Datasets:

Modalities:
Tabular
Text
Formats:
parquet
ArXiv:
License:
kvignesh1420 commited on
Commit
4e35fb4
·
verified ·
1 Parent(s): 38ee8aa

Drop join-amazon-home and join-se-askubuntu (preprocessed node blobs infeasible)

Browse files

Their nodes.rkyv artifacts (~208 GB and ~132 GB) never made it into the-join-preprocessed and are not worth regenerating; both were already excluded from the curated rt-j.json training list. Removing the raw copies keeps the two repos consistent. STATS: 643->641 DBs, 16,791->16,677 tasks.

This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. STATS/databases.parquet +2 -2
  2. STATS/tasks.parquet +2 -2
  3. join-amazon-home/README.md +0 -26
  4. join-amazon-home/db/products.parquet +0 -3
  5. join-amazon-home/db/reviewers.parquet +0 -3
  6. join-amazon-home/db/reviews.parquet +0 -3
  7. join-amazon-home/manifest.yaml +0 -19
  8. join-amazon-home/schema.svg +0 -101
  9. join-amazon-home/tasks/product-all-verified/manifest.yaml +0 -30
  10. join-amazon-home/tasks/product-all-verified/train.parquet +0 -3
  11. join-amazon-home/tasks/product-any-new-review/manifest.yaml +0 -28
  12. join-amazon-home/tasks/product-any-new-review/train.parquet +0 -3
  13. join-amazon-home/tasks/product-avg-review-length/manifest.yaml +0 -27
  14. join-amazon-home/tasks/product-avg-review-length/train.parquet +0 -3
  15. join-amazon-home/tasks/product-churn/manifest.yaml +0 -31
  16. join-amazon-home/tasks/product-churn/train.parquet +0 -3
  17. join-amazon-home/tasks/product-long-review/manifest.yaml +0 -30
  18. join-amazon-home/tasks/product-long-review/train.parquet +0 -3
  19. join-amazon-home/tasks/product-low-rating-event/manifest.yaml +0 -30
  20. join-amazon-home/tasks/product-low-rating-event/train.parquet +0 -3
  21. join-amazon-home/tasks/product-low-rating/manifest.yaml +0 -30
  22. join-amazon-home/tasks/product-low-rating/train.parquet +0 -3
  23. join-amazon-home/tasks/product-max-review-len/manifest.yaml +0 -27
  24. join-amazon-home/tasks/product-max-review-len/train.parquet +0 -3
  25. join-amazon-home/tasks/product-max-review-length/manifest.yaml +0 -27
  26. join-amazon-home/tasks/product-max-review-length/train.parquet +0 -3
  27. join-amazon-home/tasks/product-mean-rating/manifest.yaml +0 -27
  28. join-amazon-home/tasks/product-mean-rating/train.parquet +0 -3
  29. join-amazon-home/tasks/product-min-rating/manifest.yaml +0 -27
  30. join-amazon-home/tasks/product-min-rating/train.parquet +0 -3
  31. join-amazon-home/tasks/product-popular-next-window/manifest.yaml +0 -30
  32. join-amazon-home/tasks/product-popular-next-window/train.parquet +0 -3
  33. join-amazon-home/tasks/product-rating-max-heavytail/manifest.yaml +0 -29
  34. join-amazon-home/tasks/product-rating-max-heavytail/train.parquet +0 -3
  35. join-amazon-home/tasks/product-rating-mean-heavytail/manifest.yaml +0 -29
  36. join-amazon-home/tasks/product-rating-mean-heavytail/train.parquet +0 -3
  37. join-amazon-home/tasks/product-rating-min-heavytail/manifest.yaml +0 -29
  38. join-amazon-home/tasks/product-rating-min-heavytail/train.parquet +0 -3
  39. join-amazon-home/tasks/product-rating-sum-heavytail/manifest.yaml +0 -29
  40. join-amazon-home/tasks/product-rating-sum-heavytail/train.parquet +0 -3
  41. join-amazon-home/tasks/product-rating-trend/manifest.yaml +0 -31
  42. join-amazon-home/tasks/product-rating-trend/train.parquet +0 -3
  43. join-amazon-home/tasks/product-rating-volatility/manifest.yaml +0 -27
  44. join-amazon-home/tasks/product-rating-volatility/train.parquet +0 -3
  45. join-amazon-home/tasks/product-return/manifest.yaml +0 -38
  46. join-amazon-home/tasks/product-return/train.parquet +0 -3
  47. join-amazon-home/tasks/product-review-count/manifest.yaml +0 -32
  48. join-amazon-home/tasks/product-review-count/train.parquet +0 -3
  49. join-amazon-home/tasks/product-review-length-change/manifest.yaml +0 -31
  50. join-amazon-home/tasks/product-review-length-change/train.parquet +0 -3
STATS/databases.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b88aa1a4a952e4e1bc0a7baf3a390cdfa5f35651274bb3a5738e59d88cbfd9b0
3
- size 66605
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2e96475e8fb71823f239d994075e3c7a621acc6bf78beb37a4187b215701c16e
3
+ size 66463
STATS/tasks.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a16f9f2f577f77be68c75aaf8949074dedfb55527a00309218767591bc8e87a6
3
- size 670769
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4cefeb1b9d4441260970369614e82921da62ec9782f995f81c7158e647eab4d3
3
+ size 666729
join-amazon-home/README.md DELETED
@@ -1,26 +0,0 @@
1
- # join-amazon-home
2
-
3
- ## Schema
4
-
5
- ![schema diagram](schema.svg)
6
-
7
- ## Loading
8
-
9
- ```python
10
- import relbench
11
- ds = relbench.load_dataset("relbench/join/join-amazon-home")
12
- task = relbench.load_task("relbench/join/join-amazon-home", "<task>")
13
- ```
14
-
15
- ## Citation
16
-
17
- Please cite [RelBench](https://proceedings.neurips.cc/paper_files/paper/2024/hash/25cd345233c65fac1fec0ce61d0f7836-Abstract-Datasets_and_Benchmarks_Track.html):
18
-
19
- ```bibtex
20
- @inproceedings{robinson2024relbench,
21
- title = {{RelBench}: A Benchmark for Deep Learning on Relational Databases},
22
- author = {Robinson, Joshua and Ranjan, Rishabh and Hu, Weihua and Huang, Kexin and Han, Jiaqi and Dobles, Alejandro and Fey, Matthias and Lenssen, Jan E. and Yuan, Yiwen and Zhang, Zecheng and He, Xinwei and Leskovec, Jure},
23
- booktitle = {Advances in Neural Information Processing Systems 37 (NeurIPS 2024) Datasets and Benchmarks Track},
24
- year = {2024}
25
- }
26
- ```
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/db/products.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:5e652767c8e823cc717b616459598ad429fe15f914ae6646f04b822338bc6ceb
3
- size 6831034
 
