Back to Details

agent-o-rama

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

agent-o-rama

agent-o-rama

Layer 4: Learning and Pattern Extraction for Cognitive Surrogate Systems
Version: 1.0.0
Trit: +1 (Generator - produces learned patterns)
Bundle: learning
第4层：认知代理系统的学习与模式提取
版本: 1.0.0
Trit: +1（生成器 - 生成学习到的模式）
** Bundle**: learning

Overview

概述

Agent-o-rama trains learning agents on interaction sequences to discover behavioral patterns. It extracts temporal, topic, and network patterns from raw interaction data, producing models compatible with the cognitive-surrogate skill.
NEW (Langevin/Unworld Integration): Agent-o-rama now supports both:
  1. Temporal Learning (traditional): Train interaction predictor via epochs
  2. Derivational Generation (unworld): Generate equivalent patterns via seed chaining (100x faster, deterministic)
Agent-o-rama 通过在交互序列上训练学习型Agent，发现行为模式。它从原始交互数据中提取时间、主题和网络模式，生成与cognitive-surrogate skill兼容的模型。
新增（Langevin/Unworld集成）：Agent-o-rama现在同时支持两种方式：
  1. 时间学习（传统方式）：通过轮次训练交互预测器
  2. 派生生成（unworld方式）：通过种子链生成等效模式（速度提升100倍，确定性）

Capabilities

功能特性

1. train-interaction-predictor

1. train-interaction-predictor

Train a model to predict next interactions given history.
python
from agent_o_rama import InteractionPredictor

predictor = InteractionPredictor(
    learning_rate=0.01,
    epochs=100,
    batch_size=32,
    seed=0xf061ebbc2ca74d78  # SPI seed for reproducibility
)
训练模型以根据历史记录预测下一次交互。
python
from agent_o_rama import InteractionPredictor

predictor = InteractionPredictor(
    learning_rate=0.01,
    epochs=100,
    batch_size=32,
    seed=0xf061ebbc2ca74d78  # SPI种子，确保可复现
)

Train on DuckDB interaction sequences

在DuckDB交互序列上训练

predictor.fit( db_path="interactions.duckdb", table="interaction_sequences", validation_split=0.2 )
predictor.fit( db_path="interactions.duckdb", table="interaction_sequences", validation_split=0.2 )

Predict next interaction

预测下一次交互

next_pred = predictor.predict(recent_history)
undefined
next_pred = predictor.predict(recent_history)
undefined

2. extract-temporal-patterns

2. extract-temporal-patterns

Discover time-based behavioral patterns.
sql
-- Pattern query for DuckDB
SELECT 
    EXTRACT(HOUR FROM created_at) as hour,
    EXTRACT(DOW FROM created_at) as day_of_week,
    COUNT(*) as post_count,
    AVG(response_time_minutes) as avg_response_time
FROM interactions
GROUP BY hour, day_of_week
ORDER BY post_count DESC;
Output Schema:
TemporalPattern:
  - peak_hours: [9, 14, 21]
  - peak_days: [1, 3, 5]  # Mon, Wed, Fri
  - avg_response_time: 12.5 minutes
  - posting_frequency: 4.2 posts/day
  - engagement_cycles: [{start: 9, end: 11, intensity: 0.8}]
发现基于时间的行为模式。
sql
-- DuckDB模式查询
SELECT 
    EXTRACT(HOUR FROM created_at) as hour,
    EXTRACT(DOW FROM created_at) as day_of_week,
    COUNT(*) as post_count,
    AVG(response_time_minutes) as avg_response_time
FROM interactions
GROUP BY hour, day_of_week
ORDER BY post_count DESC;
输出 Schema:
TemporalPattern:
  - peak_hours: [9, 14, 21]
  - peak_days: [1, 3, 5]  # 周一、周三、周五
  - avg_response_time: 12.5 minutes
  - posting_frequency: 4.2 posts/day
  - engagement_cycles: [{start: 9, end: 11, intensity: 0.8}]

3. extract-topic-patterns

3. extract-topic-patterns

Analyze topic dynamics and correlations.
python
patterns = extract_topic_patterns(
    posts=all_posts,
    embedding_model="all-MiniLM-L6-v2",
    n_topics=20
)
分析主题动态与相关性。
python
patterns = extract_topic_patterns(
    posts=all_posts,
    embedding_model="all-MiniLM-L6-v2",
    n_topics=20
)

