popov/gogo2
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

T predictions WIP

Browse Source
This commit is contained in:
Dobromir Popov
2025-11-22 01:02:19 +02:00
parent a7def3b788
commit f967f0a142
7 changed files with 200 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
ANNOTATE/core/live_pivot_trainer.py
View File

@@ -65,8 +65,10 @@ class LivePivotTrainer:
# Williams Market Structure for pivot detection
try:
from core.williams_market_structure import WilliamsMarketStructure
self.williams_1s = WilliamsMarketStructure(num_levels=5)
self.williams_1m = WilliamsMarketStructure(num_levels=5)
# Fix: WilliamsMarketStructure.__init__ does not accept num_levels
# It defaults to 5 levels internally
self.williams_1s = WilliamsMarketStructure()
self.williams_1m = WilliamsMarketStructure()
logger.info("Williams Market Structure initialized for pivot detection")
except Exception as e:
logger.error(f"Failed to initialize Williams Market Structure: {e}")

13
ANNOTATE/core/real_training_adapter.py
View File

@@ -1336,13 +1336,16 @@ class RealTrainingAdapter:
if result_1s:
price_data_1s, norm_params_dict['1s'] = result_1s
else:
# Don't fail on missing 1s data, it's often unavailable in annotations
price_data_1s = None
result_1m = self._extract_timeframe_data(timeframes.get('1m', {}), target_seq_len) if '1m' in timeframes else None
if result_1m:
price_data_1m, norm_params_dict['1m'] = result_1m
else:
price_data_1m = None
# Warning: 1m data is critical
logger.warning(f"Missing 1m data for transformer batch (sample: {training_sample.get('test_case_id')})")
return None
result_1h = self._extract_timeframe_data(timeframes.get('1h', {}), target_seq_len) if '1h' in timeframes else None
if result_1h:
@@ -1558,6 +1561,12 @@ class RealTrainingAdapter:
# Model predicts price change ratio, not absolute price
exit_price = training_sample.get('exit_price')
# Handle 'expected_outcome' nesting from LivePivotTrainer
if exit_price is None:
expected_outcome = training_sample.get('expected_outcome', {})
if isinstance(expected_outcome, dict):
exit_price = expected_outcome.get('exit_price')
if exit_price and current_price > 0:
# Normalize: (exit_price - current_price) / current_price
# This gives the expected price change as a ratio
@@ -2547,6 +2556,7 @@ class RealTrainingAdapter:
if session['last_candle_time'] == latest_candle_time:
return # Same candle, no training needed
logger.debug(f"New candle detected: {latest_candle_time} (last: {session['last_candle_time']})")
session['last_candle_time'] = latest_candle_time
# Get the completed candle (second to last)
@@ -2613,6 +2623,7 @@ class RealTrainingAdapter:
# Convert to batch format
batch = self._convert_annotation_to_transformer_batch(training_sample)
if not batch:
logger.warning(f"Per-candle training failed: Could not convert sample to batch")
return
# Train on this batch