forecasting-tools 0.2.88__tar.gz → 0.2.89__tar.gz

{forecasting_tools-0.2.88 → forecasting_tools-0.2.89}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: forecasting-tools
-Version: 0.2.88
+Version: 0.2.89
 Summary: AI forecasting and research tools to help humans reason about and forecast the future
 License: MIT
 License-File: LICENSE
@@ -26,19 +26,19 @@ Requires-Dist: asknews (>=0.9.1,<0.14.0)
 Requires-Dist: asyncio (>=3.0.0,<5.0.0)
 Requires-Dist: exceptiongroup (>=1.2.2,<2.0.0)
 Requires-Dist: faker (>=37.0.0,<41.0.0)
-Requires-Dist: hyperbrowser (>=0.53.0,<0.84.0)
-Requires-Dist: litellm (>=1.59.1,<1.82.3,!=1.76.*,!=1.77.*)
+Requires-Dist: hyperbrowser (>=0.53.0,<1.0.0)
+Requires-Dist: litellm (>=1.59.1,<2.0.0,!=1.76.*,!=1.77.*,!=1.82.7,!=1.82.8)
 Requires-Dist: nest-asyncio (>=1.5.8,<2.0.0)
 Requires-Dist: numpy (>=1.26.0,<3.0.0)
 Requires-Dist: openai (>=1.51.0,<3.0.0)
-Requires-Dist: openai-agents[litellm] (>=0.2.0,<0.10.0)
+Requires-Dist: openai-agents[litellm] (>=0.2.0,<0.14.0)
 Requires-Dist: pandas (>=2.2.3,<4.0.0)
 Requires-Dist: pendulum (>=3.1.0,<4.0.0)
-Requires-Dist: pillow (>=9.0.0,<12.0.0)
+Requires-Dist: pillow (>=9.0.0,<13.0.0)
 Requires-Dist: plotly (>=5.24.1,<7.0.0)
 Requires-Dist: pydantic (>=2.9.2,<3.0.0)
 Requires-Dist: python-dotenv (>=1.0.0,<2.0.0)
-Requires-Dist: regex (>=2024.11.6,<2026.0.0)
+Requires-Dist: regex (>=2024.11.6,<2027.0.0)
 Requires-Dist: requests (>=2.32.3,<3.0.0)
 Requires-Dist: scikit-learn (>=1.5.2,<2.0.0)
 Requires-Dist: streamlit (>=1.20.0,<2.0.0)

{forecasting_tools-0.2.88 → forecasting_tools-0.2.89}/forecasting_tools/data_models/numeric_report.py

@@ -84,8 +84,11 @@ class DatePercentile(BaseModel):
 
 class NumericDistribution(BaseModel):
     declared_percentiles: list[Percentile]
-    open_upper_bound: bool
-    open_lower_bound: bool
+    open_upper_bound: bool  # If False, cdf[-1] must be 1.0 (no mass above upper bound)
+    open_lower_bound: bool  # If False, cdf[0] must be 0.0 (no mass strictly below lower
+    # bound). Note: cdf[0] = P(outcome < lower_bound), so a closed lower bound does not
+    # prevent assigning probability to the outcome equalling lower_bound exactly —
+    # that mass goes in the first inbound bucket: cdf[1] - cdf[0].
     upper_bound: float
     lower_bound: float
     zero_point: float | None
@@ -344,12 +347,22 @@ class NumericDistribution(BaseModel):
         between upper and lower bound (taking into account probability assigned above and below the bounds)
         that is compatible with Metaculus questions.
 
-        cdf stands for 'continuous distribution function'
+        cdf stands for 'cumulative distribution function'
 
         At Metaculus CDFs are often represented with 201 points. Each point has:
-        - percentile ("X% of values are below this point". This is the y axis of the cdf graph)
+        - percentile (the y axis of the cdf graph — see boundary notes below)
         - 'value' or 'nominal location' (The real world number that answers the question)
         - cdf location (a number between 0 and 1 representing where the point is on the cdf x axis, where 0 is range min, and 1 is range max)
+
+        Important boundary semantics (note the asymmetry):
+        - cdf[0].percentile  = P(outcome < lower_bound)  — strictly less than (not equal to)
+        - cdf[-1].percentile = P(outcome <= upper_bound) — less than or equal to
+        For questions with a closed lower bound, cdf[0].percentile must be 0.0 because
+        outcomes strictly below the lower bound are impossible. This does NOT mean zero
+        probability at the lower bound itself — probability of the outcome equalling
+        lower_bound belongs in the first inbound bucket: cdf[1].percentile - cdf[0].percentile.
+        For example, to express an 80% chance the outcome is exactly lower_bound, set
+        cdf[1].percentile = 0.8 (and cdf[0].percentile = 0.0).
         """
 
         cdf_size = self.cdf_size or NumericDefaults.DEFAULT_CDF_SIZE
@@ -527,6 +540,11 @@ class NumericDistribution(BaseModel):
         - caps the maximum growth to 0.2
 
