Add NumPy-compatible API layer for out-of-core matrix operations

[Copilot]

Provides familiar NumPy interface to enable user migration while leveraging Paper's out-of-core capabilities for datasets larger than memory.

Implementation

Core API (paper/numpy_api.py)

• ndarray class: Lazy evaluation with NumPy-like operations (+, *, @), properties (shape, dtype, T), and public to_numpy() method
• Creation functions: array(), zeros(), ones(), eye(), random_rand()
• I/O: load(), save()

Bug Fix (paper/plan.py)

Fixed pre-existing scalar multiplication bug where EagerNode matrices were closed prematurely, causing segfaults on reuse:

# Only close intermediate computed results, not input matrices
if not isinstance(self.left, EagerNode):
    TMP.close()

Testing

26 new tests covering array creation, operations, error handling, I/O, and large array support. All 62 tests passing.

Usage

from paper import numpy_api as pnp

# Create arrays with familiar syntax
a = pnp.array([[1, 2], [3, 4]])
b = pnp.array([[5, 6], [7, 8]])

# Build lazy computation plan
c = (a + b) * 2  # Automatic operator fusion

# Execute and convert to NumPy
result = c.compute().to_numpy()

Limitations

• Currently supports 2D matrices only (foundational version)
• Scalar addition not yet implemented

Documentation

• Quick start in README
• examples/numpy_api_example.py: 9 comprehensive examples
• examples/simple_demo.py: minimal demo
• NUMPY_API_SUMMARY.md: complete implementation details

Original prompt

---

This section details on the original issue you should resolve

<issue_title>NumPy-compatible API Layer</issue_title>
<issue_description>Rationale: This is the foundation for user adoption. Without NumPy compatibility, users won't migrate from existing solutions. It leverages your deep understanding of matrix operations from SystemDS while providing the familiar interface developers expect.​
</issue_description>

Comments on the Issue (you are @copilot in this section)

• Fixes NumPy-compatible API Layer #34

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.

[j143]

as a first draft this is good.