Returns:

返回结果:

- topic_distribution: {topic_id: frequency}

- topic_distribution: {topic_id: frequency}

- topic_transitions: Markov chain P(topic_j | topic_i)

- topic_transitions: 马尔可夫链 P(topic_j | topic_i)

- topic_entropy: Shannon entropy of topic usage

- topic_entropy: 主题使用的香农熵

- topic_clusters: Hierarchical clustering of related topics

- topic_clusters: 相关主题的层次聚类

undefined
undefined

4. skill-discovery

4. skill-discovery

Identify latent skills from behavioral patterns.
python
skills = discover_skills(
    interactions=interaction_log,
    min_frequency=5,
    coherence_threshold=0.7
)
从行为模式中识别潜在技能。
python
skills = discover_skills(
    interactions=interaction_log,
    min_frequency=5,
    coherence_threshold=0.7
)

Example output:

示例输出:

[

[

{skill: "category-theory-explanation", frequency: 23, coherence: 0.89},

{skill: "category-theory-explanation", frequency: 23, coherence: 0.89},

{skill: "code-review-feedback", frequency: 45, coherence: 0.92},

{skill: "code-review-feedback", frequency: 45, coherence: 0.92},

{skill: "community-bridge-building", frequency: 18, coherence: 0.85}

{skill: "community-bridge-building", frequency: 18, coherence: 0.85}

]

]

undefined
undefined

5. derive-patterns-via-unworld

5. derive-patterns-via-unworld

Generate patterns via derivational chaining (NEW - Langevin/Unworld path).
python
from agent_o_rama import UnworldPatternDeriver
通过派生链生成模式（新增 - Langevin/Unworld路径）。
python
from agent_o_rama import UnworldPatternDeriver

Instead of train_interaction_predictor(epochs=100)

替代 train_interaction_predictor(epochs=100)

Now also support:

现在还支持:

deriver = UnworldPatternDeriver( genesis_seed=0xDEADBEEF, interaction_schema=schema )
deriver = UnworldPatternDeriver( genesis_seed=0xDEADBEEF, interaction_schema=schema )

Generate learned patterns deterministically

确定性生成学习到的模式

patterns = deriver.derive_patterns( depth=100, # Derivation depth instead of epochs verify_gf3=True # Verify GF(3) conservation )
patterns = deriver.derive_patterns( depth=100, # 派生深度，替代轮次 verify_gf3=True # 验证GF(3)守恒 )

Cost comparison

成本对比

cost_analysis = { "temporal_training": { "time": "5-10 minutes", "cost": "high (compute)", "determinism": "stochastic" }, "derivational_generation": { "time": "5-10 seconds", "cost": "low", "determinism": "deterministic ✓" } }
undefined
cost_analysis = { "temporal_training": { "time": "5-10 minutes", "cost": "high (compute)", "determinism": "stochastic" }, "derivational_generation": { "time": "5-10 seconds", "cost": "low", "determinism": "deterministic ✓" } }
undefined

6. verify-equivalence-via-bisimulation

6. verify-equivalence-via-bisimulation

Prove temporal and derivational patterns are behaviorally equivalent.
python
from bisimulation_game import BisimulationGame
证明时间模式与派生模式在行为上等效。
python
from bisimulation_game import BisimulationGame

Verify that temporal and derivational patterns are equivalent

验证时间模式与派生模式是否等效

are_equivalent = BisimulationGame( system1=learned_patterns, # from temporal training system2=derived_patterns, # from unworld derivation seed=0xDEADBEEF ).play()
if are_equivalent: print("✓ Patterns are behaviorally equivalent") print("✓ Can safely switch from temporal to derivational")
undefined
are_equivalent = BisimulationGame( system1=learned_patterns, # 来自时间训练 system2=derived_patterns, # 来自unworld派生 seed=0xDEADBEEF ).play()
if are_equivalent: print("✓ 模式在行为上等效") print("✓ 可安全从时间方式切换为派生方式")
undefined

7. validate-held-out

7. validate-held-out

Cross-validate models on held-out test sets.
python
validation = validate_held_out(
    predictor=trained_model,
    test_set=held_out_interactions,
    metrics=["accuracy", "perplexity", "topic_match", "style_match"]
)
在预留测试集上交叉验证模型。
python
validation = validate_held_out(
    predictor=trained_model,
    test_set=held_out_interactions,
    metrics=["accuracy", "perplexity", "topic_match", "style_match"]
)