 
 
 
join-amazon-home/db/reviewers.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:3858a41d090501e45e8ba005a27617b6c323a2b061008a374e7aa50c4df38917
3
- size 8091480
 
 
 
 
join-amazon-home/db/reviews.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:24e9c6048250f8a6d6eaa67a12dafb83fad94bec4b92352d27cbfc7439480cf6
3
- size 48200398
 
 
 
 
join-amazon-home/manifest.yaml DELETED
@@ -1,19 +0,0 @@
1
- name: join-amazon-home
2
- manifest_version: 1
3
- val_timestamp: null
4
- test_timestamp: null
5
- tables:
6
- products:
7
- pkey: product_id
8
- time_col: null
9
- fkeys: {}
10
- reviewers:
11
- pkey: reviewer_id
12
- time_col: null
13
- fkeys: {}
14
- reviews:
15
- pkey: null
16
- time_col: timestamp
17
- fkeys:
18
- reviewer_id: reviewers
19
- product_id: products
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/schema.svg DELETED
join-amazon-home/tasks/product-all-verified/manifest.yaml DELETED
@@ -1,30 +0,0 @@
1
- name: product-all-verified
2
- kind: forecast
3
- task_type: binary_classification
4
- description: Whether all reviews for a product are verified (mean verified == 1) next window.
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: all_verified
8
- time_col: timestamp
9
- timedelta: 54 days 14:24:00
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-all-verified/train.parquet
13
- WITH eligible AS (
14
- SELECT DISTINCT t.timestamp AS ts, a.product_id AS e
15
- FROM timestamps t JOIN reviews a
16
- ON a."timestamp" > t.timestamp - 3 * INTERVAL '{timedelta}' AND a."timestamp" <= t.timestamp
17
- WHERE a.product_id IS NOT NULL
18
- ),
19
- agg AS (
20
- SELECT t.timestamp AS ts, CAST(a.product_id AS BIGINT) AS e,
21
- avg(TRY_CAST(a.verified AS DECIMAL(38, 18))) AS v
22
- FROM timestamps t
23
- JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
24
- JOIN eligible lb ON lb.ts = t.timestamp AND lb.e = a.product_id
25
- WHERE a.verified IS NOT NULL AND a.product_id IS NOT NULL
26
- GROUP BY 1, 2
27
- )
28
- SELECT ts AS "timestamp", e AS product_id, CAST(CASE WHEN v >= 1.0 THEN 1 ELSE 0 END AS BIGINT) AS all_verified
29
- FROM agg WHERE v IS NOT NULL
30
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-all-verified/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:c4914211c35d72be6bffb55fdbbb1da1125491bc1d592def63505219426cad9c
3
- size 2389836
 
 
 
 
join-amazon-home/tasks/product-any-new-review/manifest.yaml DELETED
@@ -1,28 +0,0 @@
1
- name: product-any-new-review
2
- kind: forecast
3
- task_type: binary_classification
4
- description: Whether the product gets any new review next window.
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: book_has_review
8
- time_col: timestamp
9
- timedelta: 54 days 14:24:00
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-any-new-review/train.parquet
13
- WITH eligible AS (
14
- SELECT DISTINCT t.timestamp AS ts, a.product_id AS e
15
- FROM timestamps t JOIN reviews a
16
- ON a."timestamp" > t.timestamp - 3 * INTERVAL '{timedelta}' AND a."timestamp" <= t.timestamp
17
- WHERE a.product_id IS NOT NULL
18
- ),
19
- ev AS (
20
- SELECT DISTINCT t.timestamp AS ts, CAST(v.product_id AS BIGINT) AS e
21
- FROM timestamps t
22
- JOIN reviews v ON v."timestamp" > t.timestamp AND v."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
23
- WHERE v.product_id IS NOT NULL
24
- )
25
- SELECT el.ts AS "timestamp", el.e AS product_id, CAST(CASE WHEN ev.e IS NULL THEN 0 ELSE 1 END AS BIGINT) AS book_has_review
26
- FROM eligible el
27
- LEFT JOIN ev ON ev.ts = el.ts AND ev.e = el.e
28
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-any-new-review/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:5542cb7082fc9e0fc522635c27a184139090c346dd9ce2c1072bc713026ae3d1
3
- size 5557161
 
