Data science course in Peru

Our Spanish page, curso de data science en inglés, starts from the spreadsheet, uses a university enrolment dataset from Huánuco and teaches how a fake zero distorts an average. Here we take a weather dataset and a different trap.

import csv

FILE = "IGP_EstacionEMA_data_2018_2025.csv"

coldest = {}                                  # (year, month, day) -> lowest TT
with open(FILE, encoding="utf-8") as f:
    for row in csv.DictReader(f):
        if row["TT"] == "":
            continue
        day = (row["year"], row["month"], row["day"])
        coldest[day] = min(coldest.get(day, 99.0), float(row["TT"]))

frost = {}
for (year, month, day), low in coldest.items():
    if low < 0:
        frost[year] = frost.get(year, 0) + 1

for year in range(2018, 2026):
    print(year, "frost nights:", frost.get(str(year), 0))

If winters really got colder, the number of cold hours would rise too, not only the hours below zero. Version 2 counts, for June and July of each year, every hour under 3 °C, and how many of those were under 0 °C. It also finds the first sub-zero reading in the whole file and counts the blanks.

import csv

FILE = "IGP_EstacionEMA_data_2018_2025.csv"

cold = {}       # year -> hours under 3 °C in June and July
below = {}      # year -> of those, hours under 0 °C
blank = 0
first_negative = None
with open(FILE, encoding="utf-8") as f:
    for row in csv.DictReader(f):
        if row["TT"] == "":
            blank += 1
            continue
        t = float(row["TT"])
        if t < 0 and first_negative is None:
            first_negative = (row["year"], row["month"], row["day"], row["hour"])
        if row["month"] in ("6", "7") and t < 3:
            cold[row["year"]] = cold.get(row["year"], 0) + 1
            if t < 0:
                below[row["year"]] = below.get(row["year"], 0) + 1

print("blank TT readings:", blank)
print("first reading below 0:", first_negative)
for year in range(2018, 2026):
    y = str(year)
    print(y, "hours under 3 C:", cold.get(y, 0), "of which under 0 C:", below.get(y, 0))

Data: Instituto Geofísico del Perú, Estación Meteorológica Automática, Huancayo Observatory, Huayao, via the Plataforma Nacional de Datos Abiertos, Open Data Commons Attribution License (ODC-By). Counts are our own, computed on 9 November 2026 with the programs above.

The cold hours do not jump in 2021. They wander between 222 and 349 a season across all eight years. What jumps is the share of them below zero: none for three years, then a large share every year. In the earlier years, the coldest July reading never went lower than about 0.2 °C, and the first negative value in the whole file appears on 27 May 2021. That pattern fits a change in how temperatures were recorded or processed at least as well as it fits a change in the weather.

We have not asked the IGP, and this page does not claim to know the cause. That is the point of the lesson. The honest report says: "In this file, sub-zero readings begin in May 2021; the number of cold hours did not change in the same way; before comparing frost across that date, confirm with the publisher whether the sensor or processing changed." A data scientist who writes that sentence has done more useful work than one who fits a trend line to the first output.

1. Look at the shape. Plot a histogram of July temperatures for 2019 and for 2022 on the same axes. Where do the two shapes differ?
2. Test one idea. If negative values had lost their minus sign before 2021, what would you expect to see between 0 and 3 °C in 2018 to 2020? Does the table above agree?
3. Report the blanks. 700 hourly readings are blank. Which months do they fall in, and could they affect any count on this page?
4. Write to the publisher. Draft a short, polite email to a data provider asking about a suspected change, with the evidence in three lines.

Our plan, adjusted to each learner. Our main adult course runs 10 to 12 months from zero; the teen course is six months for learners who already know basic Python. We make no job or salary promises.

America/Lima stays at UTC minus 5 and has had no summer time since 1994 (tz database); Asia/Kolkata stays at UTC plus 5:30. India is 10 hours 30 minutes ahead throughout the year.

You pick in your own time zone when you book, and the mentor sees the converted time. Huancayo, Lima and the rest of Peru share one time zone, so the same slot works wherever you are in the country.