Rebuild dbinfer from the original 4DBInfer archives

Regenerated by provenance/dbinfer.py from https://data.dgl.ai/mtbench/20240304-<name>.tar (sha256-pinned), replacing the port derived from dbinfer-relbench-adapter's db.zip artifacts.

Fixes: foreign keys were 99.9-100% null (validated against the parent's column count); real primary keys were overwritten before that validation; foreign keys to non-materialized key domains were dropped; task entity ids used a task-local remapping unrelated to the database's; task time columns were cast from integers into 1970 timestamps; classification targets were relabelled by sorted-string order and task_type inferred from cardinality; undeclared label and full-history-aggregate columns were published.

Results computed against the previous revision are not comparable.

README.md

@@ -3,6 +3,8 @@ tags:
 - relbench
 - relational-deep-learning
 pretty_name: RelBench dbinfer datasets
+size_categories:
+- 100M<n<1B
 configs:
 - config_name: databases
   data_files:
@@ -19,33 +21,51 @@ configs:
 This repository hosts the **dbinfer** family of relational datasets in the RelBench 3.0
 manifest format, one subdirectory per dataset. The datasets originate from the
 [4DBInfer benchmark](https://github.com/awslabs/multi-table-benchmark) (data version
-`20240304`) and are exposed to RelBench via the `dbinfer-relbench-adapter` package. Their
-labels are built externally and served as-is (every task has `kind: external`).
+`20240304`), built directly from the original archives that `dbinfer_bench` itself
+downloads (`https://data.dgl.ai/mtbench/20240304-<name>.tar`). Labels are the source's own, served as-is
+(every task has `kind: external`).
 
-Each subdirectory is a self-describing RelBench dataset (`manifest.yaml` + plain `db/*.parquet`
-+ `tasks/<task>/`); open its `schema.svg` for a zoomable entity-relationship diagram.
+Each subdirectory is a self-describing RelBench dataset (`manifest.yaml` + plain
+`db/*.parquet` + `tasks/<task>/`); open its `schema.svg` for a zoomable
+entity-relationship diagram.
 
 ## Datasets
 
-| dataset | domain | tasks |
-|---|---|---|
-| [`dbinfer-avs`](dbinfer-avs) | Acquire Valued Shoppers retail transactions | `repeater` |
-| [`dbinfer-diginetica`](dbinfer-diginetica) | E-commerce browsing/purchase sessions (CIKM Cup 2016) | `ctr`, `purchase` |
-| [`dbinfer-retailrocket`](dbinfer-retailrocket) | E-commerce visitor events | `cvr` |
-| [`dbinfer-seznam`](dbinfer-seznam) | Seznam.cz advertising account charges | `charge`, `prepay` |
-| [`dbinfer-amazon`](dbinfer-amazon) | Amazon product reviews | `rating`, `purchase`, `churn` |
-| [`dbinfer-stackexchange`](dbinfer-stackexchange) | StackExchange community Q&A | `churn`, `upvote` |
-| [`dbinfer-outbrain-small`](dbinfer-outbrain-small) | Outbrain content recommendation | `ctr` |
+| dataset | domain | tables | tasks |
+|---|---|---|---|
+| [`dbinfer-amazon`](dbinfer-amazon) | E-commerce (reviews) | 3 | `churn`, `purchase`, `rating` |
+| [`dbinfer-avs`](dbinfer-avs) | Retail (Acquire Valued Shoppers) | 8 | `repeater` |
+| [`dbinfer-diginetica`](dbinfer-diginetica) | E-commerce (sessions) | 12 | `ctr`, `purchase` |
+| [`dbinfer-outbrain-small`](dbinfer-outbrain-small) | Content recommendation | 9 | `ctr` |
+| [`dbinfer-retailrocket`](dbinfer-retailrocket) | E-commerce (behaviour) | 7 | `cvr` |
+| [`dbinfer-seznam`](dbinfer-seznam) | Digital advertising | 4 | `charge`, `prepay` |
+| [`dbinfer-stackexchange`](dbinfer-stackexchange) | Online community (Q&A) | 9 | `churn`, `upvote` |
 