Target: >80% accuracy on next-topic prediction

目标：下一个主题预测准确率>80%

assert validation.accuracy > 0.80
undefined
assert validation.accuracy > 0.80
undefined

DuckDB Integration

DuckDB集成

Training Data Schema

训练数据Schema

sql
CREATE TABLE interaction_sequences (
    sequence_id VARCHAR PRIMARY KEY,
    user_id VARCHAR,
    interactions JSON,  -- Array of interaction objects
    created_at TIMESTAMP,
    topic_labels VARCHAR[],
    sentiment_arc FLOAT[]
);

CREATE TABLE learned_patterns (
    pattern_id VARCHAR PRIMARY KEY,
    pattern_type VARCHAR,  -- 'temporal', 'topic', 'network', 'skill'
    pattern_data JSON,
    confidence FLOAT,
    learned_at TIMESTAMP,
    seed BIGINT  -- SPI seed for reproducibility
);
sql
CREATE TABLE interaction_sequences (
    sequence_id VARCHAR PRIMARY KEY,
    user_id VARCHAR,
    interactions JSON,  -- 交互对象数组
    created_at TIMESTAMP,
    topic_labels VARCHAR[],
    sentiment_arc FLOAT[]
);

CREATE TABLE learned_patterns (
    pattern_id VARCHAR PRIMARY KEY,
    pattern_type VARCHAR,  -- 'temporal', 'topic', 'network', 'skill'
    pattern_data JSON,
    confidence FLOAT,
    learned_at TIMESTAMP,
    seed BIGINT  -- SPI种子，确保可复现
);

GF(3) Triad Integration

GF(3)三元组集成

Agent-o-rama forms triads with:
TritSkillRole
-1self-validation-loopValidates learned patterns
0cognitive-surrogateConsumes patterns for prediction
+1agent-o-ramaGenerates learned patterns
Conservation: (-1) + (0) + (+1) = 0 ✓
Agent-o-rama与以下组件形成三元组：
TritSkill角色
-1self-validation-loop验证学习到的模式
0cognitive-surrogate消费模式以进行预测
+1agent-o-rama生成学习到的模式
守恒性：(-1) + (0) + (+1) = 0 ✓

Configuration

配置

yaml
undefined
yaml
undefined

agent-o-rama.yaml

agent-o-rama.yaml

training: learning_rate: 0.01 epochs: 100 batch_size: 32 early_stopping: true patience: 10
patterns: temporal: granularity: hour lookback_days: 90 topic: n_topics: 20 min_topic_size: 5 skill: min_frequency: 5 coherence_threshold: 0.7
reproducibility: seed: 0xf061ebbc2ca74d78 deterministic: true
undefined
training: learning_rate: 0.01 epochs: 100 batch_size: 32 early_stopping: true patience: 10
patterns: temporal: granularity: hour lookback_days: 90 topic: n_topics: 20 min_topic_size: 5 skill: min_frequency: 5 coherence_threshold: 0.7
reproducibility: seed: 0xf061ebbc2ca74d78 deterministic: true
undefined

Example Workflow

示例工作流

bash
undefined
bash
undefined

1. Extract patterns from interaction data

1. 从交互数据中提取模式

just agent-train interactions.duckdb --epochs 100
just agent-train interactions.duckdb --epochs 100

2. Discover skills

2. 发现技能

just agent-discover-skills --min-freq 5
just agent-discover-skills --min-freq 5

3. Validate on held-out set

3. 在预留数据集上验证

just agent-validate --test-split 0.2
just agent-validate --test-split 0.2

4. Export patterns for cognitive-surrogate

4. 导出模式供cognitive-surrogate使用

just agent-export patterns.json
undefined
just agent-export patterns.json
undefined

Related Skills

相关技能

  • cognitive-surrogate
    (Layer 6) - Consumes learned patterns
  • entropy-sequencer
    (Layer 5) - Arranges training data
  • acsets
    (Layer 3) - Structured pattern storage
  • gay-mcp
    - Deterministic seeding via SPI
  • cognitive-surrogate
    （第6层）- 消费学习到的模式
  • entropy-sequencer
    （第5层）- 整理训练数据
  • acsets
    （第3层）- 结构化模式存储
  • gay-mcp
    - 通过SPI实现确定性种子生成