dantArrays 0.0.3

A thin wrapper over np.ndarray to provide an intuitive interface for attaching pydantic models as metadata to arrays.

dantArrays: Enhanced NumPy Arrays with Rich, Computed Metadata

---

Overview

dantArrays extends standard NumPy arrays by embedding rich, computed metadata directly into array slices, rows, or columns. This integration allows you to keep data tightly coupled with its descriptive context, computed metrics, or dynamic annotations—without losing NumPy's efficiency and versatility.

---

Key Features

• Rich Metadata Integration: Attach structured metadata (using Pydantic models) to slices of NumPy arrays.
• Automatic Computations: Define metadata fields computed directly from array data.
• Data Safety: Safe mechanisms to access and modify underlying array data without corrupting computed metadata.
• Flexibility: Easily update array slices or metadata, automatically recalculating dependent computed fields.
• Usability: Clear, Pythonic API for both simple and complex data manipulation tasks.

---

Installation

pip install dantarrays

---

Quickstart Example

Here's how easily you can add metadata to your arrays:

import numpy as np
from pydantic import BaseModel
from dantarrays import DantArray, mean_value_field

class StockMetadata(BaseModel):
    symbol: str
    mean_price: float = mean_value_field()

price_data = np.array([
    [100, 101, 102],
    [50, 49, 51],
])

stocks = DantArray(data=price_data, metadata_class=StockMetadata, major_axis=0)

stocks.update_metadata(0, symbol="AAPL")
stocks.update_metadata(1, symbol="MSFT")

print(stocks.meta(0).mean_price)  # Automatically computed mean price
# Output: 101.0

---

Usage

Defining Metadata Models

Use Pydantic models to define structured metadata:

from pydantic import BaseModel
from dantarrays import max_value_field, mean_value_field

class ExperimentMetadata(BaseModel):
    experiment_id: str
    max_measurement: float = max_value_field()
    avg_measurement: float = mean_value_field()

Creating a DantArray

import numpy as np
from dantarrays import DantArray

data = np.random.rand(5, 10)
experiments = DantArray(data=data, metadata_class=ExperimentMetadata, major_axis=0)

Updating Metadata

experiments.update_metadata(0, experiment_id="EXP001")

Accessing Metadata

metadata = experiments.get_metadata(0)
print(metadata.max_measurement)

Using the .meta() API for Convenient Field Access

The .meta() method provides ergonomic, attribute-style access to metadata fields:

experiments.meta(0).experiment_id = "EXP002"
print(experiments.meta(0).experiment_id)
# Output: EXP002

You can also update multiple fields conveniently:

experiments.meta(0).update(experiment_id="EXP003", avg_measurement=0.75)

Editing Data Safely

with experiments.edit_data() as data:
    data[0, :] = np.random.rand(10)

Immutable Updates

updated_experiments = experiments.with_updated_slice(0, np.zeros(10))

---

Advanced Examples

Batch Metadata Creation

indices = [0, 1, 2]
experiments.batch_create(indices, experiment_id="default")

Computed Fields Based on Data

Create a metadata field dynamically computed from the data slice:

from dantarrays import data_based_field

class ImageMetadata(BaseModel):
    brightness: float = data_based_field(lambda slice: float(np.mean(slice)))

Accessing Full Metadata

for idx, meta in experiments.get_existing_metadata():
    print(f"Experiment {meta.experiment_id} avg: {meta.avg_measurement}")

Data Validation and Safety

Direct access to the internal data array triggers warnings, ensuring safe data manipulation:

experiments._data_unsafe  # Warns about unsafe access

Contextual Computations

Computed fields have access to context about their array slice:

from dantarrays import ComputedFieldContext

class CustomComputation:
    def __call__(self, ctx: ComputedFieldContext):
        return ctx.size * np.mean(ctx.get_slice())

---

API Reference

DantArray

• .update_metadata(idx, **kwargs): Update metadata fields.
• .get_metadata(idx): Retrieve metadata.
• .meta(idx): Convenient, attribute-style metadata accessor.
• .batch_create(indices, **kwargs): Batch create metadata entries.
• .with_updated_data(new_data): Immutable data update.
• .apply_with_metadata(func): Apply function to array slices and metadata.
• .refresh_computed_fields(): Refresh computed fields after data changes.

---

Use Cases

• Financial time-series analysis
• Image datasets with annotations
• Scientific experiment tracking
• Feature metadata management in machine learning