-(Only datasets actually present as subdirectories are available; see each subdirectory's card
-for details.)
+Table counts include the single-column key tables materialized for the foreign-key targets
+that 4DBInfer declares without a payload of their own (`Item`, `Visitor`, `Customer`,
+`Chain`, `Brand`, `Category`, `Company`, `Session`, `User`, `Orders`, `Token`); see each
+dataset card.
+
+> **`dbinfer-outbrain-small` has almost no referential integrity, in the source.** 4DBInfer
+> subsampled its tables independently, so ~99.9% of its foreign keys -- and all but 58 of
+> 69,543 distinct train entities of its `ctr` task -- point at rows that were not kept. The
+> previous revision hid this behind all-null keys. There is no full-size `outbrain` archive
+> upstream. Its card has the numbers.
+
+Several tasks derive their label from a column that is also in the database
+(`retailrocket/cvr` <- `View.added_to_cart`, `seznam/charge` <- `Dobito.sluzba`,
+`seznam/prepay` <- `Probehnuto.sluzba`, `outbrain-small/ctr` <- `Click.clicked`,
+`amazon/rating` <- `Review.rating`). Each such task declares `remove_columns`; the column is
+kept in `db/` so the database stays faithful to 4DBInfer. RelBench's loader currently
+honours `remove_columns` only for `kind: autocomplete` tasks, so drop it yourself for these
+`external` ones. Test labels are outside `get_db(upto_test_timestamp=True)` regardless.
 
 ## Loading
 
 ```python
 import relbench
-ds = relbench.load_dataset("dbinfer-diginetica")     # or any dataset above
-task = relbench.load_task("dbinfer-diginetica", "ctr")
+ds = relbench.load_dataset("relbench/dbinfer/dbinfer-diginetica")
+task = relbench.load_task("relbench/dbinfer/dbinfer-diginetica", "ctr")
 db = ds.get_db()
 train = task.get_table("train")
 ```
@@ -54,15 +74,62 @@ See the RelBench
 [CONTRIBUTING guide](https://github.com/snap-stanford/relbench/blob/main/CONTRIBUTING.md)
 for the manifest layout.
 
+## Provenance and revision history
+
+Generated by
+[`provenance/dbinfer.py`](https://github.com/snap-stanford/relbench/blob/main/provenance/dbinfer.py),
+which pins the sha256 of each source archive; verified by
+[`provenance/check_dbinfer.py`](https://github.com/snap-stanford/relbench/blob/main/provenance/check_dbinfer.py).
+Port decisions -- which columns are kept, how implicit key domains are materialized, how
+keys are reindexed -- are documented in that file's module docstring, and each dataset card
+lists the payload columns it drops.
+
+**This collection was rebuilt from the original 4DBInfer archives.** The previous revision
+was derived from pre-built `db.zip` artifacts produced by the upstream
+`dbinfer-relbench-adapter` export pipeline, which corrupted the data in ways that were not
+visible from the schema:
+
+* **Every declared foreign key was 99.9-100% null.** The adapter validated foreign keys
+  against `len(parent_table)` where the parent was a `{column: array}` dict -- i.e.
+  against the parent's *column count* -- so all larger key values were set to `NaN`. None
+  of the seven databases could be joined.
+* **Real primary keys were overwritten** with `np.arange(n)` before that validation, so the
+  key correspondence was already gone.
+* **Foreign keys to non-materialized key domains were dropped**, amputating the schema
+  (`retailrocket/View` was left with no foreign keys at all).
+* **Task entity ids were remapped through a task-local mapping** unrelated to the
+  database's, so label rows did not reference the database.
+* **Task time columns were destroyed** (`1970-01-01 00:00:00.000000022`) by casting the
+  source's integer-valued columns with `astype('datetime64[ns]')`.
+* **Classification targets were silently relabelled** by sorted-string order, and
+  `task_type` was inferred from target cardinality -- turning `amazon/rating`, a 4DBInfer
+  regression/RMSE task, into multiclass, and both retrieval tasks into multiclass.
+* **Undeclared payload columns were published**, including `Posts.Score` (from which
+  `stackexchange/upvote`'s label is derived), `History.repeater` (`avs/repeater`'s label
+  itself), and full-history aggregates like `Users.Reputation/UpVotes/Views`.
+* **`val_timestamp` / `test_timestamp` fell after the end of the data**, so
+  `get_db(upto_test_timestamp=True)` trimmed nothing and gave no temporal protection at all
+  (`dbinfer-retailrocket` claimed `2015-09-21` against a last event of `2015-09-18`;
+  `dbinfer-diginetica` claimed `2016-11-12`; `dbinfer-seznam` `2015-10-04` against
+  `2015-10-01`). They also did not bracket the source's own splits -- seznam's val labels
+  start at `2015-04-01` and its test labels at `2015-07-01`.
+
+The current revision fixes all of the above: foreign keys resolve, primary keys are dense,
+implicit key domains are materialized as tables, task entity columns index their entity
+table, time columns are real timestamps, targets keep their source values, `task_type`
+follows the source, and each task declares `remove_columns` for any database column its
+label is derived from. **Results computed against the previous revision are not
+comparable.**
+
 ## Citation
 
