numpy-structured

Overview

Structured arrays allow ndarrays to contain data with different types in named "fields," mimicking C structs. They are used primarily for interfacing with binary data from external sources and interpreting complex memory layouts without converting to high-level objects like Pandas DataFrames.

When to Use

• Interpreting binary blobs or file headers from C/C++ applications.
• Storing tabular data where each row has multiple related attributes (e.g., ID, Timestamp, Value).
• Mapping hardware-aligned buffers to named fields for easier access.
• Performing multi-field updates on a shared data buffer.

Decision Tree

1. Need to map names to columns in a single array buffer?
   • Use a structured array with a compound dtype.
2. Working with binary data from a file?
   • Use arr.view(dtype=your_struct_dtype) to interpret bytes without copying.
3. Selecting multiple fields?
   • Use a list of names arr[['id', 'name']]. This returns a view.

Workflows

1. Defining a Complex Data Record

   • Construct a dtype using a list of tuples: [('id', 'i4'), ('coord', 'f8', (3,))].
   • Instantiate the array with the custom dtype.
   • Access the 'coord' field to get an (N, 3) view of the coordinates.

2. Mapping Binary Data to Structs

   • Load raw bytes from a file or buffer.
   • View the buffer as a structured array: arr.view(dtype=my_struct_dtype).
   • Access named fields to extract typed data from the raw binary.

3. Multi-field View Modification

   • Select a subset of fields using a list of names: arr[['field1', 'field2']].
   • Update the resulting view with a new value.
   • Verify the original structured array has been modified in the specified fields.

Non-Obvious Insights

• Positional Assignment: Assignment between two structured arrays is based on field position, not field name. If field 1 is 'id' in Array A and 'val' in Array B, Array B's 'val' will be assigned to Array A's 'id'.
• Scalar Broadcasting: Assigning a single scalar to a structured element (a whole row) will assign that value to all fields in the row.
• Pandas vs Structured: While structured arrays handle tabular data, they lack the high-level analysis features of Pandas; use them for low-level memory mapping and binary I/O, not statistical analysis.

Evidence

• "Structured datatypes are designed to be able to mimic ‘structs’ in the C language, and share a similar memory layout." Source
• "Assignment between two structured arrays occurs as if the source elements had been converted to tuples... regardless of field names." Source

Scripts

• scripts/numpy-structured_tool.py: Demonstrates defining compound dtypes and field views.
• scripts/numpy-structured_tool.js: Simulated struct field access.

Dependencies

• numpy (Python)

References

• references/README.md