openaivec 1.0.10__py3-none-any.whl → 1.0.12__py3-none-any.whl

openaivec/_cache/optimize.py

@@ -21,7 +21,10 @@ class BatchSizeSuggester:
     min_batch_size: int = 10
     min_duration: float = 30.0
     max_duration: float = 60.0
-    step_ratio: float = 0.2
+    step_ratio_up: float = 0.1
+    step_ratio_down: float = 0.2
+    max_step: int | None = None
+    min_step: int = 1
     sample_size: int = 4
     _history: list[PerformanceMetric] = field(default_factory=list)
     _lock: threading.RLock = field(default_factory=threading.RLock, repr=False)
@@ -34,8 +37,14 @@ class BatchSizeSuggester:
             raise ValueError("current_batch_size must be >= min_batch_size")
         if self.sample_size <= 0:
             raise ValueError("sample_size must be > 0")
-        if self.step_ratio <= 0:
-            raise ValueError("step_ratio must be > 0")
+        if self.step_ratio_up <= 0:
+            raise ValueError("step_ratio_up must be > 0")
+        if self.step_ratio_down <= 0:
+            raise ValueError("step_ratio_down must be > 0")
+        if self.max_step is not None and self.max_step <= 0:
+            raise ValueError("max_step must be > 0")
+        if self.min_step <= 0:
+            raise ValueError("min_step must be > 0")
         if self.min_duration <= 0 or self.max_duration <= 0:
             raise ValueError("min_duration and max_duration must be > 0")
         if self.min_duration >= self.max_duration:
@@ -94,9 +103,15 @@ class BatchSizeSuggester:
             current_size = self.current_batch_size
 
             if average_duration < self.min_duration:
-                new_batch_size = int(current_size * (1 + self.step_ratio))
+                delta = max(self.min_step, int(current_size * self.step_ratio_up))
+                if self.max_step is not None:
+                    delta = min(delta, self.max_step)
+                new_batch_size = current_size + delta
             elif average_duration > self.max_duration:
-                new_batch_size = int(current_size * (1 - self.step_ratio))
+                delta = max(self.min_step, int(current_size * self.step_ratio_down))
+                if self.max_step is not None:
+                    delta = min(delta, self.max_step)
+                new_batch_size = current_size - delta
             else:
                 new_batch_size = current_size
 

{openaivec-1.0.10.dist-info → openaivec-1.0.12.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: openaivec
-Version: 1.0.10
+Version: 1.0.12
 Summary: Generative mutation for tabular calculation
 Project-URL: Homepage, https://microsoft.github.io/openaivec/
 Project-URL: Repository, https://github.com/microsoft/openaivec
@@ -63,7 +63,7 @@ print(sentiment.tolist())
 # Output: ['Positive sentiment', 'Negative sentiment']
 ```
 
-**Try it live:** https://microsoft.github.io/openaivec/examples/pandas/
+**Pandas tutorial (GitHub Pages):** https://microsoft.github.io/openaivec/examples/pandas/
 
 ## Benchmarks
 
@@ -82,6 +82,7 @@ Batching alone removes most HTTP overhead, and letting batching overlap with con
 ## Contents
 
 - [Why openaivec?](#why-openaivec)
+- [Overview](#overview)
 - [Core Workflows](#core-workflows)
 - [Using with Apache Spark UDFs](#using-with-apache-spark-udfs)
 - [Building Prompts](#building-prompts)
@@ -93,14 +94,13 @@ Batching alone removes most HTTP overhead, and letting batching overlap with con
 ## Why openaivec?
 
 - Drop-in `.ai` and `.aio` accessors keep pandas analysts in familiar tooling.
-- OpenAI batch-optimized: `BatchingMapProxy`/`AsyncBatchingMapProxy` coalesce requests, dedupe prompts, and keep column order stable.
-- Smart batching (`BatchingMapProxy`/`AsyncBatchingMapProxy`) dedupes prompts, preserves order, and releases waiters on failure.
+- OpenAI batch-optimized: `BatchingMapProxy`/`AsyncBatchingMapProxy` coalesce requests, dedupe prompts, preserve order, and release waiters on failure.
 - Reasoning support mirrors the OpenAI SDK; structured outputs accept Pydantic `response_format`.
 - Built-in caches and retries remove boilerplate; helpers reuse caches across pandas, Spark, and async flows.
 - Spark UDFs and Microsoft Fabric guides move notebooks into production-scale ETL.
 - Prompt tooling (`FewShotPromptBuilder`, `improve`) and the task library ship curated prompts with validated outputs.
 
-# Overview
+## Overview
 
 Vectorized OpenAI batch processing so you handle many inputs per call instead of one-by-one. Batching proxies dedupe inputs, enforce ordered outputs, and unblock waiters even on upstream errors. Cache helpers (`responses_with_cache`, Spark UDF builders) plug into the same layer so expensive prompts are reused across pandas, Spark, and async flows. Reasoning models honor SDK semantics. Requires Python 3.10+.
 
@@ -186,7 +186,7 @@ result = df.assign(
 
 ### Using with reasoning models
 
-Reasoning models (o1-preview, o1-mini, o3-mini, etc.) work without special flags. `reasoning` mirrors the OpenAI SDK.
+Reasoning models (o1-preview, o1-mini, o3-mini, etc.) follow OpenAI SDK semantics. Pass `reasoning` when you want to override model defaults.
 
