Exercise 4: Cached Time-Series Adapter

Goal

Add caching to an adapter that generates many derived objects.

Exercise 3 created windows every time the adapter was instantiated.

That is correct, but wasteful when the same series and settings are adapted repeatedly.

---

Story

The forecasting team now trains for many epochs.

Each epoch needs the same training windows:

for epoch in range(10):
    windows = SeriesToWindowAdapter(series, window_size=24)
    train(windows)

If the raw series did not change, generating the same windows again is unnecessary.

So you will create a cached adapter.

---

Your task

Open exercise4.py and implement CachedSeriesToWindowAdapter.

The adapter should:

1. validate window_size and horizon,
2. build a cache key,
3. generate windows only on a cache miss,
4. store generated windows in a class-level cache,
5. implement __iter__,
6. implement __len__,
7. implement clear_cache() for tests.

---

Cache key

Use this key:

(series.sensor_id, series.version, window_size, horizon)

The key should include everything that affects the generated windows.

For this exercise, assume version changes whenever the raw series changes.

---

Expected behavior

CachedSeriesToWindowAdapter.clear_cache()

adapter1 = CachedSeriesToWindowAdapter(series, window_size=3)
adapter2 = CachedSeriesToWindowAdapter(series, window_size=3)

assert CachedSeriesToWindowAdapter.generation_count == 1
assert list(adapter1) == list(adapter2)

The second adapter should reuse the cached windows.

---

Run the tests

python exercise4.py