 These datasets are from the 4DBInfer benchmark. If you use them, please cite:
 
 ```bibtex
-@article{dbinfer,
-  title={4DBInfer: A 4D Benchmarking Toolbox for Graph-Centric Predictive Modeling on Relational DBs},
-  author={Wang, Minjie and Gan, Quan and Wipf, David and Cai, Zhenkun and Li, Ning and Tang, Jianheng and Zhang, Yanlin and Zhang, Zizhao and Mao, Zunyao and Song, Yakun and Wang, Yanbo and Li, Jiahang and Zhang, Han and Yang, Guang and Qin, Xiao and Lei, Chuan and Zhang, Muhan and Zhang, Weinan and Faloutsos, Christos and Zhang, Zheng},
-  journal={arXiv preprint arXiv:2404.18209},
-  year={2024}
+@inproceedings{wang2024fourdbinfer,
+  title     = {{4DBInfer}: A {4D} Benchmarking Toolbox for Graph-Centric Predictive Modeling on Relational Databases},
+  author    = {Wang, Minjie and Gan, Quan and Wipf, David and Cai, Zhenkun and Li, Ning and Tang, Jianheng and Zhang, Yanlin and Zhang, Zizhao and Mao, Zunyao and Song, Yakun and Wang, Yanbo and Li, Jiahang and Zhang, Han and Yang, Guang and Qin, Xiao and Lei, Chuan and Zhang, Muhan and Zhang, Weinan and Faloutsos, Christos and Zhang, Zheng},
+  booktitle = {Advances in Neural Information Processing Systems 37 (NeurIPS 2024) Datasets and Benchmarks Track},
+  year      = {2024}
 }
 ```

STATS/databases.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7eebe80dd44a58a60e68375c9635235d19b85cbfde1612d6f5a4f898082e0aea
-size 10691
+oid sha256:14c962ef7d84bcb863052e68adbaae4c554442d162d5ddc65165d3c7113fca02
+size 14300

STATS/tasks.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e6074513c9f778c2d49770c78c3de64df1989ad326677421c343d3a8fed6b981
-size 11901
+oid sha256:1860e13c5cd347fdaf977fdd87ff66cae92be52e12c3d0d92e884817238eb797
+size 14961

dbinfer-amazon/README.md

@@ -10,9 +10,33 @@ Amazon from the 4DBInfer benchmark: a large product-review dataset linking users
 
 | task | kind | type | description |
 |---|---|---|---|
-| `churn` | external | binary_classification | Predict whether a user churns (stops purchasing). |
-| `purchase` | external | multiclass_classification | Predict whether a user purchases a product. |
-| `rating` | external | multiclass_classification | Predict the star rating a user gives a product. |
+| `rating` | external | regression | Predict the star rating a customer gives a product (`product_id` names the product; 4DBInfer scores this as regression/RMSE). |
+| `purchase` | external | link_prediction | Rank products a customer will purchase. 4DBInfer's MRR retrieval protocol: `train` holds positives only; `val`/`test` enumerate candidates with `label` and `query_idx`. Score with MRR over those candidates -- RelBench's default link metric (MAP) is a different protocol. |
+| `churn` | external | binary_classification | Predict whether a customer churns (stops reviewing/purchasing). |
+
+## Port notes
+
+Built from the original 4DBInfer archive (`https://data.dgl.ai/mtbench/20240304-amazon.tar`), keeping exactly the columns its `metadata.yaml` declares. Primary keys are reindexed to `0..n-1` and every foreign key -- in the database and in the task labels -- is remapped through the same mapping.
+
+Undeclared payload columns dropped:
+
+* `Customer`: `customer_name`
+
+### Label columns in the database
+
+4DBInfer derives some labels from a column that is itself part of the database, so a model reading that row can read its own label:
+
+* `rating` -> `Review.rating`
+* `purchase` -> `Review.product_id`
+
+Each such task declares `remove_columns`, and the column is left in `db/` so the database stays faithful to the source. Note that RelBench's loader currently applies `remove_columns` only to `kind: autocomplete` tasks (`Dataset.get_db` calls `get_modified_db` only when `target_col` is set), so for these `external` tasks **drop the column yourself** before training. Because `get_db(upto_test_timestamp=True)` trims the database at `test_timestamp`, test labels are outside the visible database either way -- the exposure is on the train and val rows.
+
+### Known upstream defects
+
+The three tasks are split at different points in time (`churn` from 2015-10-03,
+`purchase` from 2015-12-29, `rating` from 2015-12-30). The dataset-level
+`val_timestamp`/`test_timestamp` take the earliest of each, so trimming the database at a
+cutoff is conservative for every task rather than exact for one.
 
