277 lines
No EOL
10 KiB
Python
277 lines
No EOL
10 KiB
Python
"""
|
|
agent.py
|
|
========
|
|
|
|
Kerangka kerja untuk semua 'otak' atau 'agent' pembuat keputusan.
|
|
Setiap agent harus mewarisi dari BaseAgent dan mengimplementasikan
|
|
metode `decide()`.
|
|
"""
|
|
|
|
import logging
|
|
from abc import ABC, abstractmethod
|
|
from typing import Dict, Any, List, Optional
|
|
|
|
# Konfigurasi logging dasar
|
|
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
|
|
|
|
class BaseAgent(ABC):
|
|
"""
|
|
Kelas dasar abstrak untuk semua agent.
|
|
"""
|
|
def __init__(self, name: str):
|
|
self.name = name
|
|
self.logger = logging.getLogger(self.name)
|
|
|
|
@abstractmethod
|
|
def decide(self, opportunity: Dict[str, Any]) -> str:
|
|
"""
|
|
Metode utama untuk membuat keputusan trading.
|
|
|
|
Args:
|
|
opportunity (Dict[str, Any]): Sebuah dictionary yang berisi semua
|
|
informasi tentang peluang trading
|
|
(skor, komponen, fitur, dll.).
|
|
|
|
Returns:
|
|
str: Keputusan trading, contoh: "ACCEPT", "REJECT", "WAIT".
|
|
"""
|
|
pass
|
|
|
|
def __str__(self):
|
|
return f"Agent(name={self.name})"
|
|
|
|
|
|
class RuleBasedAgent(BaseAgent):
|
|
"""
|
|
Agent sederhana yang membuat keputusan berdasarkan aturan-aturan dasar.
|
|
Contoh: Cek skor dan beberapa fitur kunci.
|
|
"""
|
|
def __init__(self, confidence_threshold: float = 5.0):
|
|
super().__init__(name="RuleBasedAgent")
|
|
self.confidence_threshold = confidence_threshold
|
|
self.logger.info(f"Agent diinisialisasi dengan threshold: {self.confidence_threshold}")
|
|
|
|
def decide(self, opportunity: Dict[str, Any]) -> str:
|
|
"""
|
|
Membuat keputusan berdasarkan skor dari sinyal.
|
|
"""
|
|
score = opportunity.get('score', 0)
|
|
|
|
if abs(score) >= self.confidence_threshold:
|
|
decision = "ACCEPT"
|
|
self.logger.info(f"Keputusan: {decision} (Skor {score:.2f} >= Threshold {self.confidence_threshold})")
|
|
else:
|
|
decision = "REJECT"
|
|
self.logger.info(f"Keputusan: {decision} (Skor {score:.2f} < Threshold {self.confidence_threshold})")
|
|
|
|
return decision
|
|
|
|
|
|
# --- Agent Cerdas dengan Model Machine Learning ---
|
|
import joblib
|
|
import numpy as np
|
|
from sklearn.base import BaseEstimator
|
|
|
|
class NeuralAgent(BaseAgent):
|
|
"""
|
|
Agent yang menggunakan model Jaringan Syaraf Tiruan (atau model scikit-learn lainnya)
|
|
yang sudah dilatih untuk membuat keputusan.
|
|
"""
|
|
def __init__(self, model_path: str, probability_threshold: float = 0.70):
|
|
super().__init__(name="NeuralAgent")
|
|
self.model = self._load_model(model_path)
|
|
self.probability_threshold = probability_threshold
|
|
self.logger.info(f"Agent diinisialisasi dengan model dari: {model_path} dan threshold probabilitas: {self.probability_threshold}")
|
|
|
|
def _load_model(self, model_path: str) -> Optional[BaseEstimator]:
|
|
"""Memuat model yang sudah dilatih dari file."""
|
|
try:
|
|
model = joblib.load(model_path)
|
|
self.logger.info(f"Model {type(model).__name__} berhasil dimuat.")
|
|
return model
|
|
except FileNotFoundError:
|
|
self.logger.error(f"File model tidak ditemukan di {model_path}. Agent tidak akan berfungsi.")