 
 
 
join-amazon-home/tasks/product-avg-review-length/manifest.yaml DELETED
@@ -1,27 +0,0 @@
1
- name: product-avg-review-length
2
- kind: forecast
3
- task_type: regression
4
- description: Average review length for a home product next window.
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: avg_review_len
8
- time_col: timestamp
9
- timedelta: 54 days 19:26:24
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-avg-review-length/train.parquet
13
- WITH eligible AS (
14
- SELECT DISTINCT t.timestamp AS ts, a.product_id AS e
15
- FROM timestamps t JOIN reviews a
16
- ON a."timestamp" > t.timestamp - 3 * INTERVAL '{timedelta}' AND a."timestamp" <= t.timestamp
17
- WHERE a.product_id IS NOT NULL
18
- )
19
- SELECT t.timestamp AS "timestamp", CAST(a.product_id AS BIGINT) AS product_id,
20
- CAST(avg(CAST(a.review_len AS DOUBLE)) AS DOUBLE) AS avg_review_len
21
- FROM timestamps t
22
- JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
23
- JOIN eligible lb ON lb.ts = t.timestamp AND lb.e = a.product_id
24
- WHERE a.review_len IS NOT NULL AND a.product_id IS NOT NULL
25
- GROUP BY 1, 2
26
- HAVING avg_review_len IS NOT NULL
27
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-avg-review-length/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:46972bbc1a79d81d66b8cf4a60b6bf048a489a3e6b5cfce9e0852185f2455c51
3
- size 4011517
 
 
 
 
join-amazon-home/tasks/product-churn/manifest.yaml DELETED
@@ -1,31 +0,0 @@
1
- name: product-churn
2
- kind: forecast
3
- task_type: binary_classification
4
- description: Predict whether a home product receives no reviews in the next time window (loss of interest)
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: predict_product_churn
8
- time_col: timestamp
9
- timedelta: 54 days 14:24:00
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-churn/train.parquet
13
- WITH eligible AS (
14
- SELECT DISTINCT t.timestamp AS ts, a.product_id AS e
15
- FROM timestamps t JOIN reviews a
16
- ON a."timestamp" > t.timestamp - 3 * INTERVAL '{timedelta}' AND a."timestamp" <= t.timestamp
17
- WHERE a.product_id IS NOT NULL
18
- ),
19
- fut AS (
20
- SELECT DISTINCT t.timestamp AS ts, CAST(a.product_id AS BIGINT) AS e
21
- FROM timestamps t
22
- JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
23
- JOIN eligible lb ON lb.ts = t.timestamp AND lb.e = a.product_id
24
- WHERE a.product_id IS NOT NULL
25
- ),
26
- live_ts AS (SELECT DISTINCT ts FROM fut)
27
- SELECT el.ts AS "timestamp", el.e AS product_id, CAST(CASE WHEN f.e IS NULL THEN 1 ELSE 0 END AS BIGINT) AS predict_product_churn
28
- FROM eligible el
29
- JOIN live_ts lt ON lt.ts = el.ts
30
- LEFT JOIN fut f ON f.ts = el.ts AND f.e = el.e
31
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-churn/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:b210c9eb8c9e3c999a4dcdca66daaa598698e4dd0879935a63547817ea08c853
3
- size 5557211
 
 
 
 
join-amazon-home/tasks/product-long-review/manifest.yaml DELETED
@@ -1,30 +0,0 @@
1
- name: product-long-review
2
- kind: forecast
3
- task_type: binary_classification
4
- description: Whether any review exceeds 1000 characters next window.
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: long_review
8
- time_col: timestamp
9
- timedelta: 54 days 14:24:00
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-long-review/train.parquet
13
- WITH eligible AS (
14
- SELECT DISTINCT t.timestamp AS ts, a.product_id AS e
15
- FROM timestamps t JOIN reviews a
16
- ON a."timestamp" > t.timestamp - 3 * INTERVAL '{timedelta}' AND a."timestamp" <= t.timestamp
17
- WHERE a.product_id IS NOT NULL
18
- ),
19
- agg AS (
20
- SELECT t.timestamp AS ts, CAST(a.product_id AS BIGINT) AS e,
21
- max(TRY_CAST(a.review_len AS DECIMAL(38, 18))) AS v
22
- FROM timestamps t
23
- JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
24
- JOIN eligible lb ON lb.ts = t.timestamp AND lb.e = a.product_id
25
- WHERE a.review_len IS NOT NULL AND a.product_id IS NOT NULL
26
- GROUP BY 1, 2
27
- )
28
- SELECT ts AS "timestamp", e AS product_id, CAST(CASE WHEN v > 1000.0 THEN 1 ELSE 0 END AS BIGINT) AS long_review
29
- FROM agg WHERE v IS NOT NULL
30
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-long-review/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:6af2cc2b78301be540b6fa441aaccc6b33b00480b37f4c2e7421ebe14ae08228
3
- size 2366633
 