 ## Loading
 

dbinfer-amazon/db/Customer.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2b0c351cedf1591c31f8dd902494a3510c0d3a676640c580e0abcfb2590c93b4
-size 27108420
+oid sha256:bc2933612361bcaf9d6309fc5a52d270408f0a1590d3e4aa9e1dd23584dfd134
+size 7955070

dbinfer-amazon/db/Product.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c0207519f78c0651f2ce7fe16159ddfee1520b006864f7379f8043a155344bc8
-size 205561338
+oid sha256:cf1b894ef612e9b4ce4a0fbb7f32b394ba3487d401a9a7fb487ce0c100aa4276
+size 205561342

dbinfer-amazon/db/Review.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2157a00380d90948a1d6517f3a6306d3937f5fb4562ec2c103db98fddb75c717
-size 5669621770
+oid sha256:061db08142c1123130e2880503df1faba3f2184432c526d6e7996cc61882ebf2
+size 5497783668

dbinfer-amazon/manifest.yaml

@@ -1,20 +1,20 @@
 name: dbinfer-amazon
 manifest_version: 1
 description: 'Amazon from the 4DBInfer benchmark: a large product-review dataset linking users, products and reviews, used for rating prediction and user purchase/churn prediction.'
-val_timestamp: '2016-01-03 00:00:00'
-test_timestamp: '2016-01-04 00:00:00'
+val_timestamp: '2015-10-03 00:00:00'
+test_timestamp: '2015-12-30 00:00:00'
 tables:
   Customer:
     pkey: customer_id
     time_col: null
     fkeys: {}
+  Product:
+    pkey: product_id
+    time_col: null
+    fkeys: {}
   Review:
-    pkey: __synthetic_pk__
+    pkey: null
     time_col: review_time
     fkeys:
       customer_id: Customer
       product_id: Product
-  Product:
-    pkey: product_id
-    time_col: null
-    fkeys: {}

dbinfer-amazon/tasks/churn/manifest.yaml

@@ -1,10 +1,9 @@
 name: churn
 kind: external
 task_type: binary_classification
-description: Predict whether a user churns (stops purchasing).
+description: Predict whether a customer churns (stops reviewing/purchasing).
 entity_table: Customer
-entity_col: timestamp
+entity_col: customer_id
 target_col: churn
 time_col: timestamp
-timedelta: 1 days 00:00:00
 manifest_version: 1

dbinfer-amazon/tasks/churn/test.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fba46135cff418414ea31a30b47235a161792ff53e2a045a1827b2f6be4ce880
+oid sha256:1f7f4869d7c61553ad7742b631fa132242339b8aa82b5013a939b010b2d03d90
 size 914941

dbinfer-amazon/tasks/churn/train.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1c074b0892108123dfc7a2493e44793abee1096689e5b47b232a9e1b15367472
-size 2813393
+oid sha256:e286e9b25e89043ea5ec152a0a978b92343affe4823962159d053051382f2c3e
+size 2813478

dbinfer-amazon/tasks/churn/val.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3d83ccf909264eb5fef9d54cad96f3e1f981dcea18389f7ccbc2a5ecdd68016f
+oid sha256:c7313a16ee2650746763104dc7ecc54f8be06ad50c57cccb2a8bde1b46ebe00a
 size 901837

