Overview
The export.py module provides utilities to export evolved genomes from the genetic algorithm into standalone Python trading bots that can run independently for paper trading or real trading on Kalshi.
Command-Line Interface
# Export the best bot ever trained
python -m genetic.export
# Export from a specific generation
python -m genetic.export --generation 5
# Export a specific genome by ID
python -m genetic.export --genome-id a1b2c3d4
# Just inspect the hall of fame (no export)
python -m genetic.export --show
Functions
show_hall_of_fame()
Print the hall of fame leaderboard to console.
Output Example:
===========================================================================
HALL OF FAME (updated: 2023-12-31 10:30:45)
Latest generation: 10
===========================================================================
Rank Genome Gen ROI% Trades WinRate PnL Signal
---------------------------------------------------------------------------
1 a1b2c3d4 7 25.3% 234 68.2% $25.30 momentum
2 e5f6g7h8 9 22.1% 189 64.5% $22.10 price_level
3 i9j0k1l2 10 19.8% 156 61.3% $19.80 value
...
Best ever: genome a1b2c3d4 (25.3% ROI, gen 7)
To export as a real bot:
python -m genetic.export --genome-id a1b2c3d4
from genetic.export import show_hall_of_fame
show_hall_of_fame()
get_genome_from_hof()
Get a genome entry from the hall of fame.
Genome ID to retrieve (if None, returns best ever)
Tuple of (entry_dict, genome_dict) if found, None otherwise
from genetic.export import get_genome_from_hof
# Get best ever
result = get_genome_from_hof()
if result:
entry, genome_dict = result
print(f"Best: {entry['fitness_roi_pct']:.1f}% ROI")
# Get specific genome
result = get_genome_from_hof("a1b2c3d4")
if result:
entry, genome_dict = result
print(f"Found genome from generation {entry['generation']}")
get_genome_from_generation()
Get a genome from a specific generation file.
Generation number to load from
Specific genome ID (if None, returns best from generation)
Tuple of (entry_dict, genome_dict) if found, None otherwise
from genetic.export import get_genome_from_generation
# Get best from generation 5
result = get_genome_from_generation(5)
if result:
entry, genome_dict = result
print(f"Best in gen 5: {entry['fitness_roi_pct']:.1f}% ROI")
# Get specific genome from generation 5
result = get_genome_from_generation(5, "e5f6g7h8")
export_genome()
Export a genome as a standalone real trading bot.
Entry dictionary from hall of fame or generation file
Genome dictionary (serialized Genome object)
data/evolution/exports/genome_{genome_id}.json - Genome data with decoded paramsdata/evolution/exports/bot_{genome_id}.py - Standalone trading bot script
Output Example:
Exported genome a1b2c3d4:
Genome JSON: data/evolution/exports/genome_a1b2c3d4.json
Trading bot: data/evolution/exports/bot_a1b2c3d4.py
Decoded strategy:
Signal type: momentum
Categories: ['politics', 'economics']
Bankroll frac: 0.050
Max concurrent: 5
Max trades/day: 20
Price range: $0.20 - $0.80
Daily loss limit: 5.0%
To paper trade (dry run):
DRY_RUN=true python data/evolution/exports/bot_a1b2c3d4.py
To trade for real (CAREFUL):
DRY_RUN=false python data/evolution/exports/bot_a1b2c3d4.py
from genetic.export import get_genome_from_hof, export_genome
result = get_genome_from_hof()
if result:
entry, genome_dict = result
export_genome(entry, genome_dict)
Generated Bot Structure
The exported bot is a standalone Python file with:
Hardcoded Genome
All gene values embedded as constants
Environment Variables
DRY_RUN - Set to “true” for paper trading, “false” for real tradingBANKROLL_USD - Initial bankroll amount (default: 100)
Trading Modes
Paper Trading Mode (DRY_RUN=true)
- Uses
PaperTradingEngine for local simulation
- No real orders placed
- Logs progress every 60 ticks
- Safe for testing strategies
DRY_RUN=true BANKROLL_USD=1000 python bot_a1b2c3d4.py
Real Trading Mode (DRY_RUN=false)
- Places actual orders via Kalshi API
- Requires valid API credentials
- Respects all genome parameters (max trades/day, loss limits, etc.)
- USE WITH CAUTION
DRY_RUN=false BANKROLL_USD=100 python bot_a1b2c3d4.py
Signal Generation
The bot includes a _generate_signal() function that implements all supported signal types:
price_level - Trade when price is in threshold rangemomentum - Follow price momentum over lookback periodmean_reversion - Trade against Z-score deviationsvalue - Trade when edge exceeds minimumcontrarian - Fade extreme prices
Risk Management
The real trading mode enforces:
- Maximum concurrent positions
- Maximum trades per day
- Daily loss limits
- Price range filters
- Volume filters
- Time to expiry constraints
Complete Export Workflow
from genetic.export import (
show_hall_of_fame,
get_genome_from_hof,
export_genome
)
# 1. View hall of fame
show_hall_of_fame()
# 2. Select and export best genome
result = get_genome_from_hof()
if result:
entry, genome_dict = result
# Show strategy details
print(f"\nExporting genome {genome_dict['id']}:")
print(f" Generation: {entry['generation']}")
print(f" Training ROI: {entry['fitness_roi_pct']:.1f}%")
print(f" Win rate: {entry['win_rate'] * 100:.1f}%")
print(f" Signal: {entry['signal_type']}")
# Export
export_genome(entry, genome_dict)
print("\n✓ Ready for deployment!")
Genome JSON Structure
The exported JSON file contains:
{
"genome": {
"id": "a1b2c3d4",
"generation": 7,
"signal_type_idx": 1,
"bankroll_fraction": 0.05,
...
},
"decoded_params": {
"signal_type": "momentum",
"bankroll_fraction": 0.05,
"max_concurrent": 5,
"momentum_lookback_ticks": 12,
...
},
"categories": ["politics", "economics"],
"source": {
"fitness_roi_pct": 25.3,
"total_trades": 234,
"settled_trades": 189,
"win_rate": 0.682,
"realized_pnl": 25.30,
"generation": 7,
"signal_type": "momentum"
}
}
Safety Notes
Real Trading Risks
- Exported bots place real orders when
DRY_RUN=false
- Always test in paper trading mode first
- Past performance (training ROI) does not guarantee future results
- Monitor bots closely when running in production
- Start with small bankrolls to limit risk
Best Practices
- Run
--show first to review all candidates
- Test exported bots in paper trading mode for at least 24 hours
- Compare paper trading results to training ROI before going live
- Use environment variables to easily switch between modes
- Keep exported bots version-controlled