Merge pull request #708 from bioimage-io/float_update

Disallow inf and nan values throughout the spec

Author: FynnBe

.vscode/settings.json

@@ -15,5 +15,12 @@
             "bioimageio.yaml",
             "*.bioimageio.yaml"
         ],
-    }
+    },
+    "python-envs.pythonProjects": [
+        {
+            "path": "",
+            "envManager": "ms-python.python:conda",
+            "packageManager": "ms-python.python:conda"
+        }
+    ]
 }

bioimageio/spec/_internal/io_utils.py

@@ -20,7 +20,7 @@
 import numpy
 from loguru import logger
 from numpy.typing import NDArray
-from pydantic import FilePath, NewPath, RootModel
+from pydantic import BaseModel, FilePath, NewPath, RootModel
 from ruyaml import YAML
 from typing_extensions import Unpack
 
@@ -70,7 +70,7 @@ def read_yaml(
 
 
 def write_yaml(
-    content: Union[YamlValue, BioimageioYamlContentView],
+    content: Union[YamlValue, BioimageioYamlContentView, BaseModel],
     /,
     file: Union[NewPath, FilePath, IO[str], IO[bytes], ZipPath],
 ):
@@ -79,6 +79,9 @@ def write_yaml(
     else:
         cm = nullcontext(file)
 
+    if isinstance(content, BaseModel):
+        content = content.model_dump(mode="json")
+
     with cm as f:
         _yaml_dump.dump(content, f)
 

bioimageio/spec/_internal/node.py

@@ -27,15 +27,16 @@ def _node_title_generator(node: Type[Node]) -> str:
 
 class Node(
     pydantic.BaseModel,
+    allow_inf_nan=False,
     extra="forbid",
     frozen=False,
+    model_title_generator=_node_title_generator,
     populate_by_name=True,
     revalidate_instances="never",
+    use_attribute_docstrings=True,
     validate_assignment=True,
     validate_default=False,
     validate_return=True,  # TODO: check if False here would bring a speedup and can still be safe
-    use_attribute_docstrings=True,
-    model_title_generator=_node_title_generator,
 ):
     """"""  # empty docstring to remove all pydantic docstrings from the pdoc spec docs
 

bioimageio/spec/_internal/url.py

@@ -17,6 +17,10 @@ def _validate_url(url: Union[str, pydantic.HttpUrl]) -> pydantic.HttpUrl:
     return _validate_url_impl(url, request_mode="head")
 
 
+_KNOWN_VALID_URLS = ("https://zenodo.org/records/3446812/files/unet2d_weights.torch",)
+"""known valid urls to bypass validation for to avoid sporadic 503 errors in tests etc."""
+
+
 def _validate_url_impl(
     url: Union[str, pydantic.HttpUrl],
     request_mode: Literal["head", "get_stream", "get"],
@@ -30,7 +34,11 @@ def _validate_url_impl(
 
     val_url = url
 
-    if url.startswith("http://example.com") or url.startswith("https://example.com"):
+    if (
+        url.startswith("http://example.com")
+        or url.startswith("https://example.com")
+        or url in _KNOWN_VALID_URLS
+    ):
         return pydantic.HttpUrl(url)
 
     if url.startswith("https://colab.research.google.com/github/"):

bioimageio/spec/model/v0_5.py

@@ -76,6 +76,7 @@
 )
 from .._internal.io_utils import load_array
 from .._internal.node_converter import Converter
+from .._internal.type_guards import is_dict, is_sequence
 from .._internal.types import (
     AbsoluteTolerance,
     LowerCaseIdentifier,
@@ -878,6 +879,27 @@ class IntervalOrRatioDataDescr(Node):
     offset: Optional[float] = None
     """Offset for data on a ratio scale."""
 
+    @model_validator(mode="before")
+    def _replace_inf(cls, data: Any):
+        if is_dict(data):
+            if "range" in data and is_sequence(data["range"]):
+                forbidden = (
+                    "inf",
+                    "-inf",
+                    ".inf",
+                    "-.inf",
+                    float("inf"),
+                    float("-inf"),
+                )
+                if any(v in forbidden for v in data["range"]):
+                    issue_warning("replaced 'inf' value", value=data["range"])
+
+                data["range"] = tuple(
+                    (None if v in forbidden else v) for v in data["range"]
+                )
+
+        return data
+
 
 TensorDataDescr = Union[NominalOrOrdinalDataDescr, IntervalOrRatioDataDescr]
 

changelog.md

@@ -10,6 +10,10 @@ In this file we log both:
 
 This changelog includes implementation details and my reference the [changes to the Resource Description Format](#changes-to-the-resource-description-format), e.g. in entry [bioimageio.spec 0.5.2](#bioimageiospec-052).
 
+#### bioimageio.spec (next release)
+
+- infinity and not-a-number values are no longer allowed (when used in a tensor description under data.range they are replaced with `None`)
+
 #### bioimageio.spec 0.5.4.3
 
 - fix root determination of cacheless downloads of non-zip files

tests/test_bioimageio_collection.py

@@ -26,6 +26,7 @@
     "philosophical-panda/0.1.0": {"outputs"},  # int -> float
     "polite-pig/1.1": {"inputs", "outputs"},
     "charismatic-whale/1.0.2": {"inputs", "outputs"},  # int -> float
+    "dynamic-t-rex/1.1": {"inputs"},  # int -> float
 }
 
 

tests/test_model/test_v0_5.py

@@ -129,7 +129,7 @@ def test_tensor_base_invalid(kwargs: Dict[str, Any]):
         {
             "id": "input_1",
             "description": "Input 1",
-            "data": {"type": "float32"},
+            "data": {"type": "float32", "range": ["-inf", float("inf")]},
             "axes": [
                 dict(type="space", id="x", size=10),
                 dict(type="space", id="y", size=11),
@@ -146,7 +146,7 @@ def test_tensor_base_invalid(kwargs: Dict[str, Any]):
                 }
             ],
             "test_tensor": {"source": UNET2D_ROOT / "test_input.npy"},
-        },
+        }
     ],
 )
 def test_input_tensor(kwargs: Dict[str, Any]):