pip install pandas pydantic huggingface-hub

from praisonaiagents.tools import cot_save

# Save reasoning data
cot_save(
    input_data="Solve: If a train travels 120 miles in 2 hours, what is its average speed?",
    output_data={
        "step1": "Identify given information: distance = 120 miles, time = 2 hours",
        "step2": "Apply formula: speed = distance / time",
        "step3": "Calculate: speed = 120 miles / 2 hours = 60 mph",
        "answer": "60 miles per hour"
    },
    filename="reasoning_examples.csv"
)

from praisonaiagents.tools import cot_upload_to_huggingface

# Upload dataset to Hugging Face
result = cot_upload_to_huggingface(
    dataset_path="reasoning_examples.csv",
    dataset_name="my-cot-dataset",
    token="your-huggingface-token"
)

from praisonaiagents import Agent, Task
from praisonaiagents.tools import cot_save

# Create a reasoning agent
reasoning_agent = Agent(
    name="Reasoning Agent",
    instructions="""You are an expert at breaking down complex problems into steps.
    For each problem, provide:
    1. Problem understanding
    2. Step-by-step solution
    3. Final answer
    4. Verification""",
    tools=[cot_save]
)

# Create a task
task = Task(
    description="Solve this problem step-by-step: A bakery sells 120 cookies per day. If they're open 6 days a week, how many cookies do they sell in 4 weeks?",
    agent=reasoning_agent
)

# Execute and save reasoning
result = task.execute()

from pydantic import BaseModel, Field
from typing import List
from praisonaiagents.tools import cot_save

class ReasoningStep(BaseModel):
    step_number: int = Field(description="Step number in the reasoning process")
    description: str = Field(description="What this step accomplishes")
    calculation: str = Field(description="Any calculations performed")
    result: str = Field(description="Result of this step")

class ChainOfThought(BaseModel):
    problem: str = Field(description="The original problem statement")
    steps: List[ReasoningStep] = Field(description="List of reasoning steps")
    final_answer: str = Field(description="The final answer")
    confidence: float = Field(description="Confidence level (0-1)")

# Example usage
cot_example = ChainOfThought(
    problem="Calculate compound interest for $1000 at 5% for 3 years",
    steps=[
        ReasoningStep(
            step_number=1,
            description="Identify the compound interest formula",
            calculation="A = P(1 + r)^t",
            result="Formula identified"
        ),
        ReasoningStep(
            step_number=2,
            description="Substitute values",
            calculation="A = 1000(1 + 0.05)^3",
            result="Values substituted"
        ),
        ReasoningStep(
            step_number=3,
            description="Calculate the result",
            calculation="A = 1000(1.05)^3 = 1000 × 1.157625 = 1157.63",
            result="$1,157.63"
        )
    ],
    final_answer="$1,157.63",
    confidence=0.95
)

# Save structured reasoning
cot_save(
    input_data=cot_example.problem,
    output_data=cot_example.model_dump(),
    filename="structured_reasoning.csv"
)

from praisonaiagents import Agent, Task, Process
from praisonaiagents.tools import cot_save, cot_upload_to_huggingface

# Problem decomposer agent
decomposer = Agent(
    name="Problem Decomposer",
    instructions="Break down complex problems into smaller sub-problems",
    tools=[cot_save]
)

# Step solver agent
solver = Agent(
    name="Step Solver",
    instructions="Solve each sub-problem step by step",
    tools=[cot_save]
)

# Verifier agent
verifier = Agent(
    name="Solution Verifier",
    instructions="Verify the solution and check for errors",
    tools=[cot_save]
)

# Tasks
decompose_task = Task(
    description="Break down: How many different 5-card poker hands contain exactly 3 aces?",
    agent=decomposer
)

solve_task = Task(
    description="Solve each sub-problem identified",
    agent=solver
)

verify_task = Task(
    description="Verify the solution using a different approach",
    agent=verifier
)