 
 
 
join-amazon-home/tasks/product-low-rating-event/manifest.yaml DELETED
@@ -1,30 +0,0 @@
1
- name: product-low-rating-event
2
- kind: forecast
3
- task_type: binary_classification
4
- description: Whether the product receives any 1-star rating next window.
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: low_rating_event
8
- time_col: timestamp
9
- timedelta: 54 days 14:24:00
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-low-rating-event/train.parquet
13
- WITH eligible AS (
14
- SELECT DISTINCT t.timestamp AS ts, a.product_id AS e
15
- FROM timestamps t JOIN reviews a
16
- ON a."timestamp" > t.timestamp - 3 * INTERVAL '{timedelta}' AND a."timestamp" <= t.timestamp
17
- WHERE a.product_id IS NOT NULL
18
- ),
19
- agg AS (
20
- SELECT t.timestamp AS ts, CAST(a.product_id AS BIGINT) AS e,
21
- min(TRY_CAST(a.rating AS DECIMAL(38, 18))) AS v
22
- FROM timestamps t
23
- JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
24
- JOIN eligible lb ON lb.ts = t.timestamp AND lb.e = a.product_id
25
- WHERE a.rating IS NOT NULL AND a.product_id IS NOT NULL
26
- GROUP BY 1, 2
27
- )
28
- SELECT ts AS "timestamp", e AS product_id, CAST(CASE WHEN v <= 1.0 THEN 1 ELSE 0 END AS BIGINT) AS low_rating_event
29
- FROM agg WHERE v IS NOT NULL
30
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-low-rating-event/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:6dabd06cbdd1c60882b5c0c2f0b190141108082da436ff2f383a9cf22cdccd19
3
- size 2397978
 
 
 
 
join-amazon-home/tasks/product-low-rating/manifest.yaml DELETED
@@ -1,30 +0,0 @@
1
- name: product-low-rating
2
- kind: forecast
3
- task_type: binary_classification
4
- description: Whether a product's average rating drops below 3.5 next window.
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: low_rating
8
- time_col: timestamp
9
- timedelta: 54 days 14:24:00
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-low-rating/train.parquet
13
- WITH eligible AS (
14
- SELECT DISTINCT t.timestamp AS ts, a.product_id AS e
15
- FROM timestamps t JOIN reviews a
16
- ON a."timestamp" > t.timestamp - 3 * INTERVAL '{timedelta}' AND a."timestamp" <= t.timestamp
17
- WHERE a.product_id IS NOT NULL
18
- ),
19
- agg AS (
20
- SELECT t.timestamp AS ts, CAST(a.product_id AS BIGINT) AS e,
21
- avg(TRY_CAST(a.rating AS DECIMAL(38, 18))) AS v
22
- FROM timestamps t
23
- JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
24
- JOIN eligible lb ON lb.ts = t.timestamp AND lb.e = a.product_id
25
- WHERE a.rating IS NOT NULL AND a.product_id IS NOT NULL
26
- GROUP BY 1, 2
27
- )
28
- SELECT ts AS "timestamp", e AS product_id, CAST(CASE WHEN v < 3.5 THEN 1 ELSE 0 END AS BIGINT) AS low_rating
29
- FROM agg WHERE v IS NOT NULL
30
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-low-rating/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:2be364259ad4c7220a0fc1fbb58acb2a38d3552112d1d8cf463ef53c6760dd3a
3
- size 2400282
 
 
 
 
join-amazon-home/tasks/product-max-review-len/manifest.yaml DELETED
@@ -1,27 +0,0 @@
1
- name: product-max-review-len
2
- kind: forecast
3
- task_type: regression
4
- description: Longest review length posted on a product next window.
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: max_review_len
8
- time_col: timestamp
9
- timedelta: 54 days 19:26:24
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-max-review-len/train.parquet
13
- WITH eligible AS (
14
- SELECT DISTINCT t.timestamp AS ts, a.product_id AS e
15
- FROM timestamps t JOIN reviews a
16
- ON a."timestamp" > t.timestamp - 3 * INTERVAL '{timedelta}' AND a."timestamp" <= t.timestamp
17
- WHERE a.product_id IS NOT NULL
18
- )
19
- SELECT t.timestamp AS "timestamp", CAST(a.product_id AS BIGINT) AS product_id,
20
- CAST(max(CAST(a.review_len AS DOUBLE)) AS DOUBLE) AS max_review_len
21
- FROM timestamps t
22
- JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
23
- JOIN eligible lb ON lb.ts = t.timestamp AND lb.e = a.product_id
24
- WHERE a.review_len IS NOT NULL AND a.product_id IS NOT NULL
25
- GROUP BY 1, 2
26
- HAVING max_review_len IS NOT NULL
27
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-max-review-len/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:09f53426ec84a955c291487bed9054fdfc92977c135aa7405f54b53039ae4f1b
3
- size 3611990
 
