NumCircBuf: High-performance numerical circular buffers for Python, featuring O(1) statistical accumulators.

Overview
Features
Installation
Common Buffer API
- The Buffer View (ViewProtocol)
Main Buffers
- 1. OverwriteCircBuffer
- 2. BlockingCircBuffer
Utility Buffers
Exception Handling
- Exceptions
- Warnings
Performance
Documentation
Changelog
License
Support
Citation
Acknowledgements

Overview

NumCircBuf is a high-performance Python library providing numerical circular buffers, featuring O(1) accumulators and specialized calculation variants. Built with Cython for a balance between speed and maintainability, it provides efficient data structures for real-time signal processing, time-series analysis, and other performance-critical applications.

Features

Multiple Buffer Types – Specialized implementations for different use cases:
- BlockingCircBuffer – Blocking producer/consumer circular buffer for multi-threaded applications.
- OverwriteCircBuffer – Optimized for high-throughput writes
- IntegratedGatedBuffer – Specifically for calculating gated loudness statistics
- O(1) Accumulators – Constant-time operations for statistics, implemented in specialized buffers:
  - RunningMeanBuffer – O(1) mean
  - RunningMeanSqBuffer – O(1) mean-square
High Performance – Bypasses Python/NumPy overhead to saturate the hardware bandwidth. Details in PERFORMANCE.md:
- vs. collections.deque & Python lists: 500–1500× faster for bulk extend, and 1.5–3× faster for single append.
- vs. Optimized NumPy Ring Buffers: Up to 10× faster.
Familiar API: Buffers use append, extend, and clear methods for drop-in compatibility.
NumPy Integration: Direct integration with NumPy arrays.
Type Safety: Fully typed; supports fp32, fp64, int32, int64, uint32, and uint64.
Docs & Benchmarks: Includes extensive API documentation, usage examples, and detailed performance benchmarks.

Installation

pip install numcircbuf

Or install from source:

git clone https://github.com/basimali-ai/NumCircBuf.git
cd NumCircBuf
pip install .

For development installation with all dependencies (from source):

git clone https://github.com/basimali-ai/NumCircBuf.git
cd NumCircBuf
pip install -e .[dev]

Verification

To verify the installation and check the version:

pip show numcircbuf && python -c "import numcircbuf; print('\n' + '-'*20 + '\nRunning smoke test...'); print(f'Library version: {numcircbuf.__version__}'); print(numcircbuf.OverwriteCircBuffer(10, 'never')); print(numcircbuf.RunningMeanBuffer(10, 'calculation')); print('-'*20 + '\n--- Installation verification successful ---')"

Common Buffer API

All buffer types expose the following methods:

clear(): Clears the buffer.
clear_nan(): Removes NaN values.
clear_infs(): Removes Inf values.
__len__(): Current buffer size.
maxlen: Maximum buffer capacity.
view(): Returns a read-only logical view of the buffer (implements ViewProtocol).

All bulk extend methods have an additional warn_size boolean argument which enables or disables warnings when the block size exceeds the buffer's maxlen. It is enabled by default.

Note: In all buffers nan/inf values are allowed to be appended/extended with, as they are valuable data points.

The Buffer View (ViewProtocol)

The object returned by the view() method provides a zero-copy, read-only logical view of the circular buffer. It is exposed as a structural Protocol to provide strong type hints and an explicit contract, while the underlying implementation remains internal.

Behaves like a 1D NumPy array in logical order.
Has a to_numpy() function which provides a contiguous NumPy array copy of the data in logical order.
All returned arrays are independent (no shared memory).
Any slicing produces a 1D NumPy array copy containing only the selected elements.
Indexing or iteration preserves logical order, but yields native Python objects (int, float) for each element.

Note: This view is strictly read-only. Attempts to modify the view via indexing or deletion will raise an InvalidModification exception at runtime.