dbinfer-amazon/tasks/purchase/manifest.yaml

@@ -1,10 +1,13 @@
 name: purchase
 kind: external
-task_type: multiclass_classification
-description: Predict whether a user purchases a product.
-entity_table: Review
-entity_col: customer_id
-target_col: product_id
+task_type: link_prediction
+description: 'Rank products a customer will purchase. 4DBInfer''s MRR retrieval protocol: `train` holds positives only; `val`/`test` enumerate candidates with `label` and `query_idx`. Score with MRR over those candidates -- RelBench''s default link metric (MAP) is a different protocol.'
 time_col: prediction_timestamp
-timedelta: 1 days 00:00:00
+src_entity_table: Customer
+src_entity_col: customer_id
+dst_entity_table: Product
+dst_entity_col: product_id
+remove_columns:
+- - Review
+  - product_id
 manifest_version: 1

dbinfer-amazon/tasks/purchase/test.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:18397b96fe7d26a62e2d7479773bf906acbe701cc1bb40fa89b8713103abbec1
-size 1410670
+oid sha256:f7a9af1315ec92890ae10024c8c4412c71d77958cd1973a4ea0e7aef2367c005
+size 1421894

dbinfer-amazon/tasks/purchase/train.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d6aee988d2c6b9029a3efa82d9157f08ddb644c0b86383c47e340401390486b8
-size 663145
+oid sha256:2e829f0516c4316eddd08872880ad1ca6444d799c0fbcc1887ba3738f6956668
+size 708776

dbinfer-amazon/tasks/purchase/val.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bcf4fede317ba52095db4d29a0fe4be1b53e0e1018710c548718d7988705ca42
-size 765218
+oid sha256:c37af4304e645f79368b208dc2679ea2773215199960bc996a6083b4c860134f
+size 771793

dbinfer-amazon/tasks/rating/manifest.yaml

@@ -1,10 +1,12 @@
 name: rating
 kind: external
-task_type: multiclass_classification
-description: Predict the star rating a user gives a product.
-entity_table: Review
-entity_col: review_time
+task_type: regression
+description: Predict the star rating a customer gives a product (`product_id` names the product; 4DBInfer scores this as regression/RMSE).
+entity_table: Customer
+entity_col: customer_id
 target_col: rating
 time_col: review_time
-timedelta: 1 days 00:00:00
+remove_columns:
+- - Review
+  - rating
 manifest_version: 1

dbinfer-amazon/tasks/rating/test.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:16a4f375bc1957eaca4e77b7f992e9eb719d46223d4154511c483901e3029d2a
-size 144638
+oid sha256:ced8f84cf4a72c859be9c19c36334525f3284e5727397cb26a4e02699017fdd0
+size 144640

dbinfer-amazon/tasks/rating/train.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b9f831e1351c0a461fb967582d9e5cc841f1916661a22dc27a44f8d2e216909b
-size 738553
+oid sha256:0fe5df269383904de2659124fe76c1c5f9fda54ff18f1be6c08e4f42a747d9d4
+size 738597

dbinfer-amazon/tasks/rating/val.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:521550b51bfac612497ed6f09349ef03546dc84d9a33db779ddfa5a1af0c5b1b
-size 84948
+oid sha256:fa6347cf68fa17b405c6fe306c8623bb71a908c6d9344fc6b564e6d4f14bcd50
+size 84956

dbinfer-avs/README.md

@@ -10,7 +10,17 @@ Acquire Valued Shoppers (AVS) from the 4DBInfer benchmark: a retail dataset of c
 
 | task | kind | type | description |
 |---|---|---|---|
-| `repeater` | external | binary_classification | Predict whether a shopper becomes a repeat buyer of an offer. |
+| `repeater` | external | binary_classification | Predict whether a shopper becomes a repeat buyer of an offer (`offer`, `chain`, `market`, `offerdate` describe the offer instance). Source labels `'t'`/`'f'` are encoded 1/0. |
+
+## Port notes
+
+Built from the original 4DBInfer archive (`https://data.dgl.ai/mtbench/20240304-avs.tar`), keeping exactly the columns its `metadata.yaml` declares. Primary keys are reindexed to `0..n-1` and every foreign key -- in the database and in the task labels -- is remapped through the same mapping.
+
+4DBInfer links some foreign keys to key domains that have no payload table of their own; those are materialized here as single-column key tables: `Brand`, `Category`, `Chain`, `Company`, `Customer`.
+
+Undeclared payload columns dropped:
+
+* `History`: `repeattrips`, `repeater`
 
