reduce number of predictions
This commit is contained in:
rishubm
294
ai_intelligence_layer/RACE_CONTEXT.md
Normal file
294
ai_intelligence_layer/RACE_CONTEXT.md
Normal file
@@ -0,0 +1,294 @@
|
||||
# Race Context Guide
|
||||
|
||||
## Why Race Context is Separate from Telemetry
|
||||
|
||||
**Enrichment Service** (port 8000):
|
||||
- Provides: **Enriched telemetry** (changes every lap)
|
||||
- Example: tire degradation, aero efficiency, ERS charge
|
||||
|
||||
**Client/Frontend**:
|
||||
- Provides: **Race context** (changes less frequently)
|
||||
- Example: driver name, current position, track info, competitors
|
||||
|
||||
This separation is intentional:
|
||||
- Telemetry changes **every lap** (real-time HPC data)
|
||||
- Race context changes **occasionally** (position changes, pit stops)
|
||||
- Keeps enrichment service simple and focused
|
||||
|
||||
## How to Call Brainstorm with Both
|
||||
|
||||
### Option 1: Client Provides Both (Recommended)
|
||||
|
||||
```bash
|
||||
curl -X POST http://localhost:9000/api/strategy/brainstorm \
|
||||
-H "Content-Type: application/json" \
|
||||
-d '{
|
||||
"enriched_telemetry": [
|
||||
{
|
||||
"lap": 27,
|
||||
"aero_efficiency": 0.85,
|
||||
"tire_degradation_index": 0.72,
|
||||
"ers_charge": 0.78,
|
||||
"fuel_optimization_score": 0.82,
|
||||
"driver_consistency": 0.88,
|
||||
"weather_impact": "low"
|
||||
}
|
||||
],
|
||||
"race_context": {
|
||||
"race_info": {
|
||||
"track_name": "Monaco",
|
||||
"current_lap": 27,
|
||||
"total_laps": 58,
|
||||
"weather_condition": "Dry",
|
||||
"track_temp_celsius": 42
|
||||
},
|
||||
"driver_state": {
|
||||
"driver_name": "Hamilton",
|
||||
"current_position": 4,
|
||||
"current_tire_compound": "medium",
|
||||
"tire_age_laps": 14,
|
||||
"fuel_remaining_percent": 47
|
||||
},
|
||||
"competitors": []
|
||||
}
|
||||
}'
|
||||
```
|
||||
|
||||
### Option 2: AI Layer Fetches Telemetry, Client Provides Context
|
||||
|
||||
```bash
|
||||
# Enrichment service POSTs telemetry to webhook
|
||||
# Then client calls:
|
||||
|
||||
curl -X POST http://localhost:9000/api/strategy/brainstorm \
|
||||
-H "Content-Type: application/json" \
|
||||
-d '{
|
||||
"race_context": {
|
||||
"race_info": {...},
|
||||
"driver_state": {...},
|
||||
"competitors": []
|
||||
}
|
||||
}'
|
||||
```
|
||||
|
||||
AI layer will use telemetry from:
|
||||
1. **Buffer** (if webhook has pushed data) ← CURRENT SETUP
|
||||
2. **GET /enriched** from enrichment service (fallback)
|
||||
|
||||
## Creating a Race Context Template
|
||||
|
||||
Here's a reusable template:
|
||||
|
||||
```json
|
||||
{
|
||||
"race_context": {
|
||||
"race_info": {
|
||||
"track_name": "Monaco",
|
||||
"current_lap": 27,
|
||||
"total_laps": 58,
|
||||
"weather_condition": "Dry",
|
||||
"track_temp_celsius": 42
|
||||
},
|
||||
"driver_state": {
|
||||
"driver_name": "Hamilton",
|
||||
"current_position": 4,
|
||||
"current_tire_compound": "medium",
|
||||
"tire_age_laps": 14,
|
||||
"fuel_remaining_percent": 47
|
||||
},
|
||||
"competitors": [
|
||||
{
|
||||
"position": 1,
|
||||
"driver": "Verstappen",
|
||||
"tire_compound": "hard",
|
||||
"tire_age_laps": 18,
|
||||
"gap_seconds": -12.5
|
||||
},
|
||||
{
|
||||
"position": 2,
|
||||
"driver": "Leclerc",
|
||||
"tire_compound": "medium",
|
||||
"tire_age_laps": 10,
|
||||
"gap_seconds": -5.2
|
||||
},
|
||||
{
|
||||
"position": 3,
|
||||
"driver": "Norris",
|
||||
"tire_compound": "medium",
|
||||
"tire_age_laps": 12,
|
||||
"gap_seconds": -2.1
|
||||
},
|
||||
{
|
||||
"position": 5,
|
||||
"driver": "Sainz",
|
||||
"tire_compound": "soft",
|
||||
"tire_age_laps": 5,
|
||||
"gap_seconds": 3.8
|
||||
}
|
||||
]
