Validation

Validation#

Pre-flight checks for input, model endpoints, and output destinations.

Overview#

The SDK provides comprehensive validation to catch configuration errors before running expensive inference jobs.

Validation checks:

Input - Imagery availability, band compatibility, resolution
Model Endpoint - Connectivity and authentication
Output - GeoCatalog and blob storage access

When Validation Runs#

Validation happens automatically in:

model.run() - Always validates before execution
model.estimate() - Validates before estimating job scope

Optional: Explicit validation with model.validate_input()

Validation Method#

validation = await model.validate_input(
    input=input_source,
    constraint=constraint,
    params=params,
    output=output  # Optional: validates output if provided
)

if validation.is_valid:
    print("✅ Ready to run!")
else:
    for error in validation.errors:
        print(f" {error}")

What Gets Validated#

Input Validation#

Imagery Availability:

Queries STAC catalog to verify data exists for AOI
Reports number of imagery items found
Detects resolution from STAC metadata

Band Compatibility:

Checks collection has required bands (e.g., RGB for EOOS/MARS)
Validates against model specifications

Resolution Compatibility:

Verifies resolution matches model requirements
Warns if suboptimal but acceptable

Collection Support:

Checks if model supports the specified collection
Validates filter compatibility

Model Endpoint Validation#

Connectivity:

Tests that endpoint URL is reachable
Detects network/DNS issues

Authentication:

Validates API key or Azure credential works
Catches 401/403 errors before running
Prevents the “wrong token but validation passed” issue!

Output Validation#

GeoCatalog Access:

Tests endpoint is reachable
Validates credential has write permissions

Blob Storage Access:

Tests storage account accessibility
Validates credential can list containers
Checks if blob container exists or can be created

Validation Results#

ValidationResult Object#

class ValidationResult:
    is_valid: bool              # Overall validation status
    errors: List[str]           # List of errors (empty if valid)
    warnings: List[str]         # Non-fatal warnings
    detected_resolution: float  # Detected imagery resolution (m/pixel)
    bands_found: List[str]      # Available bands in imagery
    stac_items_count: int       # Number of STAC items found

Example Response#

ValidationResult(
    is_valid=True,
    errors=[],
    warnings=["Resolution 0.3m differs from optimal 0.6m"],
    detected_resolution=0.3,
    bands_found=['red', 'green', 'blue'],
    stac_items_count=2
)

Common Validation Errors#

Input Errors#

No imagery found:

❌ No STAC items found for collection 'naip' in specified AOI/datetime

Fix: Adjust datetime range or check AOI coordinates

Missing bands:

❌ Collection 'sentinel-2-l2a' missing required bands: ['red']. Found: ['B02', 'B03', 'B04']

Fix: Use correct band names or choose different collection

Unsupported collection:

❌ Model 'EOOS' does not support collection 'landsat-c2-l2'. Supported: ['naip']

Fix: Use a supported collection

Model Endpoint Errors#

Authentication failure:

❌ Model endpoint authentication failed: Authentication failed (401 Unauthorized). Check your credential/API key.

Fix: Verify API key or Azure credential has proper role assignments

Cannot reach endpoint:

❌ Model endpoint authentication failed: Cannot connect to endpoint: [connection error details]

Fix: Check endpoint URL, network connectivity, firewall rules

Output Errors#

GeoCatalog access denied:

❌ GeoCatalog authentication failed: Credential failed to obtain access token

Fix: Run az login or check credential has access to GeoCatalog

Blob storage access denied:

❌ Blob storage access denied: [error]. Ensure your identity has 'Storage Blob Data Contributor' role.

Fix: Assign required RBAC roles to your identity

Examples#

Basic Validation#

import geoai
from azure.identity import DefaultAzureCredential

# Configure
input_source = geoai.Input(collection="naip")
constraint = geoai.Constraint(
    bbox=[-122.5, 37.5, -122.0, 38.0],
    datetime="2020-01-01/2024-12-31"
)

model = geoai.models.EOOS(
    endpoint="https://eoos.eastus.inference.ml.azure.com/score",
    credential="your-api-key"
)

# Validate before running
validation = await model.validate_input(
    input=input_source,
    constraint=constraint,
    params={"chip_size": 1024, "threshold": 0.5}
)

if validation.is_valid:
    print(f"✅ Found {validation.stac_items_count} imagery items")
    print(f"📏 Resolution: {validation.detected_resolution}m/pixel")
else:
    print("❌ Validation failed:")
    for error in validation.errors:
        print(f"   {error}")

Validate with Output#

# Also validate output destination
output = geoai.Output(
    geocatalog_uri="https://geocatalog.contoso.com",
    collection_name="detections",
    credential=DefaultAzureCredential(),
    storage_url="https://myaccount.blob.core.windows.net",
    blob_container="results"
)

validation = await model.validate_input(
    input=input_source,
    constraint=constraint,
    params=params,
    output=output  # Validates GeoCatalog + blob storage access
)

if validation.is_valid:
    print("✅ Input, model endpoint, and output all validated!")

Handle Warnings#

if validation.is_valid:
    if validation.warnings:
        print("⚠️  Warnings:")
        for warning in validation.warnings:
            print(f"   {warning}")
        
        # Decide whether to proceed
        proceed = input("Continue anyway? (y/n): ")
        if proceed.lower() != 'y':
            exit(0)
    
    # Run model
    result = await model.run(...)

Best Practices#

Always validate on new configurations - First time using a new AOI, collection, or endpoint
Validate output - Pass output parameter to catch destination issues early
Check warnings - Non-fatal but may indicate suboptimal configuration
Use validation to debug - If run() fails, explicit validation provides clearer error messages