Exercise 3: Time-Series Window Adapter

Goal

Build an adapter that creates derived objects.

The source object is a long temperature series:

TemperatureSeries(
    sensor_id="greenhouse-1",
    version=1,
    values=(20.0, 21.0, 22.0, 23.0, 24.0),
)

The client code wants training windows:

TrainingWindow(
    sensor_id="greenhouse-1",
    features=(20.0, 21.0, 22.0),
    target=23.0,
)

One series should become many windows.

---

Story

The forecasting team wants to train a model that predicts the next temperature reading.

The raw data is stored as one long series. The model code, however, wants many small examples:

previous readings -> next reading

So you will create an adapter that exposes a time series as an iterable of TrainingWindow objects.

---

Your task

Open exercise3.py and implement SeriesToWindowAdapter.

The adapter should:

1. accept a TemperatureSeries,
2. accept window_size,
3. accept horizon with default value 1,
4. validate that window_size > 0,
5. validate that horizon > 0,
6. generate TrainingWindow objects,
7. implement __iter__,
8. implement __len__.

---

Window rule

For this series:

values = (20.0, 21.0, 22.0, 23.0, 24.0)

With window_size=3 and horizon=1, generate:

features=(20.0, 21.0, 22.0), target=23.0
features=(21.0, 22.0, 23.0), target=24.0

With window_size=2 and horizon=2, generate:

features=(20.0, 21.0), target=23.0
features=(21.0, 22.0), target=24.0

The target index is:

start + window_size + horizon - 1

---

Run the tests

python exercise3.py