 
 
 
join-amazon-home/tasks/product-max-review-length/manifest.yaml DELETED
@@ -1,27 +0,0 @@
1
- name: product-max-review-length
2
- kind: forecast
3
- task_type: regression
4
- description: Longest review written for a home product next window.
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: max_review_len
8
- time_col: timestamp
9
- timedelta: 54 days 19:26:24
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-max-review-length/train.parquet
13
- WITH eligible AS (
14
- SELECT DISTINCT t.timestamp AS ts, a.product_id AS e
15
- FROM timestamps t JOIN reviews a
16
- ON a."timestamp" > t.timestamp - 3 * INTERVAL '{timedelta}' AND a."timestamp" <= t.timestamp
17
- WHERE a.product_id IS NOT NULL
18
- )
19
- SELECT t.timestamp AS "timestamp", CAST(a.product_id AS BIGINT) AS product_id,
20
- CAST(max(CAST(a.review_len AS DOUBLE)) AS DOUBLE) AS max_review_len
21
- FROM timestamps t
22
- JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
23
- JOIN eligible lb ON lb.ts = t.timestamp AND lb.e = a.product_id
24
- WHERE a.review_len IS NOT NULL AND a.product_id IS NOT NULL
25
- GROUP BY 1, 2
26
- HAVING max_review_len IS NOT NULL
27
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-max-review-length/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:09f53426ec84a955c291487bed9054fdfc92977c135aa7405f54b53039ae4f1b
3
- size 3611990
 
 
 
 
join-amazon-home/tasks/product-mean-rating/manifest.yaml DELETED
@@ -1,27 +0,0 @@
1
- name: product-mean-rating
2
- kind: forecast
3
- task_type: regression
4
- description: Predict the mean rating a home product receives in the next time window
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: pred_rating
8
- time_col: timestamp
9
- timedelta: 54 days 14:24:00
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-mean-rating/train.parquet
13
- WITH eligible AS (
14
- SELECT DISTINCT t.timestamp AS ts, a.product_id AS e
15
- FROM timestamps t JOIN reviews a
16
- ON a."timestamp" > t.timestamp - 3 * INTERVAL '{timedelta}' AND a."timestamp" <= t.timestamp
17
- WHERE a.product_id IS NOT NULL
18
- )
19
- SELECT t.timestamp AS "timestamp", CAST(a.product_id AS BIGINT) AS product_id,
20
- CAST(avg(CAST(a.rating AS DOUBLE)) AS DOUBLE) AS pred_rating
21
- FROM timestamps t
22
- JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
23
- JOIN eligible lb ON lb.ts = t.timestamp AND lb.e = a.product_id
24
- WHERE a.rating IS NOT NULL AND a.product_id IS NOT NULL
25
- GROUP BY 1, 2
26
- HAVING pred_rating IS NOT NULL
27
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-mean-rating/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:1e8bccbd16a03e0736b4f3f96bc2c745d44f2b10184b51988bdad0d7b76b7e58
3
- size 3028528
 
 
 
 
join-amazon-home/tasks/product-min-rating/manifest.yaml DELETED
@@ -1,27 +0,0 @@
1
- name: product-min-rating
2
- kind: forecast
3
- task_type: regression
4
- description: Lowest rating a product receives next window.
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: min_rating
8
- time_col: timestamp
9
- timedelta: 54 days 14:24:00
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-min-rating/train.parquet
13
- WITH eligible AS (
14
- SELECT DISTINCT t.timestamp AS ts, a.product_id AS e
15
- FROM timestamps t JOIN reviews a
16
- ON a."timestamp" > t.timestamp - 3 * INTERVAL '{timedelta}' AND a."timestamp" <= t.timestamp
17
- WHERE a.product_id IS NOT NULL
18
- )
19
- SELECT t.timestamp AS "timestamp", CAST(a.product_id AS BIGINT) AS product_id,
20
- CAST(min(CAST(a.rating AS DOUBLE)) AS DOUBLE) AS min_rating
21
- FROM timestamps t
22
- JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
23
- JOIN eligible lb ON lb.ts = t.timestamp AND lb.e = a.product_id
24
- WHERE a.rating IS NOT NULL AND a.product_id IS NOT NULL
25
- GROUP BY 1, 2
26
- HAVING min_rating IS NOT NULL
27
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-min-rating/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:854348b9d69bb2c3b2e47a660c14ec5fe14bae7b69cbb9bae7a3bf7689aa72d4
3
- size 2570400
 
 
 
 
join-amazon-home/tasks/product-popular-next-window/manifest.yaml DELETED
@@ -1,30 +0,0 @@
1
- name: product-popular-next-window
2
- kind: forecast
3
- task_type: binary_classification
4
- description: Whether a product's rating sum exceeds 25 next window.
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: popular
8
- time_col: timestamp
9
- timedelta: 54 days 14:24:00
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-popular-next-window/train.parquet
13
- WITH eligible AS (
14
- SELECT DISTINCT t.timestamp AS ts, a.product_id AS e
15
- FROM timestamps t JOIN reviews a
16
- ON a."timestamp" > t.timestamp - 3 * INTERVAL '{timedelta}' AND a."timestamp" <= t.timestamp
17
- WHERE a.product_id IS NOT NULL
18
- ),
19
- agg AS (
20
- SELECT t.timestamp AS ts, CAST(a.product_id AS BIGINT) AS e,
21
- sum(TRY_CAST(a.rating AS DECIMAL(38, 18))) AS v
22
- FROM timestamps t
23
- JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
24
- JOIN eligible lb ON lb.ts = t.timestamp AND lb.e = a.product_id
25
- WHERE a.rating IS NOT NULL AND a.product_id IS NOT NULL
26
- GROUP BY 1, 2
27
- )
28
- SELECT ts AS "timestamp", e AS product_id, CAST(CASE WHEN v > 25 THEN 1 ELSE 0 END AS BIGINT) AS popular
29
- FROM agg WHERE v IS NOT NULL
30
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-popular-next-window/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:6bca7fdf5daba440d7484f66e1eb8dd1b33b538f4bdd061f00433bd8be9a7466
3
- size 2382553
 