 ```python
 pandas_ext.set_responses_model("o1-mini")  # Set your reasoning model
@@ -194,7 +194,7 @@ pandas_ext.set_responses_model("o1-mini")  # Set your reasoning model
 result = df.assign(
     analysis=lambda df: df.text.ai.responses(
         "Analyze this text step by step",
-        reasoning={"effort": "none"}  # Optional: mirrors the OpenAI SDK argument
+        reasoning={"effort": "none"},  # Optional: mirrors the OpenAI SDK argument
     )
 )
 ```
@@ -254,7 +254,7 @@ df = pd.DataFrame({"text": [
 async def process_data():
     return await df["text"].aio.responses(
         "Analyze sentiment and classify as positive/negative/neutral",
-        reasoning={"effort": "none"},  # Required for gpt-5.1
+        reasoning={"effort": "none"},  # Recommended for reasoning models
         max_concurrency=12    # Allow up to 12 concurrent requests
     )
 

{openaivec-1.0.10.dist-info → openaivec-1.0.12.dist-info}/RECORD

@@ -11,7 +11,7 @@ openaivec/_util.py,sha256=XfueAycVCQvgRLS7wF7e306b53lebORvZOBzbQjy4vE,6438
 openaivec/pandas_ext.py,sha256=_y48qlG-npZsCCJJL1yev-yEU1YBZT83EiVl-lH0__o,87305
 openaivec/spark.py,sha256=XosDAcbzhnaIGyHBJ-p_ZBVJALroOXOFTjWWNRpSG3o,35022
 openaivec/_cache/__init__.py,sha256=IYUH5GKsJXuCX-k3XtT259rEz49EZm9KW2TIOTGW4uQ,314
-openaivec/_cache/optimize.py,sha256=3nS8VehbS7iGC1tPDDQh-iAgyKHbVYmMbCRBWM77U_U,3827
+openaivec/_cache/optimize.py,sha256=9JKU7HYABx-sYzp9SqoHVljG0aRqzmCY5_KCJc_Uq3M,4545
 openaivec/_cache/proxy.py,sha256=aVjH_hmJIIso6SetV_-Ct3VaOSG-n9Dpil7TttnbYkE,30556
 openaivec/_schema/__init__.py,sha256=XUj3Jv6ZVDjyYzSmH6Q5lmDj-hBMfUg_eBNeZACXR6Q,368
 openaivec/_schema/infer.py,sha256=VyvORgmpkcPa8pITClOJYjNzF4VzgSWe_n-9kFJVUjE,15644
@@ -33,7 +33,7 @@ openaivec/task/nlp/sentiment_analysis.py,sha256=P1AFazqmlE9Dy0OShNOXcY8X5rvsGg7X
 openaivec/task/nlp/translation.py,sha256=IgTy0PQZVF_Q6qis60STim7Vd7rYPVTfTfwP_U1kAKk,6603
 openaivec/task/table/__init__.py,sha256=kJz15WDJXjyC7UIHKBvlTRhCf347PCDMH5T5fONV2sU,83
 openaivec/task/table/fillna.py,sha256=nMlXvlUvyWgM9DxJDeRX3M37jxlqg0MgRet1Ds3ni5Y,6571
-openaivec-1.0.10.dist-info/METADATA,sha256=xJPQaqzzlN7rkL519vee4l8ZQRQfVwl9UxkXyzXGToM,14195
-openaivec-1.0.10.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
-openaivec-1.0.10.dist-info/licenses/LICENSE,sha256=ws_MuBL-SCEBqPBFl9_FqZkaaydIJmxHrJG2parhU4M,1141
-openaivec-1.0.10.dist-info/RECORD,,
+openaivec-1.0.12.dist-info/METADATA,sha256=D5-CBm8n5ReO2vYO5Fpbqby9C2S8xT6IFVOmV-2PjQY,14165
+openaivec-1.0.12.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+openaivec-1.0.12.dist-info/licenses/LICENSE,sha256=ws_MuBL-SCEBqPBFl9_FqZkaaydIJmxHrJG2parhU4M,1141
+openaivec-1.0.12.dist-info/RECORD,,