MHkhan50 commited on
Commit
2ca913d
·
verified ·
1 Parent(s): 6c9372b

DialysisFlow-AgentBench v0.2.1 scenarios and policy results

Browse files
Files changed (4) hide show
  1. .gitattributes +1 -0
  2. README.md +80 -0
  3. data/policy_results.csv +0 -0
  4. data/scenarios.jsonl +3 -0
.gitattributes CHANGED
@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
 
 
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
61
+ data/scenarios.jsonl filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,80 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ pretty_name: DialysisFlow-AgentBench
4
+ language:
5
+ - en
6
+ tags:
7
+ - synthetic
8
+ - operations-research
9
+ - scheduling
10
+ - multi-agent
11
+ - healthcare-operations
12
+ - benchmark
13
+ size_categories:
14
+ - 1K<n<10K
15
+ task_categories:
16
+ - other
17
+ configs:
18
+ - config_name: scenarios
19
+ data_files: data/scenarios.jsonl
20
+ - config_name: policy_results
21
+ data_files: data/policy_results.csv
22
+ ---
23
+
24
+ # DialysisFlow-AgentBench
25
+
26
+ 1,200 synthetic dialysis networks in which one unit closes some or all of its shifts and its patients must be re-placed in free chairs at nearby units, with a shared patient-transport fleet. The dataset also holds the result of nine placement policies on every network, including two exact MILP optima.
27
+
28
+ > **Synthetic operational data only.** No patient data, no clinical information. Nothing here supports clinical or treatment decisions.
29
+
30
+ - Code, paper and full method: [github.com/MHKhan50/dialysisflow-agentbench](https://github.com/MHKhan50/dialysisflow-agentbench)
31
+ - Software DOI (all versions): [10.5281/zenodo.24354744](https://doi.org/10.5281/zenodo.24354744)
32
+ - Matches release v0.2.1. Every scenario is regenerated from its seed by `scripts/export_hf.py`.
33
+
34
+ ## Configurations
35
+
36
+ ### `scenarios` (1,200 rows, 41,794 displaced patients)
37
+
38
+ | Field | Meaning |
39
+ |---|---|
40
+ | `scenario_id` | `S01000` to `S02199`; the number is the generator seed |
41
+ | `scarcity` | `low`, `medium` or `high`: 60–75%, 75–85% or 85–95% of chairs already in use |
42
+ | `severity` | `partial` (two shifts closed) or `full` (all four) |
43
+ | `units` | receiving units: `unit`, position `x_km`, `y_km` (the closed unit is at 20, 20) and `free_chairs` per shift |
44
+ | `transport_seats` | spare seats in the shared transport fleet, per shift, for the whole network |
45
+ | `patients` | displaced patients: `pid`, current `shift`, `cohort` (`transport` or `independent`), `travel_limit_min` (45 or 75), `travel_min` to each unit, and `options`, the feasible `unit|shift` pairs |
46
+
47
+ Shifts are `MWF-AM`, `MWF-PM`, `TTS-AM`, `TTS-PM`. A patient keeps their weekday pattern; moving between morning and afternoon is allowed and adds 20 minutes of burden. Travel time is 5 minutes plus 1.5 minutes per km.
48
+
49
+ ### `policy_results` (10,800 rows: 1,200 scenarios × 9 policies)
50
+
51
+ `scenario_id`, `scarcity`, `severity`, `policy`, `n_patients`, `fulfilment` (share placed), `unmet_transport`, `unmet_independent`, `cohort_gap`, `mean_burden` (minutes), `time_changes`, `transport_overbooked`, `chair_overbooked`, `travel_violations`, `feasible` (1 if the plan respects every chair, seat and travel limit).
52
+
53
+ Policies: `list_order_greedy`, `most_constrained_greedy`, `independent_units`, `independent_units_gated`, `quota_units`, `repeated_independent_units`, `coordinated_units`, `optimal_throughput`, `optimal_fair`.
54
+
55
+ ## Use
56
+
57
+ ```python
58
+ from datasets import load_dataset
59
+
60
+ scenarios = load_dataset("MHkhan50/dialysisflow-agentbench", "scenarios", split="train")
61
+ results = load_dataset("MHkhan50/dialysisflow-agentbench", "policy_results", split="train")
62
+ ```
63
+
64
+ A new placement policy can be scored against the same networks and compared with the exact optima in `policy_results`.
65
+
66
+ ## Limitations
67
+
68
+ The networks are stylised: straight-line travel, one closure at a time, a fixed transport share of about 35%, and no clinical priority between patients. Results describe these networks, not any real dialysis service.
69
+
70
+ ## Citation
71
+
72
+ ```bibtex
73
+ @software{hussain_dialysisflow_agentbench,
74
+ author = {Hussain, Muzammal},
75
+ title = {DialysisFlow-AgentBench: re-placing displaced dialysis patients after a unit closure},
76
+ version = {0.2.1},
77
+ doi = {10.5281/zenodo.24354744},
78
+ url = {https://github.com/MHKhan50/dialysisflow-agentbench}
79
+ }
80
+ ```
data/policy_results.csv ADDED
The diff for this file is too large to render. See raw diff
 
data/scenarios.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2e282fcec8788b485d8000c0d9900d775aee253568a8e7bca3cde2ea87047fa0
3
+ size 11565768