Usage Example

# Assuming `buffer` is a populated circular buffer instance from numcircbuf
view = buffer.view()

def view_usage(v):
    print("All elements (slice):", v[:])
    print("Single element:", v[0])
    print("Number of elements:", len(v))
    print("Max capacity:", v.maxlen)
    print("Dtype:", v.dtype)
    print("-" * 20)
    print("Iterating over view:")
    for value in v:
        print("-" * 10)
        print(f"Value = {value}")
        print(f"Type = {type(value)}")  # Will be native <class 'int'> or <class 'float'>
    print("-" * 20)

print("\n--- View Usage ---")
view_usage(view)

# Copy behavior
# All returned arrays are independent; no shared memory
arr = view.to_numpy()
print("Before: ", view[:])
arr *= 2.0  # modify the array copy
print("After: ", view[:])  # View remains unchanged

Main Buffers

1. OverwriteCircBuffer

Write-optimized circular buffer with auto-overwrite and non-destructive reads.

Provides simple vectorized mathematical metrics over the buffer contents.

Note: This buffer is not thread safe.

Concurrent reads during writes can provide inaccurate data
Concurrent writes can cause data corruption.

Constructor

def __init__(
    self,
    maxlen: int,
    return_overwritten_policy: Literal["never", "always", "conditional"],
    dtype: (
        type[np.float32]
        | type[np.float64]
        | type[np.int32]
        | type[np.int64]
        | type[np.uint32]
        | type[np.uint64]
    ) = np.float64,
) -> None:

Overwrite Return Policy

Controls whether overwritten elements are returned when the buffer wraps.

return_overwritten_policy: Literal["always", "never", "conditional"]

"never"

from numcircbuf import OverwriteCircBuffer
buf = OverwriteCircBuffer(maxlen=3, return_overwritten_policy="never")
print(buf.extend([1, 2]))  # Empty NumPy array []
print(buf.extend([3, 4]))  # Empty NumPy array []
print(buf.append(5))  # None

"always"

from numcircbuf import OverwriteCircBuffer
buf = OverwriteCircBuffer(maxlen=3, return_overwritten_policy="always")
print(buf.extend([1, 2]))  # Empty NumPy array []
print(buf.extend([3, 4]))  # NumPy array [1.0]
print(buf.append(5))  # float 2.0

"conditional"

The returned values depend on the return_overwritten flag, it defaults to False.

from numcircbuf import OverwriteCircBuffer
buf = OverwriteCircBuffer(maxlen=3, return_overwritten_policy="conditional")
print(buf.extend([1, 2]))  # Empty NumPy array []
print(buf.extend([3, 4]))  # Empty NumPy array []
print(buf.extend([5, 6], return_overwritten=True))  # NumPy array [2.0, 3.0]
print(buf.append(7))  # None
print(buf.append(8, return_overwritten=True))  # float 5.0

Supported input types

.append(float | int)
- Accepts a single numeric value.
- Value is stored internally as the buffer’s dtype.
.extend(Iterable[float | int])
- Accepts any Iterable of numeric values.
- The iterable is converted to a contiguous NumPy array of the buffer’s dtype.
- If you pass an np.ndarray, it is only copied if,
  - Its dtype does not match the buffer’s
  - It is not C-contiguous
.extend_unchecked(np.ndarray)
- Expects only a C-contiguous 1-D np.ndarray with the same dtype as the buffer.
- This method skips dtype, contiguous array and dimension conversions/checks.
Note: Using this yields the best performance, but if the array's dtype is different or if it is not contiguous, it will cause silent data corruption or crashes.

Mathematical Metrics

These metrics are computed on all elements.

.mean(): Mean of all elements.
.sum(): Sum of all elements.
.sum_squares(): Sum of squares.
.mean_squares(): Mean of squares.
.sum_and_count_gt(threshold: float | int): Returns a sum and count of elements above a threshold.

2. BlockingCircBuffer

Blocking producer/consumer circular buffer, suitable for multi-threaded applications.

This buffer blocks under these conditions:

The writer will wait if the buffer is full.
The reader will wait if the buffer is empty.