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
## Where Does Race Context Come From?
|
||||
|
||||
In a real system, race context typically comes from:
|
||||
|
||||
1. **Timing System** - Official F1 timing data
|
||||
- Current positions
|
||||
- Gap times
|
||||
- Lap numbers
|
||||
|
||||
2. **Team Database** - Historical race data
|
||||
- Track information
|
||||
- Total laps for this race
|
||||
- Weather forecasts
|
||||
|
||||
3. **Pit Wall** - Live observations
|
||||
- Competitor tire strategies
|
||||
- Weather conditions
|
||||
- Track temperature
|
||||
|
||||
4. **Telemetry Feed** - Some data overlaps
|
||||
- Driver's current tires
|
||||
- Tire age
|
||||
- Fuel remaining
|
||||
|
||||
## Recommended Architecture
|
||||
|
||||
```
|
||||
┌─────────────────────┐
|
||||
│ Timing System │
|
||||
│ (Race Control) │
|
||||
└──────────┬──────────┘
|
||||
│
|
||||
▼
|
||||
┌─────────────────────┐ ┌─────────────────────┐
|
||||
│ Frontend/Client │ │ Enrichment Service │
|
||||
│ │ │ (Port 8000) │
|
||||
│ Manages: │ │ │
|
||||
│ - Race context │ │ Manages: │
|
||||
│ - UI state │ │ - Telemetry │
|
||||
│ - User inputs │ │ - HPC enrichment │
|
||||
└──────────┬──────────┘ └──────────┬──────────┘
|
||||
│ │
|
||||
│ │ POST /ingest/enriched
|
||||
│ │ (telemetry only)
|
||||
│ ▼
|
||||
│ ┌─────────────────────┐
|
||||
│ │ AI Layer Buffer │
|
||||
│ │ (telemetry only) │
|
||||
│ └─────────────────────┘
|
||||
│ │
|
||||
│ POST /api/strategy/brainstorm │
|
||||
│ (race_context + telemetry) │
|
||||
└───────────────────────────────┤
|
||||
│
|
||||
▼
|
||||
┌─────────────────────┐
|
||||
│ AI Strategy Layer │
|
||||
│ (Port 9000) │
|
||||
│ │
|
||||
│ Generates 3 │
|
||||
│ strategies │
|
||||
└─────────────────────┘
|
||||
```
|
||||
|
||||
## Python Example: Calling with Race Context
|
||||
|
||||
```python
|
||||
import httpx
|
||||
|
||||
async def get_race_strategies(race_context: dict):
|
||||
"""
|
||||
Get strategies from AI layer.
|
||||
|
||||
Args:
|
||||
race_context: Current race state
|
||||
|
||||
Returns:
|
||||
3 strategies with pit plans and risk assessments
|
||||
"""
|
||||
url = "http://localhost:9000/api/strategy/brainstorm"
|
||||
|
||||
payload = {
|
||||
"race_context": race_context
|
||||
# enriched_telemetry is optional - AI will use buffer or fetch
|
||||
}
|
||||
|
||||
async with httpx.AsyncClient(timeout=60.0) as client:
|
||||
response = await client.post(url, json=payload)
|
||||
response.raise_for_status()
|
||||
return response.json()
|
||||
|
||||
# Usage:
|
||||
race_context = {
|
||||
"race_info": {
|
||||
"track_name": "Monaco",
|
||||
"current_lap": 27,
|
||||
"total_laps": 58,
|
||||
"weather_condition": "Dry",
|
||||
"track_temp_celsius": 42
|
||||
},
|
||||
"driver_state": {
|
||||
"driver_name": "Hamilton",
|
||||
"current_position": 4,
|
||||
"current_tire_compound": "medium",
|
||||
"tire_age_laps": 14,
|
||||
"fuel_remaining_percent": 47
|
||||
},
|
||||
"competitors": []
|
||||
}
|
||||
|
||||
strategies = await get_race_strategies(race_context)
|
||||
print(f"Generated {len(strategies['strategies'])} strategies")
|
||||
```
|
||||
|
||||
## Alternative: Enrichment Service Sends Full Payload
|
||||
|
||||
If you really want enrichment service to send race context too, you'd need to:
|
||||
|
||||
### 1. Store race context in enrichment service
|
||||
|
||||
```python
|
||||
# In hpcsim/api.py
|
||||
_race_context = {
|
||||
"race_info": {...},
|
||||
"driver_state": {...},
|
||||
"competitors": []
|
||||
}
|
||||
|
||||
@app.post("/set_race_context")
|
||||
async def set_race_context(context: Dict[str, Any]):
|
||||
"""Update race context (call this when race state changes)."""
|
||||
global _race_context
|
||||
_race_context = context
|
||||
return {"status": "ok"}
|
||||
```
|
||||
|
||||
### 2. Send both in webhook
|
||||
|
||||
```python
|
||||
# In ingest_telemetry endpoint
|
||||
if _CALLBACK_URL:
|
||||
payload = {
|
||||
"enriched_telemetry": [enriched],
|
||||
"race_context": _race_context
|
||||
}
|
||||
await client.post(_CALLBACK_URL, json=payload)
|
||||
```
|
||||
|
||||
### 3. Update AI webhook to handle full payload
|
||||
|
||||
But this adds complexity. **I recommend keeping it simple**: client provides race_context when calling brainstorm.
|
||||
|
||||
---
|
||||
|
||||
## Current Working Setup
|
||||
|
||||
✅ **Enrichment service** → POSTs telemetry to `/api/ingest/enriched`
|
||||
✅ **AI layer** → Stores telemetry in buffer
|
||||
✅ **Client** → Calls `/api/strategy/brainstorm` with race_context
|
||||
✅ **AI layer** → Uses buffer telemetry + provided race_context → Generates strategies
|
||||
|
||||
This is clean, simple, and follows single responsibility principle!
|
||||
Reference in New Issue
Block a user