Upload README.md with huggingface_hub

README.md

@@ -2,7 +2,7 @@
 license: cc-by-nc-sa-4.0
 language:
 - en
-pretty_name: SurvCancel
+pretty_name: SurvBench Order Snapshots
 tags:
 - survival-analysis
 - transportation
@@ -10,24 +10,50 @@ tags:
 - ride-pooling
 - jsonl
 configs:
-- config_name: default
+- config_name: all
   data_files:
   - split: train
-    path: train-*.jsonl
+    path: data/R*/*/*.jsonl
+- config_name: R1
+  data_files:
+  - split: train
+    path: data/R1/*/*.jsonl
+- config_name: R2
+  data_files:
+  - split: train
+    path: data/R2/*/*.jsonl
+- config_name: R3
+  data_files:
+  - split: train
+    path: data/R3/*/*.jsonl
+- config_name: R4
+  data_files:
+  - split: train
+    path: data/R4/*/*.jsonl
 ---
 
-# SurvCancel
+# SurvBench Order Snapshots
+
+SurvBench Order Snapshots is an anonymized order-level temporal dataset for
+survival-analysis benchmarking in demand-responsive transportation. Each JSONL
+record represents one focal order and contains static trip attributes, terminal
+outcome information, and an ordered sequence of system snapshots observed before
+the order outcome.
 
-SurvCancel is an anonymized order-level temporal dataset for survival-analysis
-benchmarking in demand-responsive transportation. Each JSONL record represents
-one focal order and contains static trip attributes, terminal outcome
-information, and an ordered sequence of system snapshots observed before the
-order outcome.
+The dataset is intended for research on order matching, cancellation dynamics,
+pickup dynamics, and time-to-event modeling in ride-pooling or
+demand-responsive transit systems.
 
-The Hugging Face release is intentionally flat: each public region is stored in
-one JSONL file, and all region files are exposed as a single `train` split.
+## Dataset Details
 
-## Files
+- **Format:** newline-delimited JSON (`application/x-ndjson`)
+- **Scope:** four anonymized service regions, named `R1` to `R4`
+- **Organization:** records are partitioned by region and service date
+- **Primary record:** one order per JSONL line
+- **Temporal component:** nested `snapshots` sequence per order
+- **License:** CC BY-NC-SA 4.0
+
+## Repository Structure
 
 ```text
 README.md
@@ -37,46 +63,68 @@ checksums.sha256
 sample/
   example_record.json
   sample_records.jsonl
-train-R1.jsonl
-train-R2.jsonl
-train-R3.jsonl
-train-R4.jsonl
+data/
+  R1/YYYY-MM-DD/order_snapshots_R1_YYYY-MM-DD.jsonl
+  R2/YYYY-MM-DD/order_snapshots_R2_YYYY-MM-DD.jsonl
+  R3/YYYY-MM-DD/order_snapshots_R3_YYYY-MM-DD.jsonl
+  R4/YYYY-MM-DD/order_snapshots_R4_YYYY-MM-DD.jsonl
 ```
 
+Large source files may be split by line into
+`order_snapshots_<region>_<date>_part-NNN.jsonl`. File parts preserve complete
+JSONL records; no JSON object is split across files.
+
+## Data Fields
+
+Each top-level record includes:
+
+- anonymized order and region identifiers
+- passenger count
+- pickup and dropoff coordinates in standardized planar coordinates
+- terminal outcome type and timestamp
+- an ordered `snapshots` array
+
+Each snapshot includes the focal order state, active surrounding orders,
+available vehicles, and anonymized matched vehicle information when applicable.
+See `SCHEMA.md` for field-level documentation.
+
 ## Loading
 
+The files can be read directly as JSONL. For example:
+
 ```python
-from datasets import load_dataset
+import json
+from pathlib import Path
 
-dataset = load_dataset("Couteau/SurvCancel", split="train", streaming=True)
+path = Path("data/R1/2025-12-22/order_snapshots_R1_2025-12-22.jsonl")
+with path.open("r", encoding="utf-8") as f:
+    first_record = json.loads(next(f))
 ```
 
-To load one region directly:
+With Hugging Face Datasets:
 
 ```python
 from datasets import load_dataset
 
-r1 = load_dataset(
+dataset = load_dataset(
     "json",
-    data_files="hf://datasets/Couteau/SurvCancel/train-R1.jsonl",
+    data_files="data/R1/*/*.jsonl",
     split="train",
-    streaming=True,
 )
 ```
 
-## Data Fields
-
-Each top-level record includes:
+Use streaming mode when reading many files or the full dataset:
 
-- anonymized order and region identifiers
-- passenger count
-- pickup and dropoff coordinates in standardized planar coordinates
-- terminal outcome type and timestamp
-- an ordered `snapshots` array
+```python
+from datasets import load_dataset
 
-Each snapshot includes the focal order state, active surrounding orders,
-available vehicles, and anonymized matched vehicle information when applicable.
-See `SCHEMA.md` for field-level documentation.
+dataset = load_dataset(
+    "json",
+    data_files="data/R*/*/*.jsonl",
+    split="train",
+    streaming=True,
+)
+```
 
 ## Anonymization
 
@@ -96,9 +144,13 @@ benchmark features.
 
 Coordinates are standardized planar coordinates, not raw latitude/longitude.
 
-## Integrity
+## Metadata and Integrity
+
+- `SCHEMA.md` describes the record fields.
+- `croissant.jsonld` provides MLCommons Croissant metadata.
+- `checksums.sha256` contains SHA256 checksums for release files.
 
-`checksums.sha256` contains SHA256 checksums for release files.
+To verify files after download:
 
 ```bash
 sha256sum -c checksums.sha256