Constructor

def __init__(
    self,
    maxlen: int,
    dtype: (
        type[np.float32]
        | type[np.float64]
        | type[np.int32]
        | type[np.int64]
        | type[np.uint32]
        | type[np.uint64]
    ) = np.float64,
) -> None:

Writing to the buffer

.write_append(value: float | int, timeout: float = -1.0)
- Accepts a single numeric value.
- Value is stored internally as the buffer’s dtype.
- timeout: Time in seconds to wait if the buffer is empty. Default is -1.0 (waits indefinitely). Use 0.0 to make it non-blocking.
.write_extend(data: Iterable[float | int], timeout: float = -1.0)
- data: Data block to write.
  - Accepts any Iterable of numeric values.
  - The iterable is converted to a contiguous NumPy array of the buffer’s dtype.
  - If you pass an np.ndarray, it is only copied if,
    - Its dtype does not match the buffer’s
    - It is not C-contiguous
- timeout: Same as .write_append
.write_extend_unchecked(block_np: np.ndarray, timeout: float = -1.0)
- block_np: np.ndarray to write.
  - Expects only a C-contiguous 1-D np.ndarray with the same dtype as the buffer.
  - Skips dtype, contiguous array, and dimension checks.
- timeout: Same as .write_append
Note: Using this yields the best performance, but if the array's dtype is different or if it is not contiguous, it will cause silent data corruption or crashes.

Reading from the buffer

.read(n: int = NumCircBuf.constants.Limits.SIZE_MAX.value, timeout: float = -1.0, partial_read: bool = True) -> np.ndarray

Reads items from the buffer.
- n: Number of items to read. Defaults to the maximum possible buffer size.
- timeout: Time in seconds to wait if the buffer is empty. Default is -1.0 (waits indefinitely). Use 0.0 to make it non-blocking.
- partial_read: If True (default), returns available items (up to n) immediately. If False, blocks until all n items are available.
.read_into(out_array_np: np.ndarray, timeout: float = -1.0, partial_read: bool = True) -> int

Reads directly into a pre-allocated np.ndarray of the same dtype as the buffer.
- Returns the number of items actually read.
- The number of items to read is determined by the length of out_array_np.
- timeout and partial_read: Same as .read().
.read_into_unchecked(out_array_np: np.ndarray, timeout: float = -1.0, partial_read: bool = True) -> int

Similar to .read_into
- Skips dtype, contiguous array, and dimension checks.
Note: Using this yields the best performance, but if the array's dtype is different or if it is not contiguous, it will cause silent data corruption or crashes.

Usage Example

import threading
import time
import numpy as np
from numcircbuf import BlockingCircBuffer

buf = BlockingCircBuffer(maxlen=1000, dtype=np.float32)
data = np.random.randn(1000).astype(np.float32)

# Pre-allocate a NumPy array with zeros for read_into.
# We use zeros instead of np.empty so that any unwritten elements are clearly visible
# — useful when inspecting random data.
read_into_arr = np.zeros(100, dtype=np.float32)


def producer():
    # Write an initial batch of 150 items
    buf.write_extend(data[:150])

    # Simulate a delay (e.g., waiting for network packets or sensor data)
    time.sleep(0.5)

    # Write the remaining 850 items
    buf.write_extend(data[150:])


def consumer():
    # We ask for 200 items, but the producer has only written 150 so far.
    # Because partial_read=True, it returns the 150 items immediately instead of waiting.
    received = buf.read(n=200, partial_read=True)
    print(f"1. Partial read: asked for 200, got {len(received)}")  # 150
    print(f"   Items left in buffer: {len(buf)}\n")  # 0

    # We ask for 300 items. The buffer is currently empty.
    # Because partial_read=False, this will block until the producer writes the next batch,
    # ensuring we get exactly 300 items.
    received = buf.read(n=300, partial_read=False)
    print(f"2. Strict read: asked for 300, got {len(received)}")  # 300
    print(
        f"   Items left in buffer: {len(buf)}\n"
    )  # 550 (850 new items - 300 read)

    # Reads directly into the pre-allocated array (size 100)
    received_count = buf.read_into(out_array_np=read_into_arr)
    received = read_into_arr[:received_count]
    print(f"3. Read into array: got {len(received)}")  # 100
    print(f"   Items left in buffer: {len(buf)}\n")  # 450

    # Reads everything left in the buffer
    received = buf.read()
    print(f"4. Read remaining: got {len(received)}")  # 450
    print(f"   Items left in buffer: {len(buf)}\n")  # 0

    # Non-blocking read: buffer is empty, timeout=0 makes it return immediately
    received = buf.read(timeout=0)
    print(f"5. Non-blocking read: got {len(received)}")  # 0

    # Non-blocking read_into: buffer is empty, returns immediately
    received_count = buf.read_into_unchecked(
        out_array_np=read_into_arr, timeout=0
    )
    print(f"6. Non-blocking read_into: got {received_count}")  # 0


# Start producer and consumer threads
producer_thread = threading.Thread(target=producer)
consumer_thread = threading.Thread(target=consumer)

producer_thread.start()
consumer_thread.start()

producer_thread.join()
consumer_thread.join()

Utility Buffers

Common Utility Buffer API

All utility buffer types expose the methods defined in Common Buffer API, and the following:

clear_cache(): Clears the cached metric.

