Handle NaN and Infinity in JSON serialization | Fixes #4144

[sswez02]

Fixes #4144

Problem

When saving a Client to JSON using save_to_json_file(), the operation fails with ValueError: Out of range float values are not JSON compliant because NaN and Infinity float values are not valid in standard JSON format.

Solution

Modified ax/storage/json_store/encoder.py to handle non-JSON-compliant float values:

• Added import math for NaN/Infinity checking
• Added explicit handling for float type that converts NaN and Infinity to None (JSON null)

This follows the same pattern as pandas to_json() which converts NaN to null by default.

Testing

import math
from ax.storage.json_store.encoder import object_to_json

# These now work instead of failing
result = object_to_json({"value": float("nan")})
print(result)  # {"value": None}

result = object_to_json({"value": float("inf")})
print(result)  # {"value": None}

Previous Behavior

ValueError: Out of range float values are not JSON compliant
"map_key_infos": [ { "key": "step", "default_value": NaN } ]

New Behavior

NaN and Infinity values are converted to null in JSON, allowing save/load to work correctly.

Environment

• Ax version: 1.0.0+
• Python version: 3.11+
• OS: All