "time_series": "time_series.parquet"
    }
)

print(dataset['games'][0])

Loading with Pandas

import pandas as pd

# Load the global time series
df_ts = pd.read_parquet("time_series.parquet")

# Plot global network activity over time for a specific game
game_id = df_ts['game_id'].iloc[0]
subset = df_ts[df_ts['game_id'] == game_id]
subset.plot(x='tick', y='network_activity')

Note on Large Tables

The per_nxer_time_series.parquet (20GB+), associativity_events.parquet, and plasticity_events.parquet tables are very large. It is recommended to load these using streaming or by filtering for specific game_ids or round_numbers if not using a distributed framework like Spark or Dask.

Citation

@dataset{NeuraxonLife2.5-TimeSeries,
  title={Neuraxon Game of Life 2.5 Research Dataset: Deep Time Series Exploration},
  author={Vivancos, David and Sanchez, Jose},
  year={2026},
  publisher={Hugging Face},
  version={2.5.0},
  url={https://huggingface.co/datasets/DavidVivancos/NeuraxonLife2.5-100K-TimeSeries}
}

Authors & Curators

• David Vivancos / Artificiology Research - Qubic Science
• Dr. Jose Sanchez / UNIR - Qubic Science

Contact: For questions or issues, please open a GitHub issue at https://github.com/DavidVivancos/Neuraxon.