Furthermore, all the buffers cache their metric value when the metric function is called and clear the cached metric value whenever the buffer is extended or appended to.

1. RunningMeanSqBuffer

Accumulator-capable circular buffer optimized for mean-square calculations.

Features fully vectorized operations, float-drift protection, and caching for mean-square.

Note: This buffer is not thread safe.

Concurrent reads during writes can return stale mean-square values
Concurrent writes can cause data corruption.

Constructor

def __init__(
    self,
    maxlen: int,
    operation_focus: Literal["extend/append", "calculation"],
    dtype: type[np.float32] | type[np.float64] = np.float64,
    recalc_threshold: int | None = 0,
) -> None:

Operation Focus

operation_focus: Literal["calculation", "extend/append"]

"calculation": O(1) statistics, higher per-write cost.
"extend/append": O(n) statistics, lower write cost.

Use the library utility determine_operation_focus to automatically select the best operation_focus. This function runs a small runtime benchmark and returns the appropriate Literal value for your use case:

def determine_operation_focus(
    buffer_type: type[RunningMeanSqBuffer] | type[RunningMeanBuffer],
    dtype: type[np.float32] | type[np.float64],
    buffer_maxlen: int,
    block_size: int, # Use 1 if you will be appending a single element only.
    calc_every: int, # Calculate every n blocks.
    verbose: bool = False, # Logs the exact relative multipliers,
                           # as well as total time spent in the function.
) -> Literal["calculation", "extend/append"]: # Outputs the best operation focus
                                              # for this (use case) + (system),
                                              # This can be directly passed to
                                              # the buffer init

Actual performance depends on buffer's maxlen, block size, overwrite rate, and statistics frequency. See PERFORMANCE.md for relative benchmarks.

Recalculation Threshold

recalc_threshold: int: Number of operations after which the accumulator is recalculated on all values in the buffer to reduce floating-point drift. Use 0 or None to disable.

Usage Example

import numpy as np
from numcircbuf import RunningMeanSqBuffer, determine_operation_focus

DTYPE = np.float32
CALC_EVERY = 2
BUFFER_MAXLEN = 1000
BLOCK_SIZE = 100

buffer = RunningMeanSqBuffer(
    maxlen=BUFFER_MAXLEN,
    operation_focus=determine_operation_focus(
        buffer_type=RunningMeanSqBuffer,
        dtype=DTYPE,
        buffer_maxlen=BUFFER_MAXLEN,
        block_size=BLOCK_SIZE,
        calc_every=CALC_EVERY,
        verbose=True,
    ),
    dtype=DTYPE,
)

# Add arrays/blocks
rng = np.random.default_rng(25)  # For random number generation
for i in range(10):  # Simulate time
    block = rng.random(BLOCK_SIZE, dtype=DTYPE)  # Mock block arriving
    buffer.extend(block)  # Extend with the block
    if (i + 1) % CALC_EVERY == 0:  # Calc every n
        print("-" * 10)
        print(buffer.mean_square())  # Get mean-square
        print(buffer.mean_square())  # Uses cached value

2. RunningMeanBuffer

Accumulator-capable circular buffer optimized for mean calculations.

Features fully vectorized operations, and caching for mean.

Note: This buffer is not thread safe.

Concurrent reads during writes can return stale mean values
Concurrent writes can cause data corruption.

Constructor

def __init__(
    self,
    maxlen: int,
    operation_focus: Literal["extend/append", "calculation"],
    dtype: type[np.float32] | type[np.float64] = np.float64,
    recalc_threshold: int | None = 0,
) -> None:

These work the same as in RunningMeanSqBuffer:

operation_focus
recalc_threshold

Usage Example

Works identically to RunningMeanSqBuffer — use .mean() instead of .mean_square()

3. IntegratedGatedBuffer

A specialized circular buffer for calculating gated loudness.