         Note, thresholds change with different `inbound_outcome_count`s
+
+        Boundary convention: cdf[0] is P(outcome < lower_bound) — strictly less than.
+        For closed lower bounds this is forced to 0.0 after standardization; probability
+        of landing exactly on lower_bound belongs in the first inbound bucket (cdf[1]).
+        cdf[-1] is P(outcome <= upper_bound) — less than or equal to (standard convention).
         """
 
         lower_open = self.open_lower_bound

{forecasting_tools-0.2.88 → forecasting_tools-0.2.89}/forecasting_tools/data_models/questions.py

@@ -474,8 +474,10 @@ class DateQuestion(MetaculusQuestion, BoundedQuestionMixin):
     question_type: Literal["date"] = "date"
     upper_bound: datetime
     lower_bound: datetime
-    open_upper_bound: bool
-    open_lower_bound: bool
+    open_upper_bound: bool  # If False, cdf[-1] must be 1.0 (no mass above upper bound)
+    open_lower_bound: bool  # If False, cdf[0] must be 0.0. Note: cdf[0] = P(outcome <
+    # lower_bound) strictly, so probability of landing exactly on lower_bound belongs
+    # in the first inbound bucket (cdf[1]), not in cdf[0].
     zero_point: float | None = None
     cdf_size: int = 201
 
@@ -530,8 +532,10 @@ class NumericQuestion(MetaculusQuestion, BoundedQuestionMixin):
     question_type: Literal["numeric"] = "numeric"
     upper_bound: float
     lower_bound: float
-    open_upper_bound: bool
-    open_lower_bound: bool
+    open_upper_bound: bool  # If False, cdf[-1] must be 1.0 (no mass above upper bound)
+    open_lower_bound: bool  # If False, cdf[0] must be 0.0. Note: cdf[0] = P(outcome <
+    # lower_bound) strictly, so probability of landing exactly on lower_bound belongs
+    # in the first inbound bucket (cdf[1]), not in cdf[0].
     zero_point: float | None = None
     cdf_size: int = (
         201  # Normal numeric questions have 201 points, but discrete questions have fewer

{forecasting_tools-0.2.88 → forecasting_tools-0.2.89}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "forecasting-tools"
-version = "0.2.88"
+version = "0.2.89"
 description = "AI forecasting and research tools to help humans reason about and forecast the future"
 authors = ["Benjamin Wilson <ben@metaculus.com>"]
 license = "MIT"
@@ -33,12 +33,12 @@ nest-asyncio = "^1.5.8"
 requests = "^2.32.3"
 pydantic = "^2.9.2"
 python-dotenv = "^1.0.0"
-regex = ">=2024.11.6,<2026.0.0"
+regex = ">=2024.11.6,<2027.0.0"
 tenacity = ">=8.0.0,<10.0.0"
-pillow = ">=9.0.0,<12.0.0"
+pillow = ">=9.0.0,<13.0.0"
 typeguard = "^4.3.0"
 scikit-learn = "^1.5.2"
-litellm = ">=1.59.1,<1.82.3,!=1.76.*,!=1.77.* || >=1.83.1"
+litellm = ">=1.59.1,<2.0.0,!=1.76.*,!=1.77.*,!=1.82.7,!=1.82.8"
 numpy = ">=1.26.0,<3.0.0"
 exceptiongroup = "^1.2.2"
 faker = ">=37.0.0,<41.0.0"
@@ -47,15 +47,15 @@ pandas = ">=2.2.3,<4.0.0"
 streamlit = "^1.20.0"
 asknews = ">=0.9.1,<0.14.0"
 unidecode = "^1.4.0"
-hyperbrowser = ">=0.53.0,<0.84.0"
+hyperbrowser = ">=0.53.0,<1.0.0"
 pendulum = "^3.1.0"
-openai-agents = {extras = ["litellm"], version = ">=0.2.0,<0.10.0"}
+openai-agents = {extras = ["litellm"], version = ">=0.2.0,<0.14.0"}
 
 [tool.poetry.group.dev.dependencies]
 time-machine = ">=2.19.0,<4.0.0"
 pre-commit = "^4.0.1"
 matplotlib = "^3.9.2"
-ipykernel = "^6.29.5"
+ipykernel = ">=6.29.5,<8.0.0"
 notebook = "^7.2.2"
 pytest = ">=8.3.2,<10.0.0"
 pytest-xdist = "^3.3.1"