|
|
return None
|
|
except Exception as e:
|
|
self.logger.error(f"Gagal memuat model dari {model_path}: {e}")
|
|
return None
|
|
|
|
def decide(self, opportunity: Dict[str, Any]) -> str:
|
|
"""
|
|
Membuat keputusan menggunakan prediksi dari model neural network.
|
|
"""
|
|
if self.model is None:
|
|
self.logger.warning("Model tidak tersedia, keputusan otomatis REJECT.")
|
|
return "REJECT"
|
|
|
|
features = opportunity.get('features')
|
|
if features is None or not hasattr(features, 'size') or features.size == 0:
|
|
self.logger.warning("Tidak ada fitur (features) untuk dianalisis, keputusan REJECT.")
|
|
return "REJECT"
|
|
|
|
# Model scikit-learn mengharapkan input 2D
|
|
features_2d = np.array(features).reshape(1, -1)
|
|
|
|
try:
|
|
# Memeriksa apakah model memiliki metode predict_proba
|
|
if not hasattr(self.model, 'predict_proba'):
|
|
self.logger.warning(f"Model {type(self.model).__name__} tidak memiliki 'predict_proba'. Menggunakan 'predict'.")
|
|
prediction = self.model.predict(features_2d)[0]
|
|
# Asumsikan '1' atau 'ACCEPT' adalah sinyal positif
|
|
decision = "ACCEPT" if str(prediction).upper() in ["1", "ACCEPT"] else "REJECT"
|
|
else:
|
|
# Prediksi probabilitas (asumsi kelas 1 adalah 'WIN' atau 'ACCEPT')
|
|
probabilities = self.model.predict_proba(features_2d)[0]
|
|
if len(probabilities) < 2:
|
|
self.logger.error(f"Model hanya mengembalikan {len(probabilities)} probabilitas. Butuh setidaknya 2.")
|
|
return "REJECT"
|
|
|
|
probability_of_win = probabilities[1]
|
|
|
|
if probability_of_win > self.probability_threshold:
|
|
decision = "ACCEPT"
|
|
else:
|
|
decision = "REJECT"
|
|
self.logger.info(f"Probabilitas keberhasilan: {probability_of_win:.2%}. Keputusan: {decision}")
|
|
|
|
return decision
|
|
|
|
except Exception as e:
|
|
self.logger.error(f"Gagal membuat prediksi: {e}", exc_info=True)
|
|
return "REJECT"
|
|
|
|
class EnsembleAgent(BaseAgent):
|
|
"""
|
|
Agent yang menggabungkan keputusan dari beberapa agent lain untuk
|
|
mencapai konsensus.
|
|
"""
|
|
def __init__(self, agents: List[BaseAgent], strategy: str = 'majority_vote'):
|
|
super().__init__(name="EnsembleAgent")
|
|
if not agents:
|
|
raise ValueError("EnsembleAgent memerlukan setidaknya satu agent.")
|
|
self.agents = agents
|
|
self.strategy = strategy
|
|
self.logger.info(f"Agent diinisialisasi dengan {len(self.agents)} sub-agents dan strategi: {self.strategy}")
|
|
|
|
def decide(self, opportunity: Dict[str, Any]) -> str:
|
|
"""
|
|
Menjalankan strategi voting untuk membuat keputusan akhir.
|
|
"""
|
|
decisions = []
|
|
for agent in self.agents:
|
|
try:
|
|
decision = agent.decide(opportunity)
|
|
decisions.append(decision)
|
|
self.logger.debug(f"Agent {agent.name} memutuskan: {decision}")
|
|
except Exception as e:
|
|
self.logger.error(f"Agent {agent.name} gagal membuat keputusan: {e}")
|
|
decisions.append("REJECT") # Anggap REJECT jika ada error
|
|
|
|
if not decisions:
|
|
self.logger.warning("Tidak ada keputusan dari sub-agents, mengembalikan REJECT.")
|
|
return "REJECT"
|
|
|
|
if self.strategy == 'majority_vote':
|
|
return self._majority_vote(decisions)
|
|
else:
|
|
self.logger.error(f"Strategi '{self.strategy}' tidak diketahui. Menggunakan REJECT.")
|
|
return "REJECT"
|
|
|
|
def _majority_vote(self, decisions: List[str]) -> str:
|
|
"""Menentukan keputusan berdasarkan suara mayoritas."""