Features fully vectorized operations, and caching for gated mean-square.

Internal Storage: This buffer stores values representing signal power (the square of the amplitude). By default, input values are squared internally before storage. However, if already_squared is set to True during input, the values are stored as-is. Values retrieved via views will represent these squared values.

Note: This buffer is not thread safe.

Concurrent reads during writes can return stale gated mean-square values and inaccurate data
Concurrent writes can cause data corruption.

Constructor

def __init__(
    self,
    maxlen: int,
    abs_gate_lufs: float,
    rel_gate_lu: float,
    dtype: type[np.float32] | type[np.float64] = np.float64,
    recalc_threshold: int | None = 0,
) -> None:

recalc_threshold works the same as in RunningMeanSqBuffer.

Threshold Parameters

abs_gate_lufs: float The absolute loudness threshold in LUFS. Blocks with a mean-square power below this threshold are ignored during the first stage of the gating process.
rel_gate_lu: float The relative loudness threshold in LU. Blocks with a mean-square power more than this many decibels below the absolute-gated mean-square are ignored in the final integrated loudness calculation.

Usage Example

import numpy as np
from numcircbuf import IntegratedGatedBuffer

# Parameters for gated loudness calculation
# (We use the ITU-R BS.1770 standard constants for the example)
ABS_GATE_LUFS = -70.0
REL_GATE_LU = -10.0

buffer = IntegratedGatedBuffer(
    maxlen=3000,
    abs_gate_lufs=ABS_GATE_LUFS,
    rel_gate_lu=REL_GATE_LU
)

# Add the audio signal to the buffer.
# Input should be K-weighted and normalized to [-1.0, 1.0] as per ITU-R BS.1770.
block = np.random.uniform(-1.0, 1.0, 1000)
buffer.extend(block)

# Retrieve the final gated mean-square
gated_mean_sq = buffer.gated_mean_square()
print(f"Gated mean square: {gated_mean_sq}")

Exception Handling

Exceptions

NumCircBufError(Exception)
├── NumCircBufValueError(NumCircBufError, ValueError) ────────────────────────────┬─────┐
├── NumCircBufTypeError(NumCircBufError, TypeError) ─────────────────────┬─────┬─ │ ─┐  │
│   └── InvalidModification(NumCircBufTypeError)                         │     │  │  │  │
├── NumCircBufIndexError(NumCircBufError, IndexError)                    │     │  │  │  │
│   └── IndexOutOfBounds(NumCircBufIndexError)                           │     │  │  │  │
├── NumCircBufArithmeticError(NumCircBufError, ArithmeticError)          │     │  │  │  │
│   └── UnsupportedOperation(NumCircBufArithmeticError)                  │     │  │  │  │
├── NumCircBufRuntimeError(NumCircBufError, RuntimeError)                │     │  │  │  │
├── NumCircBufOSError(NumCircBufError, OSError)                          │     │  │  │  │
├── NumCircBufNotImplementedError(NumCircBufError, NotImplementedError)  │     │  │  │  │
└── NumCircBufInitError(NumCircBufError)                                 │     │  │  │  │
    ├── DataTypeError(NumCircBufInitError, NumCircBufTypeError) ─────────┘     │  │  │  │
    ├── BufferCapacityError(NumCircBufInitError)                               │  │  │  │
    │   ├── BufferCapacityTypeError(BufferCapacityError, NumCircBufTypeError) ─┘  │  │  │
    │   └── BufferCapacityValueError(BufferCapacityError, NumCircBufValueError) ──┘  │  │
    └── ConfigurationError(NumCircBufInitError)                                      │  │
        ├── ConfigurationTypeError(ConfigurationError, NumCircBufTypeError) ─────────┘  │
        └── ConfigurationValueError(ConfigurationError, NumCircBufValueError) ──────────┘

NumCircBuf exceptions are comprehensive and include context (e.g., the object/class that caused the error):

class_obj – the class where the exception occurred (may be None if no class is associated with the error).

obj – the instance that caused the error (may be None if instantiation failed or no specific instance is involved).

Various different attributes available based on the exception type.