 ## Loading
 

dbinfer-avs/db/Brand.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5b3996482f14935b37ed78ac4534cd39d5a8ed8cc4bf71d13f10ea465b620acf
+size 213076

dbinfer-avs/db/Category.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e4a283dfb5d829b18e6f2c533629438e85c258963068cbab5dabe00c4bfd8bdf
+size 5581

dbinfer-avs/db/Chain.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e30458708d44b960b4b236847c71378038960ce03e99d2737184a2d75c8076b
+size 1860

dbinfer-avs/db/Company.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f05c4d6ec1068976656ae37a8e51f78f3ffacb205df42abebbdb72c0886951f
+size 195564

dbinfer-avs/db/Customer.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:827cd003d8b36e2b0fa617d76bcef1a7be4925562b81f591736800969d3f90a4
+size 1519961

dbinfer-avs/db/History.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d863a3bb790e9f08fcfbf606467947d94ab69099534612a42fef5006b333ac5f
-size 2469250
+oid sha256:a6af897bd74c66cdb0818ac32fb1437401f0d5b7c735862daf4fb2130bf39365
+size 1302748

dbinfer-avs/db/Offer.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2f32501995256bba5b498008009213a4ce5fc0e39d540e0b51c141e33002a4ec
-size 4625
+oid sha256:5ed6b3bb3cd44e19ce7af3e52c924c755f7b3eda3e3cbf28fef5c5cbe31cab87
+size 4598

dbinfer-avs/db/Transaction.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f44db13a2e424027900a1229bb72068bdd101f5ab275f16673177d45d86f83e3
-size 5559598329
+oid sha256:adbfa91855cf766d52667f3726ada3888d5b489c55e21383c182b8e5729d8b94
+size 3284971241

dbinfer-avs/manifest.yaml

@@ -1,19 +1,49 @@
 name: dbinfer-avs
 manifest_version: 1
 description: 'Acquire Valued Shoppers (AVS) from the 4DBInfer benchmark: a retail dataset of customer transaction histories and promotional offers, used to predict shopper behavior such as offer repeat purchases.'
-val_timestamp: '2013-07-30 00:00:00'
-test_timestamp: '2013-07-31 00:00:00'
+val_timestamp: '2013-04-24 00:00:00'
+test_timestamp: '2013-04-30 00:00:00'
 tables:
+  Offer:
+    pkey: offer
+    time_col: null
+    fkeys:
+      brand: Brand
+      category: Category
+      company: Company
+  Brand:
+    pkey: id
+    time_col: null
+    fkeys: {}
+  Category:
+    pkey: id
+    time_col: null
+    fkeys: {}
+  Chain:
+    pkey: id
+    time_col: null
+    fkeys: {}
+  Company:
+    pkey: id
+    time_col: null
+    fkeys: {}
+  Customer:
+    pkey: id
+    time_col: null
+    fkeys: {}
   History:
-    pkey: __synthetic_pk__
+    pkey: null
     time_col: offerdate
     fkeys:
+      chain: Chain
+      id: Customer
       offer: Offer
   Transaction:
-    pkey: __synthetic_pk__
+    pkey: null
     time_col: date
-    fkeys: {}
-  Offer:
-    pkey: offer
-    time_col: null
-    fkeys: {}
+    fkeys:
+      brand: Brand
+      category: Category
+      chain: Chain
+      company: Company
+      id: Customer

dbinfer-avs/tasks/repeater/manifest.yaml

@@ -1,10 +1,9 @@
 name: repeater
 kind: external
 task_type: binary_classification
-description: Predict whether a shopper becomes a repeat buyer of an offer.
-entity_table: History
+description: Predict whether a shopper becomes a repeat buyer of an offer (`offer`, `chain`, `market`, `offerdate` describe the offer instance). Source labels `'t'`/`'f'` are encoded 1/0.
+entity_table: Customer
 entity_col: id
 target_col: repeater
 time_col: timestamp
-timedelta: 1 days 00:00:00
 manifest_version: 1

dbinfer-avs/tasks/repeater/test.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8c38a530eb578b583a9d8d4004c5539feb4ff26122860553fe41066f599e943c
-size 225527
+oid sha256:117a0916ac7ec0edd8bb1660c9e68d0ebc5f79f80ba57b6e41e075e061409738
+size 227993