# Process
reasoning_process = Process(
    agents=[decomposer, solver, verifier],
    tasks=[decompose_task, solve_task, verify_task]
)

result = reasoning_process.run()

import os

# Hugging Face configuration
os.environ['HUGGINGFACE_TOKEN'] = 'your-token-here'
os.environ['HUGGINGFACE_ORGANIZATION'] = 'your-org-name'

# Default save directory
os.environ['COT_DATA_DIR'] = '/path/to/cot/data'

# CSV formatting options
os.environ['COT_CSV_DELIMITER'] = ','
os.environ['COT_CSV_ENCODING'] = 'utf-8'

# Configure CSV output format
def save_cot_with_metadata(input_data, output_data, metadata=None):
    import pandas as pd
    from datetime import datetime
    import os
    
    # Add metadata
    entry = {
        'timestamp': datetime.now().isoformat(),
        'input': input_data,
        'output': output_data,
        'model': metadata.get('model', 'unknown'),
        'temperature': metadata.get('temperature', 0.7),
        'reasoning_type': metadata.get('type', 'general')
    }
    
    # Save to CSV with custom formatting
    df = pd.DataFrame([entry])
    df.to_csv(
        'cot_data_with_metadata.csv',
        mode='a',
        header=not os.path.exists('cot_data_with_metadata.csv'),
        index=False
    )

from praisonaiagents.tools import cot_save
import pandas as pd

# Process multiple problems in batch
problems = [
    "Calculate 15% of 240",
    "Find the area of a circle with radius 7",
    "Convert 72°F to Celsius"
]

for problem in problems:
    # Generate reasoning for each problem
    reasoning_steps = generate_reasoning(problem)  # Your reasoning function
    
    # Save each solution
    cot_save(
        input_data=problem,
        output_data=reasoning_steps,
        filename="batch_reasoning.csv"
    )

# Upload complete dataset
cot_upload_to_huggingface(
    dataset_path="batch_reasoning.csv",
    dataset_name="math-reasoning-dataset"
)

# Analyze reasoning quality
def analyze_reasoning_quality(csv_path):
    import ast
    df = pd.read_csv(csv_path)
    
    metrics = {
        'total_examples': len(df),
        'avg_steps': df['output'].apply(lambda x: len(ast.literal_eval(x).get('steps', []))).mean(),
        'completeness': (df['output'].notna().sum() / len(df)) * 100,
        'unique_patterns': df['input'].nunique()
    }
    
    return metrics

# Check quality before uploading
quality = analyze_reasoning_quality("reasoning_examples.csv")
print(f"Dataset quality metrics: {quality}")

# Prepare data for model training
def prepare_for_training(csv_path, output_format='jsonl'):
    df = pd.read_csv(csv_path)
    
    training_data = []
    for _, row in df.iterrows():
        entry = {
            'instruction': row['input'],
            'reasoning': row['output'],
            'model_type': 'chain-of-thought'
        }
        training_data.append(entry)
    
    # Save in training format
    if output_format == 'jsonl':
        import json
        with open('training_data.jsonl', 'w') as f:
            for item in training_data:
                f.write(json.dumps(item) + '\n')
    
    return training_data

# Generate step-by-step solutions for educational materials
educational_agent = Agent(
    name="Education Agent",
    instructions="Create detailed, educational explanations suitable for students",
    tools=[cot_save]
)

# Generate explanations for various topics
topics = ["Pythagorean theorem", "Photosynthesis", "Supply and demand"]
for topic in topics:
    task = Task(
        description=f"Explain {topic} with clear reasoning steps",
        agent=educational_agent
    )
    task.execute()

# Use CoT for debugging code
debugging_agent = Agent(
    name="Debug Agent",
    instructions="Analyze code errors step by step",
    tools=[cot_save]
)

error_task = Task(
    description="Debug: Why does 'list.append(x)' return None in Python?",
    agent=debugging_agent
)