Slight tangent: Domain-specific languages (DSLs) that blew my mind.
Here are three DSLs that left lasting and positive impressions:
1) Mathematical programming languages, first GNU MathProg, then later python pulp and pyomo. These launched a subcareer in applying LP and MIP to practical, large-scale problems. The whole field is fascinating, fun, and often highly profitable because the economic gains from optimization can be huge. Think of UPS.
2) Probabilistic programming languages, first BUGS (winbugs, openbugs, JAGS), later Stan, pymc, and various specialized tools. These helped me become a practicing Bayesian who doesn't much like p-value hypothesis testing.
3) The dplyr | ggplot universe. These are really two independent eDSLs but they're from the same folks and work great together. Writing awkward pandas|matplotlib code is just soul-wrecking tedium after using dplyr|ggplot.
Here are three DSLs that left lasting and positive impressions:
1) Mathematical programming languages, first GNU MathProg, then later python pulp and pyomo. These launched a subcareer in applying LP and MIP to practical, large-scale problems. The whole field is fascinating, fun, and often highly profitable because the economic gains from optimization can be huge. Think of UPS.
2) Probabilistic programming languages, first BUGS (winbugs, openbugs, JAGS), later Stan, pymc, and various specialized tools. These helped me become a practicing Bayesian who doesn't much like p-value hypothesis testing.
3) The dplyr | ggplot universe. These are really two independent eDSLs but they're from the same folks and work great together. Writing awkward pandas|matplotlib code is just soul-wrecking tedium after using dplyr|ggplot.