Performance Issues¶
Large arrays are slow¶
Check which backend is active. Without NumPy, all operations use pure Python loops, which can be orders of magnitude slower for large arrays.
Fix: install NumPy:
Verify it is being used (see Check which backend is active).
If you have set MARPLE_BACKEND=none, remove or change that setting:
Rank operator is slow on many cells¶
(f⍤k) calls f once per cell. For scalar functions, this is much slower than the built-in element-wise operations:
10 + M ⍝ fast: built-in scalar extension with NumPy
10 (+⍤0) M ⍝ slow: per-element dfn call overhead
Fix: use scalar primitives directly when possible. Reserve the rank operator for operations that genuinely need per-cell application (sorting rows, custom reductions, etc.).
Inner product is slow¶
The fast path for +.× uses numpy.dot. Other inner products (e.g., ⌈.+) fall back to per-element Python loops.
Fix: if you need fast matrix multiplication, use +.× which is accelerated. For other inner products, consider whether you can restructure the computation.
Memory usage¶
MARPLE stores flat arrays. When NumPy is active, numeric arrays are stored as NumPy ndarrays (typically 8 bytes per element). Without NumPy, they are Python lists (roughly 28+ bytes per element).
Fix: install NumPy to reduce memory usage for large numeric arrays.