 
 
 
join-amazon-home/tasks/product-rating-max-heavytail/manifest.yaml DELETED
@@ -1,29 +0,0 @@
1
- name: product-rating-max-heavytail
2
- kind: forecast
3
- task_type: regression
4
- description: log1p of the maximum rating received on this product next window.
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: log_rating_max
8
- time_col: timestamp
9
- timedelta: 54 days 19:26:24
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-rating-max-heavytail/train.parquet
13
- WITH universe AS (
14
- SELECT DISTINCT CAST(product_id AS BIGINT) AS e FROM reviews
15
- WHERE product_id IS NOT NULL
16
- ),
17
- agg AS (
18
- SELECT t.timestamp AS ts, CAST(a.product_id AS BIGINT) AS e,
19
- max(TRY_CAST(a.rating AS DOUBLE)) AS v
20
- FROM timestamps t
21
- JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
22
- WHERE a.product_id IS NOT NULL
23
- GROUP BY 1, 2
24
- )
25
- SELECT t.timestamp AS "timestamp", u.e AS product_id, CAST(ln(1 + greatest(COALESCE(agg.v, 0), 0)) AS DOUBLE) AS log_rating_max
26
- FROM timestamps t
27
- CROSS JOIN universe u
28
- LEFT JOIN agg ON agg.ts = t.timestamp AND agg.e = u.e
29
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-rating-max-heavytail/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:d68dc261bfa26c93f2fef1de56447e60b1e9ddfb6ec546f1e0eb2d30936993f8
3
- size 88794958
 
 
 
 
join-amazon-home/tasks/product-rating-mean-heavytail/manifest.yaml DELETED
@@ -1,29 +0,0 @@
1
- name: product-rating-mean-heavytail
2
- kind: forecast
3
- task_type: regression
4
- description: log1p of the mean rating for this product next window (zero-dominated for inactive items).
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: log_rating_mean
8
- time_col: timestamp
9
- timedelta: 54 days 19:26:24
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-rating-mean-heavytail/train.parquet
13
- WITH universe AS (
14
- SELECT DISTINCT CAST(product_id AS BIGINT) AS e FROM reviews
15
- WHERE product_id IS NOT NULL
16
- ),
17
- agg AS (
18
- SELECT t.timestamp AS ts, CAST(a.product_id AS BIGINT) AS e,
19
- avg(TRY_CAST(a.rating AS DOUBLE)) AS v
20
- FROM timestamps t
21
- JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
22
- WHERE a.product_id IS NOT NULL
23
- GROUP BY 1, 2
24
- )
25
- SELECT t.timestamp AS "timestamp", u.e AS product_id, CAST(ln(1 + greatest(COALESCE(agg.v, 0), 0)) AS DOUBLE) AS log_rating_mean
26
- FROM timestamps t
27
- CROSS JOIN universe u
28
- LEFT JOIN agg ON agg.ts = t.timestamp AND agg.e = u.e
29
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-rating-mean-heavytail/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:336812bf6c1c8a45e38b3d5c17417c9ece957cdddff7bda3922a2bff02a8c92b
3
- size 89750332
 
 
 
 
join-amazon-home/tasks/product-rating-min-heavytail/manifest.yaml DELETED
@@ -1,29 +0,0 @@
1
- name: product-rating-min-heavytail
2
- kind: forecast
3
- task_type: regression
4
- description: log1p of the minimum rating received on this product next window (worst-review tail).
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: log_rating_min
8
- time_col: timestamp
9
- timedelta: 54 days 19:26:24
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-rating-min-heavytail/train.parquet
13
- WITH universe AS (
14
- SELECT DISTINCT CAST(product_id AS BIGINT) AS e FROM reviews
15
- WHERE product_id IS NOT NULL
16
- ),
17
- agg AS (
18
- SELECT t.timestamp AS ts, CAST(a.product_id AS BIGINT) AS e,
19
- min(TRY_CAST(a.rating AS DOUBLE)) AS v
20
- FROM timestamps t
21
- JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
22
- WHERE a.product_id IS NOT NULL
23
- GROUP BY 1, 2
24
- )
25
- SELECT t.timestamp AS "timestamp", u.e AS product_id, CAST(ln(1 + greatest(COALESCE(agg.v, 0), 0)) AS DOUBLE) AS log_rating_min
26
- FROM timestamps t
27
- CROSS JOIN universe u
28
- LEFT JOIN agg ON agg.ts = t.timestamp AND agg.e = u.e
29
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-rating-min-heavytail/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:816feb6714f6684a1b9645e91a7f27afb646e3924d7ee990374679aed4595508
3
- size 88851780
 
