dsf-label-sdk 2.0.0

Professional SDK for DSF Label Adaptive Formula API

DSF Label SDK

Accelerate AI development with programmatic data classification. Reduce labeling costs and time with configurable heuristics.

---

Why DSF Label?

Manual data labeling is slow and expensive. This SDK transforms domain expertise into configurable heuristics that classify datasets at scale with confidence scoring.

Built on the DSF evaluation formula (weighted similarity with parameter scaling), ensuring consistency across all DSF products.

---

Core Concepts

Define weighted heuristics based on domain knowledge. The system evaluates data points against these rules, producing classification scores. Uncertain cases get flagged for review.

---

Installation

pip install dsf-label-sdk

---

Quick Start

Community Edition

from dsf_label import LabelSDK

sdk = LabelSDK()

# Build configuration
config = (sdk.create_config()
    .add_field('feature_a', reference=True, params={'importance': 5.0})
    .add_field('feature_b', reference=True, params={'importance': 4.0})
    .add_field('feature_c', reference=50, params={'importance': 2.0})
)

# Classify data point
data = {
    'feature_a': True,
    'feature_b': True,
    'feature_c': 45
}

result = sdk.evaluate(data, config)
print(f"Score: {result.score:.3f}")
print(f"Above threshold: {result.is_above_threshold}")

Note: For boolean fields, reference=True means expected value is True. Similarity computed as exact match.

Professional Edition

import pandas as pd

sdk = LabelSDK(
    license_key='PRO-2026-12-31-XXXX',
    tier='professional'
)

# Batch processing
df = pd.read_csv('unlabeled_data.csv')
results = sdk.batch_evaluate(
    data_points=df.to_dict('records'),
    config=config
)

# Access metrics
metrics = sdk.get_metrics()
print(f"Confidence level: {metrics['confidence_level']:.3f}")
print(f"Average score: {metrics['avg_score']:.3f}")

Enterprise Edition

sdk = LabelSDK(
    license_key='ENT-2026-12-31-XXXX',
    tier='enterprise',
    mode='adaptive'
)

# Process large datasets with auto-calibration
for batch in dataset_batches:
    results = sdk.batch_evaluate(batch, config)

# View adaptation metrics
metrics = sdk.get_metrics()
print(f"Evaluations: {metrics['evaluations']}")
print(f"Auto-calibrated: {metrics.get('auto_calibration', False)}")

---

Config Builder

sdk = LabelSDK()

# Chained calls
config = (sdk.create_config()
    .add_field('metric_a', reference=20, params={'importance': 3.0})
    .add_field('metric_b', reference=1.0, params={'importance': 2.5})
)

# Sequential calls
config = sdk.create_config()
for field_name, params in field_definitions.items():
    config.add_field(field_name, **params)

---

Context Manager

with LabelSDK(license_key='...', tier='enterprise') as sdk:
    result = sdk.evaluate(data, config)
    metrics = sdk.get_metrics()

---

Error Handling

from dsf_label import LabelSDK, LicenseError, ValidationError

try:
    sdk = LabelSDK(license_key='invalid', tier='professional')
    result = sdk.evaluate(data, config)
    
except LicenseError as e:
    sdk = LabelSDK()  # Fallback to community
    
except ValidationError as e:
    print(f"Invalid configuration: {e}")

---

Tier Comparison

Feature	Community	Professional	Enterprise
Classifications/month	Unlimited†	Unlimited	Unlimited
Single Evaluation	✅	✅	✅
Batch Processing	❌	✅	✅
DataFrame Support	❌	✅	✅
Adaptive Thresholds	❌	✅	✅
Performance Metrics	❌	✅	✅ Enhanced
Auto-Calibration	❌	❌	✅
Adaptive Modes	❌	❌	✅
Support	Community	Email	Priority SLA

†Subject to fair-use policies. Community tier free for evaluation. Production requires registration.

---

Enterprise Features

Auto-Calibration (Enterprise)

System optimizes heuristic parameters based on data patterns.

Adaptation Modes (Enterprise)

# Standard: Full history
sdk = LabelSDK(tier='enterprise', mode='standard')

# Adaptive: Recent patterns priority
sdk = LabelSDK(tier='enterprise', mode='adaptive')

Cache Management (Enterprise)

sdk.invalidate_cache()  # Reset when patterns change

---

Configuration Guidelines

config = {
    'field_name': {
        'reference': expected_value,
        'params': {'importance': <float>}  # 0.0-5.0
    }
}

Internally: All field contributions computed using DSF evaluation formula (weighted similarity with parameter scaling), ensuring consistency across all DSF SDKs.

