signal-classification

Name: signal-classification
Author: agiprolabs

235stars53forks

How to install signal-classification?

agiprolabs/claude-trading-skills/signal-classification

$npx -y skills add agiprolabs/claude-trading-skills --skill signal-classification

Installs into the current project.

Summary

ML trading signal classifiers using XGBoost and LightGBM with walk-forward validation, SHAP feature importance, and threshold optimization

Files · 1

View on GitHub

SKILL.md

1	# Signal Classification
2
3	Predict whether an asset's price will move up or down over a forward horizon using supervised machine learning classifiers. This skill covers the full pipeline: label creation, model training, walk-forward validation, feature importance analysis, and threshold optimization for trading applications.
4
5	## Why Tree-Based Models Dominate Trading ML
6
7	XGBoost and LightGBM are the workhorses of quantitative trading ML for good reason:
8
9	- Non-linear relationships: Financial features interact in complex, non-linear ways that trees capture naturally
10	- Robust to feature scale: No need to normalize or standardize inputs — trees split on rank order
11	- Built-in feature importance: Understand which features drive predictions without separate analysis
12	- Fast training and inference: Train on thousands of samples in seconds, predict in microseconds
13	- Handle missing values: Native support for NaN without imputation hacks
14	- Regularization built in: max_depth, min_child_weight, subsample all prevent overfitting
15
16	Linear models and deep learning have their place, but for tabular trading features with fewer than 100k samples, gradient-boosted trees consistently outperform alternatives.
17
18	## Classification Types
19
20	### Binary Classification
21
22	The simplest and most common setup. Predict whether forward returns exceed a threshold:
23
24	- Up signal: forward return > +1%
25	- Down signal: forward return < -1%
26	- Neutral (excluded): -1% to +1% — drop these from training to create cleaner labels
27
28	```python
29	import numpy as np
30
31	def create_binary_labels(
32	prices: np.ndarray, horizon: int = 24, threshold: float = 0.01
33	) -> np.ndarray:
34	"""Create binary labels from forward returns.
35
36	Args:
37	prices: Array of prices.
38	horizon: Forward return lookback in bars.
39	threshold: Minimum return magnitude for a label.
40
41	Returns:
42	Array of labels: 1 (up), 0 (down), NaN (neutral).
43	"""
44	fwd_returns = np.roll(prices, -horizon) / prices - 1
45	fwd_returns[-horizon:] = np.nan
46	labels = np.where(fwd_returns > threshold, 1,
47	np.where(fwd_returns < -threshold, 0, np.nan))
48	return labels
49	```
50
51	### Multi-Class Classification
52
53	Three classes for finer signal granularity:
54
55	\| Class \| Condition \| Typical threshold \|
56	\|-------\|-----------\|-------------------\|
57	\| Strong Up \| fwd_return > +2% \| High confidence long \|
58	\| Mild Up \| +0.5% to +2% \| Moderate confidence \|
59	\| Down \| fwd_return < -0.5% \| Avoid / short \|
60
61	Multi-class reduces per-class sample size. Use only with large datasets (1000+ samples per class).
62
63	### Probability Calibration
64
65	Raw model probabilities from XGBoost/LightGBM are not well-calibrated. A predicted 0.7 probability does not mean 70% chance of being correct. Use calibration to fix this:
66
67	```python
68	from sklearn.calibration import CalibratedClassifierCV
69
70	calibrated = CalibratedClassifierCV(base_model, cv=5, method="isotonic")
71	calibrated.fit(X_train, y_train)
72	probs = calibrated.predict_proba(X_test)[:, 1]
73	```
74
75	Isotonic calibration works better than Platt scaling for tree models.
76
77	## Walk-Forward Validation
78
79	This is the single most important concept in trading ML. Standard cross-validation randomly shuffles data, which creates lookahead bias. Walk-forward validation respects time ordering.
80
81	### How It Works
82
83	```
84	Window 1: [===TRAIN===][GAP][=TEST=]
85	Window 2: [===TRAIN===][GAP][=TEST=]
86	Window 3: [===TRAIN===][GAP][=TEST=]
87	Window 4: [===TRAIN===][GAP][=TEST=]
88	```
89
90	Each window:
91	1. Train on past N bars
92	2. Skip a gap (embargo) equal to the forward return horizon
93	3. Predict on next M bars
94	4. Record out-of-sample predictions
95	5. Slide forward and repeat
96
97	### Typical Parameters
98
99	\| Parameter \| Value \| Rationale \|
100	\|-----------\|-------\|-----------\|
101	\| Train window \| 30 days (720 hourly bars) \| Enough data to learn, recent enough to be relevant \|
102	\| Test window \| 7 days (168 hourly bars) \| Enough predictions for statistical significance \|
103	\| Step size \| 1 day (24 bars) \| Overlap test windows for more data points \|
104	\| Gap (embargo) \| Same as forward horizon \| Prevents label leakage \|
105
106	### Walk-Forward Implementation
107
108	```python
109	from typing import Iterator
110
111	def walk_forward_splits(
112	n_samples: int,
113	train_size: int = 720,
114	test_size: int = 168,
115	step_size: int = 24,
116	gap: int = 24,
117	) -> Iterator[tuple[np.ndarray, np.ndarray]]:
118	"""Generate walk-forward train/test index splits.
119
120	Args:
121	n_samples: Total number of samples.
122	train_size: Number of training samples per window.
123	test_size: Number of test samples per window.
124	step_size: Step between successive windows.
125	gap: Gap between train end and test start.
126
127	Yields:
128	Tuples of (train_indices, test_indices).
129	"""
130	start = 0
131	while start + tr