 
 
 
join-amazon-home/tasks/product-rating-sum-heavytail/manifest.yaml DELETED
@@ -1,29 +0,0 @@
1
- name: product-rating-sum-heavytail
2
- kind: forecast
3
- task_type: regression
4
- description: log1p of the sum of all star ratings received by this product next window (volume-weighted popularity).
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: log_rating_sum
8
- time_col: timestamp
9
- timedelta: 54 days 19:26:24
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-rating-sum-heavytail/train.parquet
13
- WITH universe AS (
14
- SELECT DISTINCT CAST(product_id AS BIGINT) AS e FROM reviews
15
- WHERE product_id IS NOT NULL
16
- ),
17
- agg AS (
18
- SELECT t.timestamp AS ts, CAST(a.product_id AS BIGINT) AS e,
19
- sum(TRY_CAST(a.rating AS DOUBLE)) AS v
20
- FROM timestamps t
21
- JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
22
- WHERE a.product_id IS NOT NULL
23
- GROUP BY 1, 2
24
- )
25
- SELECT t.timestamp AS "timestamp", u.e AS product_id, CAST(ln(1 + greatest(COALESCE(agg.v, 0), 0)) AS DOUBLE) AS log_rating_sum
26
- FROM timestamps t
27
- CROSS JOIN universe u
28
- LEFT JOIN agg ON agg.ts = t.timestamp AND agg.e = u.e
29
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-rating-sum-heavytail/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:dabeb41ca6731d2e6be1f0bb0627d0fcf28c3a916f101f0c2a86dd14794ecc57
3
- size 89813377
 
 
 
 
join-amazon-home/tasks/product-rating-trend/manifest.yaml DELETED
@@ -1,31 +0,0 @@
1
- name: product-rating-trend
2
- kind: forecast
3
- task_type: binary_classification
4
- description: Predict whether a home product's mean rating increases in the next window compared to the current window.
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: rating_trend
8
- time_col: timestamp
9
- timedelta: 54 days 14:24:00
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-rating-trend/train.parquet
13
- WITH cur AS (
14
- SELECT t.timestamp AS ts, CAST(a.product_id AS BIGINT) AS e,
15
- avg(TRY_CAST(a.rating AS DECIMAL(38, 18))) AS v
16
- FROM timestamps t JOIN reviews a
17
- ON a."timestamp" > t.timestamp - INTERVAL '{timedelta}' AND a."timestamp" <= t.timestamp
18
- WHERE a.rating IS NOT NULL
19
- GROUP BY 1, 2
20
- ),
21
- fut AS (
22
- SELECT t.timestamp AS ts, CAST(a.product_id AS BIGINT) AS e,
23
- avg(TRY_CAST(a.rating AS DECIMAL(38, 18))) AS v
24
- FROM timestamps t JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
25
- WHERE a.rating IS NOT NULL
26
- GROUP BY 1, 2
27
- )
28
- SELECT c.ts AS "timestamp", c.e AS product_id, CAST(CASE WHEN f.v > c.v THEN 1 ELSE 0 END AS BIGINT) AS rating_trend
29
- FROM cur c JOIN fut f ON f.ts = c.ts AND f.e = c.e
30
- WHERE c.v IS NOT NULL AND f.v IS NOT NULL AND c.e IS NOT NULL
31
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-rating-trend/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:72c5c3aae50295538e6312a8aa7a38a864f87f7fc8c77873517e8fd4c77658bd
3
- size 1653433
 
 
 
 
join-amazon-home/tasks/product-rating-volatility/manifest.yaml DELETED
@@ -1,27 +0,0 @@
1
- name: product-rating-volatility
2
- kind: forecast
3
- task_type: regression
4
- description: Predict the standard deviation of ratings for a home product in the next window, reflecting divided customer opinions.
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: rating_volatility
8
- time_col: timestamp
9
- timedelta: 54 days 14:24:00
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-rating-volatility/train.parquet
13
- WITH eligible AS (
14
- SELECT DISTINCT t.timestamp AS ts, a.product_id AS e
15
- FROM timestamps t JOIN reviews a
16
- ON a."timestamp" > t.timestamp - 3 * INTERVAL '{timedelta}' AND a."timestamp" <= t.timestamp
17
- WHERE a.product_id IS NOT NULL
18
- )
19
- SELECT t.timestamp AS "timestamp", CAST(a.product_id AS BIGINT) AS product_id,
20
- CAST(stddev_samp(CAST(a.rating AS DOUBLE)) AS DOUBLE) AS rating_volatility
21
- FROM timestamps t
22
- JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
23
- JOIN eligible lb ON lb.ts = t.timestamp AND lb.e = a.product_id
24
- WHERE a.rating IS NOT NULL AND a.product_id IS NOT NULL
25
- GROUP BY 1, 2
26
- HAVING COUNT(*) >= 2 AND rating_volatility IS NOT NULL
27
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-rating-volatility/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:5857f9acf848748e4999cbe9286ad10fda80b211383ab74d4b2f7acfb49ce52b
3
- size 1767545
 
 
 