Performance note: Some low-level Cython/NumPy errors may still propagate as native Python exceptions (ValueError, TypeError, etc.) to avoid redundant checks in performance-critical code.

Usage Example

try:
    buf.extend(data)
except exceptions.NumCircBufError as e:
    # Structured library errors; you can access `e.class_obj` and `e.obj`
    handle_numcircbuf_error(e)
except Exception as e:
    # Low-level errors from NumPy/Cython
    handle_low_level_error(e)

Warnings

NumCircBufWarning(Warning)
├── NumCircBufDeprecationWarning(NumCircBufWarning, DeprecationWarning)
├── NumCircBufFutureWarning(NumCircBufWarning, FutureWarning)
└── NumCircBufRuntimeWarning(NumCircBufWarning, RuntimeWarning)
    └── DataSizeWarning(NumCircBufRuntimeWarning)

All NumCircBuf warnings include class_obj and obj attributes (like exceptions).

Usage Example

import warnings
from numcircbuf import OverwriteCircBuffer, exceptions

buffer = OverwriteCircBuffer(5, "never")


# Example function that catches a NumCircBufWarning
def extend_and_catch(buffer, data):
    with warnings.catch_warnings(record=True) as caught_warnings:
        warnings.simplefilter("always")  # Capture all warnings
        buffer.extend(data)
        for w in caught_warnings:
            if isinstance(w.message, exceptions.NumCircBufWarning):
                print("Warning type:", type(w.message))
                print("Class of buffer:", w.message.class_obj)
                print("Buffer instance:", w.message.obj)
                print("Full message:", w.message)


# Trigger a warning
extend_and_catch(buffer, range(10))

Performance

NumCircBuf provides a suite of pre-allocated, contiguous-memory circular buffers engineered for low-latency ingestion and O(1) windowed analytics.

Benchmark Results

For detailed performance benchmarks, see the PERFORMANCE.md document which includes testing results across different buffer types and use cases.

Key Performance Highlights:

Throughput (extending 64 KiB of data to OverwriteCircBuffer):
- Cold cache (data):
  - AMD R7700x (DDR5): ~35 GB/s
  - AMD R5600 (DDR4): ~30 GB/s
- Warm cache (data):
  - AMD R7700x (DDR5): ~73 GB/s
  - AMD R5600 (DDR4): ~50 GB/s

Documentation

Performance Benchmarks: Benchmark results and optimization guide.
Versioning Strategy: Versioning policy, release process, and compatibility guarantees.

Changelog

See CHANGELOG.md for a detailed history of changes, including new features, bug fixes, and performance improvements.

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

Support

For issues, questions, or feature requests:

GitHub Issues: Open an issue on our GitHub repository
GitHub Discussions: Join the conversation in our Discussions forum

Citation

If you use NumCircBuf in your research or projects, please cite it as:

@software{NumCircBuf,
  author = {Syed Basim Ali},
  title = {NumCircBuf: High-Performance Numerical Circular Buffers for Python},
  year = {2026},
  url = {https://github.com/basimali-ai/NumCircBuf},
  version = {1.2.0}
}

Acknowledgements

NumCircBuf is built on top of these technologies:

Cython for performance optimization
NumPy for numerical computing and integration

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

numcircbuf-1.2.0.tar.gz (268.0 kB view details)

Uploaded Jul 31, 2026 Source

Built Distributions

If you're not sure about the file name format, learn more about wheel file names.

The dropdown lists show the available interpreters, ABIs, and platforms. Enable javascript to be able to filter the list of wheel files.

numcircbuf-1.2.0-cp314-cp314t-win_amd64.whl (191.4 kB view details)

Uploaded Jul 31, 2026 CPython 3.14tWindows x86-64

numcircbuf-1.2.0-cp314-cp314t-musllinux_1_2_x86_64.whl (1.2 MB view details)

Uploaded Jul 31, 2026 CPython 3.14tmusllinux: musl 1.2+ x86-64

numcircbuf-1.2.0-cp314-cp314t-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl (212.8 kB view details)

Uploaded Jul 31, 2026 CPython 3.14tmanylinux: glibc 2.24+ x86-64manylinux: glibc 2.28+ x86-64

numcircbuf-1.2.0-cp314-cp314t-macosx_11_0_arm64.whl (206.4 kB view details)

