Yo tengo:

- 30,000 data points
- each data point is a measurement of type float
- each measurement is associated with a date
- each date has only one measurement
- no dates are without measurements
- the data comes in the form of a text file: 30,000 lines in this form:
    - YYYY-MM-DD I,F (e.g. 1977-02-08 20.74)
- measurement appearing in the source file are already sorted by date

Necesito:

- a time-interval T with boundaries (s,e) /* start, end */
- (s - e = 14 days) the time-interval *must* be 2 weeks
- define min as the lowest value in the interval T
- define max as the greatest value in the interval T
- the chosen T needs to have the greatest distance btwn max and min of all possible Ts
- break ties among intervals T by choosing the most recent (with the greatest s value)
- the chosen T must consider all jumps in the 14 days, not just the values @ s and e
- if the overall "variance" in the interval is great but the jump 
  |max-min| is not the greatest in absolute value, T is not the right choice,
  even if it's an "exciting" interval

Estoy preguntando:

- which algorithm to employ, considering algorithms are not my specialty
- which data structure to use to keep track of the subtotals

Nota:

- an answer in pseudo code would be preferred, "prose" is fine if pressured for time
- an answer in Python would be... splendid :)

Si lo desea, puede generar datos "ficticios" y ejecutar el algoritmo propuesto como una prueba o podría compartir los datos reales.

No me interesa el rendimiento tanto aquí, aparte de querer saber la forma más rápida de hacerlo para aprender a aplicar la solución correcta y el algoritmo correcto.

Creo que puedo "probar" la corrección incluso con el algoritmo iterativo más simple porque el conjunto de datos es pequeño en las computadoras de hoy.

Hasta ahora, estoy "recorriendo y transportando 14 vectores de 14 mediciones", si pudiera enseñarme cómo hacerlo de forma incremental con sumas secundarias, eso sería realmente apreciado.