JSON encoding of non-finite floats

src/pandas_profiling/config.py

@@ -180,6 +180,16 @@ class Html(BaseModel):
     full_width: bool = False
 
 
+# Describes how non-finite numbers (such as NaN or Infinity) should be encoded in the JSON output
+class JsonNonFiniteEncoding(Enum):
+    # Use the default python behaviour, which violates the official JSON standard
+    PYTHON_NATIVE = 0
+    # Encode non-finite numbers as null values
+    NULL = 1
+    # Encode non-finite numbers as null values
+    STRING = 2
+
+
 class Duplicates(BaseModel):
     head: int = 10
     key: str = "# duplicates"
@@ -293,6 +303,9 @@ class Config:
     n_freq_table_max: int = 10
     n_extreme_obs: int = 10
 
+    # JSON output
+    json_non_finite_encoding: JsonNonFiniteEncoding = JsonNonFiniteEncoding.NULL
+
     # Report rendering
     report: Report = Report()
     html: Html = Html()

src/pandas_profiling/profile_report.py

@@ -1,5 +1,6 @@
 import copy
 import json
+import math
 import warnings
 from pathlib import Path
 from typing import Any, Dict, Optional, Union
@@ -10,7 +11,13 @@
 from tqdm.auto import tqdm
 from visions import VisionsTypeset
 
-from pandas_profiling.config import Config, PandasSettings, Settings, SparkSettings
+from pandas_profiling.config import (
+    Config,
+    PandasSettings,
+    Settings,
+    SparkSettings,
+    JsonNonFiniteEncoding
+)
 from pandas_profiling.expectations_report import ExpectationsReport
 from pandas_profiling.model.alerts import AlertType
 from pandas_profiling.model.describe import describe as describe_df
@@ -335,8 +342,24 @@ def encode_it(o: Any) -> Any:
             if isinstance(o, dict):
                 return {encode_it(k): encode_it(v) for k, v in o.items()}
             else:
-                if isinstance(o, (bool, int, float, str)):
+                if isinstance(o, (bool, int, str)):
                     return o
+                elif isinstance(o, float):
+                    if not math.isfinite(o):
+                        # Special handling for non-finite floats.
+                        # This is necessary because JSON does not support NaN/Infinity values.
+                        # The default in Python is to generate invalid JSON.
+                        # Depending on the configuration, we can encode them as null values,
+                        # stringify the non-finite value, or output it as is to keep the default
+                        # Python behaviour.
+                        if self.config.json_non_finite_encoding == JsonNonFiniteEncoding.NULL:
+                            return None
+                        elif self.config.json_non_finite_encoding == JsonNonFiniteEncoding.STRING:
+                            return str(o)
+                        else:
+                            return o
+                    else:
+                        return o
                 elif isinstance(o, list):
                     return [encode_it(v) for v in o]
                 elif isinstance(o, set):