---

Hybrid Integration

Integrate ML models as additional heuristics:

# Example: Load pre-trained models
bert_classifier = your_model_loader('sentiment')  # Conceptual
xgboost_model = your_model_loader('risk')

# Hybrid configuration
config = {
    'metric_standard': {'reference': 100, 'params': {'importance': 3.0}},
    'bert_score': {'reference': 0.8, 'params': {'importance': 4.0}},
    'xgb_score': {'reference': 0.3, 'params': {'importance': 4.5}}
}

# Process with ensemble
def process_item(item_data):
    bert_pred = bert_classifier(item_data['text'])
    xgb_pred = xgboost_model.predict(item_data['features'])
    
    hybrid_data = {
        'metric_standard': item_data['value'],
        'bert_score': bert_pred,
        'xgb_score': xgb_pred
    }
    
    return sdk.evaluate(hybrid_data, config)

---

API Reference

LabelSDK

Methods:

• __init__(tier='community', license_key=None, mode='standard')
• evaluate(data, config) - Single evaluation
• batch_evaluate(data_points, config) - Batch processing (Pro/Enterprise)
• create_config() - Config builder
• get_metrics() - Performance statistics (Pro/Enterprise)
• set_confidence_level(level) - Threshold adjustment
• invalidate_cache() - Reset cache (Enterprise)

EvaluationResult

Attributes:

• score (float): Confidence score [0.0-1.0]
• tier (str): License tier
• confidence_level (float): Current threshold
• is_above_threshold (bool): Threshold comparison
• metrics (dict): Metrics (Pro/Enterprise)

ConfigBuilder

Methods:

• add_field(name, reference, params={'importance': 1.0})
• remove_field(name)
• build()

---

Performance Metrics (Pro/Enterprise)

metrics = sdk.get_metrics()

print(f"Evaluations: {metrics['evaluations']}")
print(f"Average score: {metrics['avg_score']:.3f}")
print(f"Confidence level: {metrics['confidence_level']:.3f}")

# Enterprise metrics
if metrics.get('auto_calibration'):
    print(f"Adapted fields: {metrics['adapted_fields']}")

---

Migration Example

Before:

for data_point in dataset:
    label = human_annotator.label(data_point)
    labeled_data.append((data_point, label))

After:

for data_point in dataset:
    result = sdk.evaluate(data_point, config)
    if result.score > 0.75:
        labeled_data.append((data_point, 'POSITIVE'))
    elif result.score < 0.35:
        labeled_data.append((data_point, 'NEGATIVE'))
    else:
        review_queue.append(data_point)

---

Use Cases

Classification Tasks

config = {
    'indicator_a': {'reference': True, 'params': {'importance': 5.0}},
    'metric_b': {'reference': 0.8, 'params': {'importance': 3.0}},
    'count_c': {'reference': 2, 'params': {'importance': 2.5}}
}

Quality Assessment

config = {
    'quality_metric': {'reference': 0.8, 'params': {'importance': 4.0}},
    'completeness': {'reference': 1.0, 'params': {'importance': 3.0}},
    'consistency': {'reference': 0.9, 'params': {'importance': 3.5}}
}

---

FAQ

How accurate are classifications?
Accuracy depends on heuristic design. Well-designed configs achieve comparable results to human annotators at scale.

Can I use with active learning?
Yes. Use confidence scores to identify uncertain examples for human review.

Difference between modes?

• Standard: Uses full history for adaptation
• Adaptive: Prioritizes recent patterns (Enterprise only)

When to invalidate cache?
When data distribution changes significantly (new categories, seasonal shifts, different sources).

---

⚠️ Important Notes

Client Responsibility:
Clients must validate classifications and compliance. This SDK is a classification support tool and does not make autonomous decisions.

Data Processing:
All logic executes server-side. SDK exposes configuration interface only.

Model Outputs:
ML model inputs are client-provided and client-controlled.

---

Support

Documentation: https://dsfuptech.cloud
Issues: [e-mail]
Licensing: [e-mail]

---

Licensing

License Format:

• Professional: PRO-YYYY-MM-DD-XXXX-XXXX
• Enterprise: ENT-YYYY-MM-DD-XXXX-XXXX

Contact: [e-mail]

---

📋 Credits

Technology Architect: Jaime Alexander Jimenez

---

© 2025 DSF UpTech. All rights reserved.