Uploaded Jul 31, 2026 CPython 3.14tmacOS 11.0+ ARM64

numcircbuf-1.2.0-cp314-cp314-win_amd64.whl (179.3 kB view details)

Uploaded Jul 31, 2026 CPython 3.14Windows x86-64

numcircbuf-1.2.0-cp314-cp314-musllinux_1_2_x86_64.whl (1.2 MB view details)

Uploaded Jul 31, 2026 CPython 3.14musllinux: musl 1.2+ x86-64

numcircbuf-1.2.0-cp314-cp314-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl (216.3 kB view details)

Uploaded Jul 31, 2026 CPython 3.14manylinux: glibc 2.24+ x86-64manylinux: glibc 2.28+ x86-64

numcircbuf-1.2.0-cp314-cp314-macosx_11_0_arm64.whl (193.8 kB view details)

Uploaded Jul 31, 2026 CPython 3.14macOS 11.0+ ARM64

numcircbuf-1.2.0-cp313-cp313-win_amd64.whl (177.8 kB view details)

Uploaded Jul 31, 2026 CPython 3.13Windows x86-64

numcircbuf-1.2.0-cp313-cp313-musllinux_1_2_x86_64.whl (1.2 MB view details)

Uploaded Jul 31, 2026 CPython 3.13musllinux: musl 1.2+ x86-64

numcircbuf-1.2.0-cp313-cp313-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl (215.6 kB view details)

Uploaded Jul 31, 2026 CPython 3.13manylinux: glibc 2.24+ x86-64manylinux: glibc 2.28+ x86-64

numcircbuf-1.2.0-cp313-cp313-macosx_11_0_arm64.whl (193.7 kB view details)

Uploaded Jul 31, 2026 CPython 3.13macOS 11.0+ ARM64

numcircbuf-1.2.0-cp312-cp312-win_amd64.whl (178.4 kB view details)

Uploaded Jul 31, 2026 CPython 3.12Windows x86-64

numcircbuf-1.2.0-cp312-cp312-musllinux_1_2_x86_64.whl (1.2 MB view details)

Uploaded Jul 31, 2026 CPython 3.12musllinux: musl 1.2+ x86-64

numcircbuf-1.2.0-cp312-cp312-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl (217.3 kB view details)

Uploaded Jul 31, 2026 CPython 3.12manylinux: glibc 2.24+ x86-64manylinux: glibc 2.28+ x86-64

numcircbuf-1.2.0-cp312-cp312-macosx_11_0_arm64.whl (193.7 kB view details)

Uploaded Jul 31, 2026 CPython 3.12macOS 11.0+ ARM64

numcircbuf-1.2.0-cp311-cp311-win_amd64.whl (185.6 kB view details)

Uploaded Jul 31, 2026 CPython 3.11Windows x86-64

numcircbuf-1.2.0-cp311-cp311-musllinux_1_2_x86_64.whl (1.3 MB view details)

Uploaded Jul 31, 2026 CPython 3.11musllinux: musl 1.2+ x86-64

numcircbuf-1.2.0-cp311-cp311-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl (227.9 kB view details)

Uploaded Jul 31, 2026 CPython 3.11manylinux: glibc 2.24+ x86-64manylinux: glibc 2.28+ x86-64

numcircbuf-1.2.0-cp311-cp311-macosx_11_0_arm64.whl (195.3 kB view details)

Uploaded Jul 31, 2026 CPython 3.11macOS 11.0+ ARM64

numcircbuf-1.2.0-cp310-cp310-win_amd64.whl (185.5 kB view details)

Uploaded Jul 31, 2026 CPython 3.10Windows x86-64

numcircbuf-1.2.0-cp310-cp310-musllinux_1_2_x86_64.whl (1.3 MB view details)

Uploaded Jul 31, 2026 CPython 3.10musllinux: musl 1.2+ x86-64

numcircbuf-1.2.0-cp310-cp310-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl (230.1 kB view details)

Uploaded Jul 31, 2026 CPython 3.10manylinux: glibc 2.24+ x86-64manylinux: glibc 2.28+ x86-64

numcircbuf-1.2.0-cp310-cp310-macosx_11_0_arm64.whl (197.0 kB view details)