People who liked this also viewed

6 picks

Type

azure-costAzure cost management: query costs, forecast spending, optimize to reduce waste. WHEN: \"Azure costs\", \"Azure bill\", \"cost breakdown\", \"how much am I spending\", \"forecast spending\", \"optimizSkillsJul 2026275k1.3k
pricingWhen the user wants help with pricing decisions, packaging, or monetization strategy. Also use when the user mentions 'pricing,' 'pricing tiers,' 'freemium,' 'free trial,' 'packaging,' 'price increaseSkillsJul 202638k41k
stripe-integrationMulti-harness agentic plugin marketplace for Claude Code, Codex CLI, Cursor, OpenCode, GitHub Copilot, and Gemini CLISkillsJul 202611k38k
startup-financial-modelingMulti-harness agentic plugin marketplace for Claude Code, Codex CLI, Cursor, OpenCode, GitHub Copilot, and Gemini CLISkillsJul 20269.5k38k
tushare-finance专为 Claude Code 设计的 Tushare Pro 金融数据技能包，支持 220+ 个接口，用自然语言获取 A 股行情、财务数据、宏观经济等金融数据。SkillsJul 20265.0k75
azure-pricingFetches real-time Azure retail pricing using the Azure Retail Prices API (prices.azure.com) and estimates Copilot Studio agent credit consumption. Use when the user asks about the cost of any Azure seSkillsJul 20264.5k37k

signal-classification

235stars53forks

View on GitHub

How to install signal-classification?

agiprolabs/claude-trading-skills/signal-classification

$npx -y skills add agiprolabs/claude-trading-skills --skill signal-classification

Installs into the current project.

Summary

ML trading signal classifiers using XGBoost and LightGBM with walk-forward validation, SHAP feature importance, and threshold optimization

