Streaming Min-Max Normalization
Streaming Min-Max Normalization
Implement streaming min-max normalization: given multiple batches of data, update per-feature running min and max, and normalize each incoming batch.
Streaming normalization processes data in real-time without knowing the full dataset statistics upfront. Your implementation should maintain running minimum and maximum values for each feature, updating them as new batches arrive, then normalize each batch using the current global statistics.
Normalization Formula:
x′=max−min+ϵx−minFunction Arguments
D: int- Number of featuresstate: dict- Contains 'min' and 'max' arrays (shape D,)X_batch: array-like, shape (B, D)- Input batcheps: float- Small value to avoid division by zero
Examples
Input: streaming_minmax_init(D=2), streaming_minmax_update(state, [[1,3],[2,1]])
Output: streaming_minmax_init({'min': [inf,inf], 'max': [-inf,-inf]}), streaming_minmax_update([[0,1],[1,0]])
Input: streaming_minmax_init(D=1), streaming_minmax_update(state, [[5],[3]])
Output: streaming_minmax_init({'min': [inf], 'max': [-inf]}), streaming_minmax_update([[1],[0]])
Hint 1
Initialize with np.full() for min and max arrays.
Hint 2
Use np.minimum() and np.maximum() to update running statistics.
Hint 3
Use np.maximum() to handle constant features safely with eps.
Requirements
- On init: min=+inf, max=-inf
- On update: state['min']=min(state['min'], batch_min); same for max
- Normalize using the updated min/max (post-update) for consistency
- Handle constant columns (range≈0) via eps
- Pure NumPy implementation
Constraints
- Handles constant columns (range≈0) via eps
- Batches can arrive in any order/size
- NumPy only; time limit: 300ms
Log in to take notes on this problem
Accepts: number
Accepts: array
Streaming Min-Max Normalization
Streaming Min-Max Normalization
Implement streaming min-max normalization: given multiple batches of data, update per-feature running min and max, and normalize each incoming batch.
Streaming normalization processes data in real-time without knowing the full dataset statistics upfront. Your implementation should maintain running minimum and maximum values for each feature, updating them as new batches arrive, then normalize each batch using the current global statistics.
Normalization Formula:
x′=max−min+ϵx−minFunction Arguments
D: int- Number of featuresstate: dict- Contains 'min' and 'max' arrays (shape D,)X_batch: array-like, shape (B, D)- Input batcheps: float- Small value to avoid division by zero
Examples
Input: streaming_minmax_init(D=2), streaming_minmax_update(state, [[1,3],[2,1]])
Output: streaming_minmax_init({'min': [inf,inf], 'max': [-inf,-inf]}), streaming_minmax_update([[0,1],[1,0]])
Input: streaming_minmax_init(D=1), streaming_minmax_update(state, [[5],[3]])
Output: streaming_minmax_init({'min': [inf], 'max': [-inf]}), streaming_minmax_update([[1],[0]])
Hint 1
Initialize with np.full() for min and max arrays.
Hint 2
Use np.minimum() and np.maximum() to update running statistics.
Hint 3
Use np.maximum() to handle constant features safely with eps.
Requirements
- On init: min=+inf, max=-inf
- On update: state['min']=min(state['min'], batch_min); same for max
- Normalize using the updated min/max (post-update) for consistency
- Handle constant columns (range≈0) via eps
- Pure NumPy implementation
Constraints
- Handles constant columns (range≈0) via eps
- Batches can arrive in any order/size
- NumPy only; time limit: 300ms
Log in to take notes on this problem
Accepts: number
Accepts: array