dbinfer-avs/tasks/repeater/train.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:995a502ea9f2a215798e8839d1e488297e90782a12a65d1f7353aca66004b2f0
-size 931403
+oid sha256:d5726602bece7b5a059fbda34d171c759c87967e7d26516228a8d9d9298df2f8
+size 941649

dbinfer-avs/tasks/repeater/val.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8225e3dd5f9cfb48287df40d15afb72f9a361b9e8a3c61b935e4691072bb14fa
-size 204314
+oid sha256:19216bf35cc6da40a6d681cc19121ad38f0ad6ecdcff90b81162256cedf7f66a
+size 206200

dbinfer-diginetica/README.md

@@ -10,8 +10,31 @@ Diginetica from the 4DBInfer benchmark: an e-commerce dataset of user browsing a
 
 | task | kind | type | description |
 |---|---|---|---|
-| `ctr` | external | binary_classification | Predict whether a displayed item is clicked (click-through rate). |
-| `purchase` | external | multiclass_classification | Predict whether a viewed item is purchased. |
+| `ctr` | external | binary_classification | Predict whether a displayed item is clicked (click-through rate); `itemId` names the displayed item. |
+| `purchase` | external | link_prediction | Rank items a session will purchase. 4DBInfer's MRR retrieval protocol: `train` holds positives only; `val`/`test` enumerate candidates with `label` and `query_idx`. Score with MRR over those candidates -- RelBench's default link metric (MAP) is a different protocol. |
+
+## Port notes
+
+Built from the original 4DBInfer archive (`https://data.dgl.ai/mtbench/20240304-diginetica.tar`), keeping exactly the columns its `metadata.yaml` declares. Primary keys are reindexed to `0..n-1` and every foreign key -- in the database and in the task labels -- is remapped through the same mapping.
+
+4DBInfer links some foreign keys to key domains that have no payload table of their own; those are materialized here as single-column key tables: `Orders`, `Session`, `Token`, `User`.
+
+### Label columns in the database
+
+4DBInfer derives some labels from a column that is itself part of the database, so a model reading that row can read its own label:
+
+* `purchase` -> `Purchase.itemId`
+
+Each such task declares `remove_columns`, and the column is left in `db/` so the database stays faithful to the source. Note that RelBench's loader currently applies `remove_columns` only to `kind: autocomplete` tasks (`Dataset.get_db` calls `get_modified_db` only when `target_col` is set), so for these `external` tasks **drop the column yourself** before training. Because `get_db(upto_test_timestamp=True)` trims the database at `test_timestamp`, test labels are outside the visible database either way -- the exposure is on the train and val rows.
+
+### Known upstream defects
+
+Two source-side quirks, both faithful to the archive:
+
+* `QuerySearchstringToken.queryId` -> `Query`: 43,759 of 138,260 values (31.65%) name a
+  query that is not in `Query`, so they are null here.
+* `Query.userId`, `View.userId`, `Purchase.userId` are 64%/70%/63% null in the source --
+  Diginetica sessions are mostly anonymous. That is the data, not a porting loss.
 
 ## Loading
 

dbinfer-diginetica/db/Click.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d71af23640c331e3d68d75aafe57db56e831938c130dff92b2a969f40fae1f9f
-size 13213627
+oid sha256:0191c4fece9a1a040b7530868f26ec0eec8405d6b8ea4bb8a1f8d7d82f9d2a2e
+size 16083479

dbinfer-diginetica/db/Orders.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba01053b724416779a46c2220ae11bf0f7f305a6a319f8ea8e1786221a9c1149
+size 78926

dbinfer-diginetica/db/Product.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:49cb9d5ed922ff9da2e0bab1b2dc26d4442efbb174b48c45438e07cb6c41002f
+oid sha256:dfe3b14863402fac527f4693eb6c38026600fdbb7d523b9e435c8768d744e3ad
 size 1105589

dbinfer-diginetica/db/ProductNameToken.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e7e0440643583c710da6009876e5c6872462ed7014b0af23aef8acb707651b16
-size 7040478
+oid sha256:c34352cc65f3afef6a2291675f3891c6b3cc6ad31917ca244eec8976313ba871
+size 4635441

dbinfer-diginetica/db/Purchase.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:65ddc59398f6c8af80c092a3fec3633ab4559a78ccd06a57f6a09f09c5d8f6a8
-size 464218
+oid sha256:ba861387673acfcc8d5121277403ddceb1471df7009122c3145e407278b50a65
+size 428150