Files · 1

View on GitHub

SKILL.md

1	# Signal Classification
2
3	Predict whether an asset's price will move up or down over a forward horizon using supervised machine learning classifiers. This skill covers the full pipeline: label creation, model training, walk-forward validation, feature importance analysis, and threshold optimization for trading applications.
4
5	## Why Tree-Based Models Dominate Trading ML
6
7	XGBoost and LightGBM are the workhorses of quantitative trading ML for good reason:
8
9	- Non-linear relationships: Financial features interact in complex, non-linear ways that trees capture naturally
10	- Robust to feature scale: No need to normalize or standardize inputs — trees split on rank order
11	- Built-in feature importance: Understand which features drive predictions without separate analysis
12	- Fast training and inference: Train on thousands of samples in seconds, predict in microseconds
13	- Handle missing values: Native support for NaN without imputation hacks
14	- Regularization built in: max_depth, min_child_weight, subsample all prevent overfitting
15
16	Linear models and deep learning have their place, but for tabular trading features with fewer than 100k samples, gradient-boosted trees consistently outperform alternatives.
17
18	## Classification Types
19
20	### Binary Classification
21
22	The simplest and most common setup. Predict whether forward returns exceed a threshold:
23
24	- Up signal: forward return > +1%
25	- Down signal: forward return < -1%
26	- Neutral (excluded): -1% to +1% — drop these from training to create cleaner labels
27
28	```python
29	import numpy as np
30
31	def create_binary_labels(
32	prices: np.ndarray, horizon: int = 24, threshold: float = 0.01
33	) -> np.ndarray:
34	"""Create binary labels from forward returns.
35
36	Args:
37	prices: Array of prices.
38	horizon: Forward return lookback in bars.
39	threshold: Minimum return magnitude for a label.
40
41	Returns:
42	Array of labels: 1 (up), 0 (down), NaN (neutral).
43	"""
44	fwd_returns = np.roll(prices, -horizon) / prices - 1
45	fwd_returns[-horizon:] = np.nan
46	labels = np.where(fwd_returns > threshold, 1,
47	np.where(fwd_returns < -threshold, 0, np.nan))
48	return labels
49	```
50
51	### Multi-Class Classification
52
53	Three classes for finer signal granularity:
54
55	\| Class \| Condition \| Typical threshold \|
56	\|-------\|-----------\|-------------------\|
57	\| Strong Up \| fwd_return > +2% \| High confidence long \|
58	\| Mild Up \| +0.5% to +2% \| Moderate confidence \|
59	\| Down \| fwd_return < -0.5% \| Avoid / short \|
60
61	Multi-class reduces per-class sample size. Use only with large datasets (1000+ samples per class).
62
63	### Probability Calibration
64
65	Raw model probabilities from XGBoost/LightGBM are not well-calibrated. A predicted 0.7 probability does not mean 70% chance of being correct. Use calibration to fix this:
66
67	```python
68	from sklearn.calibration import CalibratedClassifierCV
69
70	calibrated = CalibratedClassifierCV(base_model, cv=5, method="isotonic")
71	calibrated.fit(X_train, y_train)
72	probs = calibrated.predict_proba(X_test)[:, 1]
73	```
74
75	Isotonic calibration works better than Platt scaling for tree models.
76
77	## Walk-Forward Validation
78
79	This is the single most important concept in trading ML. Standard cross-validation randomly shuffles data, which creates lookahead bias. Walk-forward validation respects time ordering.
80
81	### How It Works
82
83	```
84	Window 1: [===TRAIN===][GAP][=TEST=]
85	Window 2: [===TRAIN===][GAP][=TEST=]
86	Window 3: [===TRAIN===][GAP][=TEST=]
87	Window 4: [===TRAIN===][GAP][=TEST=]
88	```
89
90	Each window:
91	1. Train on past N bars
92	2. Skip a gap (embargo) equal to the forward return horizon
93	3. Predict on next M bars
94	4. Record out-of-sample predictions
95	5. Slide forward and repeat
96
97	### Typical Parameters
98
99	\| Parameter \| Value \| Rationale \|
100	\|-----------\|-------\|-----------\|
101	\| Train window \| 30 days (720 hourly bars) \| Enough data to learn, recent enough to be relevant \|
102	\| Test window \| 7 days (168 hourly bars) \| Enough predictions for statistical significance \|
103	\| Step size \| 1 day (24 bars) \| Overlap test windows for more data points \|
104	\| Gap (embargo) \| Same as forward horizon \| Prevents label leakage \|
105
106	### Walk-Forward Implementation
107
108	```python
109	from typing import Iterator
110
111	def walk_forward_splits(
112	n_samples: int,
113	train_size: int = 720,
114	test_size: int = 168,
115	step_size: int = 24,
116	gap: int = 24,
117	) -> Iterator[tuple[np.ndarray, np.ndarray]]:
118	"""Generate walk-forward train/test index splits.
119
120	Args:
121	n_samples: Total number of samples.
122	train_size: Number of training samples per window.
123	test_size: Number of test samples per window.
124	step_size: Step between successive windows.
125	gap: Gap between train end and test start.
126
127	Yields:
128	Tuples of (train_indices, test_indices).
129	"""
130	start = 0
131	while start + tr

People who liked this also viewed

6 picks

Type