Skip to content
SynthStats
SynthStats
ReportBrief
DemonstrationsResearch resultsSupporting resultsHow one program is scored
Dugong growth (length against age)Disease spread (infected count against time, population 50)
Dugong growth programsDisease spread programs

Files

ResourcesSlidesData

Slides

Eleven slides: what a demonstration is; the reference program; the reference and the first single attempt on both tasks; the scores of all eighteen scored programs; score and coverage tables; the dugong and disease-spread observations and program predictions; and the public sources. Open the PDF, or open the slides full screen; the arrow keys move between slides.

The title slide reads SynthStats demonstrations and names the SynthStats team. The title slide reads SynthStats demonstrations and names the SynthStats team. Eight PyMC programs predict dugong length from age: four came from single calls, four from a two-round loop. Eight PyMC programs predict dugong length from age: four came from single calls, four from a two-round loop. The slide gives the reference PyMC program for dugong length against age, with Normal priors and a Normal likelihood. The slide gives the reference PyMC program for dugong length against age, with Normal priors and a Normal likelihood. Dugong length against age: both the reference and single attempt 1 predict slowing growth, with visible and held-out observations. Dugong length against age: both the reference and single attempt 1 predict slowing growth, with visible and held-out observations. Infected count among 50 individuals against time: predictions of the reference and single attempt 1, with visible and held-out counts. Infected count among 50 individuals against time: predictions of the reference and single attempt 1, with visible and held-out counts. Held-out scores of twenty programs: every scored model-written program beats the floor model, which ignores the input. Held-out scores of twenty programs: every scored model-written program beats the floor model, which ignores the input. Table of held-out scores and coverage for ten dugong programs: single attempts 2 and 4 beat the reference. Table of held-out scores and coverage for ten dugong programs: single attempts 2 and 4 beat the reference. Table of held-out scores and coverage for ten disease-spread programs: two failed at model build, and none beat the reference. Table of held-out scores and coverage for ten disease-spread programs: two failed at model build, and none beat the reference. Dugong length against age with predictions from all ten programs; single attempts 2 and 4 start lower than the reference. Dugong length against age with predictions from all ten programs; single attempts 2 and 4 start lower than the reference. Infected count against time with predictions from eight scored programs; the two that failed at model build show observations only. Infected count against time with predictions from eight scored programs; the two that failed at model build show observations only. The public sources, as three links: synthstats.org, SynthStats on GitHub, and SynthStats on Hugging Face. The public sources, as three links: synthstats.org, SynthStats on GitHub, and SynthStats on Hugging Face.
SynthStats team

Resources

Reports, code, data and papers.

Data

Next Page