|
|
votes = {"ACCEPT": 0, "REJECT": 0, "WAIT": 0}
|
|
for d in decisions:
|
|
if d in votes:
|
|
votes[d] += 1
|
|
|
|
# Log detail voting
|
|
self.logger.info(f"Hasil voting: {votes}")
|
|
|
|
# Prioritaskan ACCEPT jika imbang antara ACCEPT dan REJECT
|
|
if votes["ACCEPT"] > 0 and votes["ACCEPT"] >= votes["REJECT"]:
|
|
final_decision = "ACCEPT"
|
|
elif votes["REJECT"] > votes["ACCEPT"]:
|
|
final_decision = "REJECT"
|
|
else: # Jika hanya ada WAIT atau tidak ada mayoritas jelas
|
|
final_decision = "WAIT"
|
|
|
|
self.logger.info(f"Keputusan akhir (mayoritas): {final_decision}")
|
|
return final_decision
|
|
|
|
def create_agent(config: Dict[str, Any]) -> BaseAgent:
|
|
"""
|
|
Factory function untuk membuat instance agent berdasarkan konfigurasi.
|
|
"""
|
|
agent_type = config.get("type")
|
|
params = config.get("params", {})
|
|
|
|
if agent_type == "rule_based":
|
|
return RuleBasedAgent(**params)
|
|
elif agent_type == "neural":
|
|
return NeuralAgent(**params)
|
|
elif agent_type == "ensemble":
|
|
sub_agents_configs = params.get("agents", [])
|
|
sub_agents = [create_agent(conf) for conf in sub_agents_configs]
|
|
strategy = params.get("strategy", "majority_vote")
|
|
return EnsembleAgent(agents=sub_agents, strategy=strategy)
|
|
else:
|
|
raise ValueError(f"Tipe agent tidak diketahui: {agent_type}")
|
|
|
|
# Contoh Penggunaan:
|
|
if __name__ == '__main__':
|
|
# Contoh ini hanya akan berjalan jika file dieksekusi secara langsung
|
|
# dan memerlukan file model dummy.
|
|
|
|
# 1. Buat file model dummy untuk pengujian
|
|
from sklearn.linear_model import LogisticRegression
|
|
import os
|
|
|
|
dummy_model_path = "dummy_model.joblib"
|
|
if not os.path.exists(dummy_model_path):
|
|
# Latih model sederhana
|
|
X_train = np.random.rand(10, 5)
|
|
y_train = (np.sum(X_train, axis=1) > 2.5).astype(int)
|
|
dummy_model = LogisticRegression()
|
|
dummy_model.fit(X_train, y_train)
|
|
joblib.dump(dummy_model, dummy_model_path)
|
|
print(f"Model dummy '{dummy_model_path}' dibuat.")
|
|
|
|
# 2. Konfigurasi untuk membuat agent
|
|
agent_config = {
|
|
"type": "ensemble",
|
|
"params": {
|
|
"strategy": "majority_vote",
|
|
"agents": [
|
|
{
|
|
"type": "rule_based",
|
|
"params": {"confidence_threshold": 6.0}
|
|
},
|
|
{
|
|
"type": "neural",
|
|
"params": {
|
|
"model_path": dummy_model_path,
|
|
"probability_threshold": 0.65
|
|
}
|
|
}
|
|
]
|
|
}
|
|
}
|
|
|
|
# 3. Buat agent utama dari konfigurasi
|
|
main_agent = create_agent(agent_config)
|
|
|
|
# 4. Siapkan data peluang dummy
|
|
dummy_opportunity = {
|
|
"score": 7.5,
|
|
"features": np.random.rand(5) # Fitur harus sesuai dengan yang diharapkan model
|
|
}
|
|
|
|
# 5. Dapatkan keputusan dari agent
|
|
print("-" * 30)
|
|
final_decision = main_agent.decide(dummy_opportunity)
|
|
print(f"\nKEPUTUSAN FINAL DARI {main_agent.name}: {final_decision}")
|
|
print("-" * 30)
|
|
|
|
# 6. Hapus file dummy
|
|
os.remove(dummy_model_path)
|
|
print(f"Model dummy '{dummy_model_path}' dihapus.") |