dbinfer-diginetica/db/Query.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cf5f7e2b06e76f4e48f5998cd7b2ef2d32d4030a4617b4c5db035b66decb0412
-size 13367023
+oid sha256:989eeec9ed1e8554a37d5825e7c99ad4ff73c0568bffceaa1ecca9c93c9e2778
+size 13294414

dbinfer-diginetica/db/QueryResult.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f167bbabc93d9f0f4420f64a13f5b011bd7ac1e45c77fad5a227b2a1a9302c71
-size 401427337
+oid sha256:e0142e7f1f6e07ba4d973c0ea350d19682f23f9e94268afe15cb0d2251f4fec7
+size 228555939

dbinfer-diginetica/db/QuerySearchstringToken.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2e593b640506608d2ab4194f688774f7e8c0e22090507f7449c5ba39ae0a1368
-size 1168161
+oid sha256:d5b6677e00b35b4ce5104f4fd2d407b2de2974546e07c096beedc83fa5e53459
+size 664675

dbinfer-diginetica/db/Session.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f97e071a5650ee16486c44c818d69f1158c34d3f76a98ca18d2481d114f170a
+size 1757616

dbinfer-diginetica/db/Token.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b216550200e27ed51000a84701de5650d1e1fde67dc621dca38530e266f9ba4b
+size 994344

dbinfer-diginetica/db/User.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:014c18f8b0109ddd34294b1066a6bc23c747532f81288d3febbc50a0f87c1acf
+size 837770

dbinfer-diginetica/db/View.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:10c885cf4a116bcdf4667fdeadbb8129a87cd540899de3ce79751ecf1e606b18
-size 19543006
+oid sha256:e6548ece42a50a7bcd708a103e3c07a98254235b59d532f7505d1b9ea72221d5
+size 17283867

dbinfer-diginetica/manifest.yaml

@@ -1,46 +1,71 @@
 name: dbinfer-diginetica
 manifest_version: 1
 description: 'Diginetica from the 4DBInfer benchmark: an e-commerce dataset of user browsing and purchasing sessions over a product catalog (CIKM Cup 2016), used for click-through-rate and purchase prediction.'
-val_timestamp: '2016-11-11 00:00:00'
-test_timestamp: '2016-11-12 00:00:00'
+val_timestamp: '2016-05-26 00:00:00'
+test_timestamp: '2016-05-31 00:00:00'
 tables:
-  Click:
-    pkey: __synthetic_pk__
-    time_col: timestamp
-    fkeys:
-      queryId: Query
-      itemId: Product
-  Purchase:
-    pkey: __synthetic_pk__
+  Product:
+    pkey: itemId
+    time_col: null
+    fkeys: {}
+  Query:
+    pkey: queryId
     time_col: timestamp
     fkeys:
-      itemId: Product
-  QuerySearchstringToken:
-    pkey: __synthetic_pk__
+      sessionId: Session
+      userId: User
+  Orders:
+    pkey: id
+    time_col: null
+    fkeys: {}
+  Session:
+    pkey: id
+    time_col: null
+    fkeys: {}
+  Token:
+    pkey: id
+    time_col: null
+    fkeys: {}
+  User:
+    pkey: id
     time_col: null
+    fkeys: {}
+  Click:
+    pkey: null
+    time_col: timestamp
     fkeys:
       queryId: Query
+      itemId: Product
   QueryResult:
-    pkey: __synthetic_pk__
+    pkey: null
     time_col: timestamp
     fkeys:
       queryId: Query
       itemId: Product
   View:
-    pkey: __synthetic_pk__
+    pkey: null
     time_col: timestamp
     fkeys:
+      sessionId: Session
+      userId: User
       itemId: Product
+  Purchase:
+    pkey: null
+    time_col: timestamp
+    fkeys:
+      sessionId: Session
+      userId: User
+      itemId: Product
+      ordernumber: Orders
   ProductNameToken:
-    pkey: __synthetic_pk__
+    pkey: null
     time_col: null
     fkeys:
       itemId: Product
-  Product:
-    pkey: itemId
+      token: Token
+  QuerySearchstringToken:
+    pkey: null
     time_col: null
-    fkeys: {}
-  Query:
-    pkey: queryId
-    time_col: timestamp
-    fkeys: {}
+    fkeys:
+      queryId: Query
+      token: Token