 
join-amazon-home/tasks/product-return/manifest.yaml DELETED
@@ -1,38 +0,0 @@
1
- name: product-return
2
- kind: forecast
3
- task_type: binary_classification
4
- description: Predict whether a home product that received no reviews in the current window will receive reviews again in the next window.
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: product_returns
8
- time_col: timestamp
9
- timedelta: 54 days 14:24:00
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-return/train.parquet
13
- WITH recent AS (
14
- SELECT DISTINCT t.timestamp AS ts, CAST(a.product_id AS BIGINT) AS e
15
- FROM timestamps t JOIN reviews a
16
- ON a."timestamp" > t.timestamp - INTERVAL '{timedelta}' AND a."timestamp" <= t.timestamp
17
- WHERE a.product_id IS NOT NULL
18
- ),
19
- prev AS (
20
- SELECT DISTINCT t.timestamp AS ts, CAST(a.product_id AS BIGINT) AS e
21
- FROM timestamps t JOIN reviews a
22
- ON a."timestamp" > t.timestamp - 3 * INTERVAL '{timedelta}' AND a."timestamp" <= t.timestamp - INTERVAL '{timedelta}'
23
- WHERE a.product_id IS NOT NULL
24
- ),
25
- inactive AS (
26
- SELECT p.ts, p.e FROM prev p ANTI JOIN recent r ON r.ts = p.ts AND r.e = p.e
27
- ),
28
- gate AS (SELECT ts FROM inactive GROUP BY 1 HAVING COUNT(*) >= 5),
29
- fut AS (
30
- SELECT DISTINCT t.timestamp AS ts, CAST(a.product_id AS BIGINT) AS e
31
- FROM timestamps t JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
32
- WHERE a.product_id IS NOT NULL
33
- )
34
- SELECT i.ts AS "timestamp", i.e AS product_id, CAST(CASE WHEN f.e IS NULL THEN 0 ELSE 1 END AS BIGINT) AS product_returns
35
- FROM inactive i
36
- JOIN gate g ON g.ts = i.ts
37
- LEFT JOIN fut f ON f.ts = i.ts AND f.e = i.e
38
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-return/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:b5915861e5e8c2021b66b48b4b1718e21235d34a721ca1ae48d859415fe519cd
3
- size 2508717
 
 
 
 
join-amazon-home/tasks/product-review-count/manifest.yaml DELETED
@@ -1,32 +0,0 @@
1
- name: product-review-count
2
- kind: forecast
3
- task_type: regression
4
- description: Predict the number of reviews a home product receives in the next time window
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: predict_product_review_count
8
- time_col: timestamp
9
- timedelta: 54 days 14:24:00
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-review-count/train.parquet
13
- WITH eligible AS (
14
- SELECT DISTINCT t.timestamp AS ts, a.product_id AS e
15
- FROM timestamps t JOIN reviews a
16
- ON a."timestamp" > t.timestamp - 3 * INTERVAL '{timedelta}' AND a."timestamp" <= t.timestamp
17
- WHERE a.product_id IS NOT NULL
18
- ),
19
- fut AS (
20
- SELECT t.timestamp AS ts, CAST(a.product_id AS BIGINT) AS e, COUNT(*) AS n
21
- FROM timestamps t
22
- JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
23
- JOIN eligible lb ON lb.ts = t.timestamp AND lb.e = a.product_id
24
- WHERE a.product_id IS NOT NULL
25
- GROUP BY 1, 2
26
- ),
27
- live_ts AS (SELECT DISTINCT ts FROM fut)
28
- SELECT el.ts AS "timestamp", el.e AS product_id, CAST(COALESCE(f.n, 0) AS DOUBLE) AS predict_product_review_count
29
- FROM eligible el
30
- JOIN live_ts lt ON lt.ts = el.ts
31
- LEFT JOIN fut f ON f.ts = el.ts AND f.e = el.e
32
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-review-count/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:8a0fbbd1ff8e08e65afcd54702c43b530aa6f1330802907600539e2abfebfdb8
3
- size 6330704
 
 
 
 
join-amazon-home/tasks/product-review-length-change/manifest.yaml DELETED
@@ -1,31 +0,0 @@
1
- name: product-review-length-change
2
- kind: forecast
3
- task_type: regression
4
- description: Predict the change in mean review length for a home product between the current and next windows.
5
- entity_table: products
6
- entity_col: product_id
7
- target_col: review_length_change
8
- time_col: timestamp
9
- timedelta: 54 days 14:24:00
10
- sql: |-
11
- -- timestamps(timestamp) holds the seed anchors; this is a train-only release:
12
- -- timestamps := SELECT DISTINCT timestamp FROM tasks/product-review-length-change/train.parquet
13
- WITH cur AS (
14
- SELECT t.timestamp AS ts, CAST(a.product_id AS BIGINT) AS e,
15
- avg(CAST(a.review_len AS DOUBLE)) AS v
16
- FROM timestamps t JOIN reviews a
17
- ON a."timestamp" > t.timestamp - INTERVAL '{timedelta}' AND a."timestamp" <= t.timestamp
18
- WHERE a.review_len IS NOT NULL
19
- GROUP BY 1, 2
20
- ),
21
- fut AS (
22
- SELECT t.timestamp AS ts, CAST(a.product_id AS BIGINT) AS e,
23
- avg(CAST(a.review_len AS DOUBLE)) AS v
24
- FROM timestamps t JOIN reviews a ON a."timestamp" > t.timestamp AND a."timestamp" <= t.timestamp + INTERVAL '{timedelta}'
25
- WHERE a.review_len IS NOT NULL
26
- GROUP BY 1, 2
27
- )
28
- SELECT c.ts AS "timestamp", c.e AS product_id, CAST(f.v - c.v AS DOUBLE) AS review_length_change
29
- FROM cur c JOIN fut f ON f.ts = c.ts AND f.e = c.e
30
- WHERE c.v IS NOT NULL AND f.v IS NOT NULL AND c.e IS NOT NULL
31
- manifest_version: 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
join-amazon-home/tasks/product-review-length-change/train.parquet DELETED
@@ -1,3 +0,0 @@
1
- version https://git-lfs.github.com/spec/v1
2
- oid sha256:4163c1635c72507d6e9c4a9cd7bb6fadaae8d41d4567f89d79e1d7d162b29ab6
3
- size 3321492