Prepare RNN data — scale, window, reshape

Background

The lab's prepare_data turns a raw 1-D series into RNN-ready tensors in three moves:

1. Min-max scale to $[0, 1]$: $\tilde{x} = (x - \min)/(\max - \min)$.
2. Sliding window: each input is seq_len consecutive scaled values; the target is the next value.
3. Reshape inputs to (N, seq_len, 1) and targets to (N, 1) — the (B, T, F) / (B, O) shapes nn.RNN expects.

Problem statement

Implement prepare_rnn_data(series, seq_len) returning (X, y).

Input

• series — array-like of length L (raw values).
• seq_len — int, the window length.

Output

Returns (X, y): X is shape (L - seq_len, seq_len, 1), y is shape (L - seq_len, 1), both min-max scaled.

Examples

Example 1

Input:  series = [10, 20, 30, 40, 50], seq_len = 2
Output: X = [[[0.0],[0.25]], [[0.25],[0.5]], [[0.5],[0.75]]]   shape (3, 2, 1)
        y = [[0.5], [0.75], [1.0]]                              shape (3, 1)

Explanation: scaled series is [0, 0.25, 0.5, 0.75, 1.0]; windows of 2 predict the next value.

Constraints

• Scale the whole series first with global min/max.
• Window with stride 1: X[i] = scaled[i:i+seq_len], y[i] = scaled[i+seq_len].
• Reshape X to (N, seq_len, 1) and y to (N, 1).

Notes

• Scaling is essential: raw prices in the hundreds make the network's job far harder than values in $[0, 1]$.