Uploaded Jul 31, 2026 CPython 3.10macOS 11.0+ ARM64

numcircbuf-1.2.0-cp39-cp39-win_amd64.whl (186.0 kB view details)

Uploaded Jul 31, 2026 CPython 3.9Windows x86-64

numcircbuf-1.2.0-cp39-cp39-musllinux_1_2_x86_64.whl (1.3 MB view details)

Uploaded Jul 31, 2026 CPython 3.9musllinux: musl 1.2+ x86-64

numcircbuf-1.2.0-cp39-cp39-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl (230.3 kB view details)

Uploaded Jul 31, 2026 CPython 3.9manylinux: glibc 2.24+ x86-64manylinux: glibc 2.28+ x86-64

numcircbuf-1.2.0-cp39-cp39-macosx_11_0_arm64.whl (197.8 kB view details)

Uploaded Jul 31, 2026 CPython 3.9macOS 11.0+ ARM64

File details

Details for the file numcircbuf-1.2.0.tar.gz.

File metadata

Download URL: numcircbuf-1.2.0.tar.gz
Upload date: Jul 31, 2026
Size: 268.0 kB
Tags: Source
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for numcircbuf-1.2.0.tar.gz
Algorithm	Hash digest
SHA256	`4eb7b4521b9ce02b0f394de51b4df2370d5031f3a6c57864382bd24417108d8e`
MD5	`ac1c63fd47307fc71d69d50e45061034`
BLAKE2b-256	`f69c9b443dbcf7c9adc3a15b12f3338e9301837aa6359633c60d6294c2bb0d59`

Algorithm	Hash digest
SHA256	`6a5f9ae39d9ddbab95a41add7a7f52e28677928f01ff2f83fcc44a973634f9a2`
MD5	`a7e35375b2f1ec4ab5acd960240ee3e2`
BLAKE2b-256	`7094f39174721f70171187e3cce86454ab7a4baa32c7de2e4d3d21d9050cef8a`

Algorithm	Hash digest
SHA256	`ef0f09aa36078650ea61896a22860a2c0f7b33dbfbbbfe68cd7e5944484070fd`
MD5	`e899e984e82365818c565ab544ac0ac0`
BLAKE2b-256	`c94c72ed514af7c4da9c868cd471fb6e62c3b3ad984a6b7310a743ff0491c51f`

Algorithm	Hash digest
SHA256	`d333d14cf7a340fbc1d08e8a6fcb50e63150303770af4723c9eea3522c7e77de`
MD5	`b3173e1ea130d251ac5cec4c12f9e1bb`
BLAKE2b-256	`fe46c25cab115c2cf096e90cee2e71edc69ee28623bd8786dcd7ac445ea08fb9`

Algorithm	Hash digest
SHA256	`380fb7a00eedb92bc06eb843d665f874422532e59b6b95bef36136b5e8ad9056`
MD5	`ccc6b869208690cd50d7a34a6c046576`
BLAKE2b-256	`89ededc8e9614dc30a12ecf0de07c6b75b9d329dfde3d35b0eba691ac37682a1`

Algorithm	Hash digest
SHA256	`a3866cf38d6667d5e66835a808a9ff6098045419f2f3146ff29e3adacc50498f`
MD5	`0a9e72c98046b88451b58f73fc8db692`
BLAKE2b-256	`201713617327a0b186b6cc42ea87e49198bc44c2bdb0d8bd84cd7679478b6cc8`

Algorithm	Hash digest
SHA256	`802df75732d673e7921f3a3f466384b1384b99b659f01f95c0cd8fa4fe3ed1bc`
MD5	`65247c48f59d9cad7cd5b1611f27d168`
BLAKE2b-256	`c97c9063417e60e1395fbd6d9aff90754526aa521909552f508b0ece0da47dde`

Algorithm	Hash digest
SHA256	`3ba7f30af372b5af98f7842b08f601ac1844e0f2f9714d59977155e0ddcc4169`
MD5	`2969a847a2a2f37892e68362e3dcad0d`
BLAKE2b-256	`2f8389d28bcd1cc878a4dc237deff120ea9e1b20aae59422cc3b153f0a652ae7`

Algorithm	Hash digest
SHA256	`94985471c84b58e231f30d7dbfc0d9fbd9aa969cbb39b5fa84fad1826ca20fc6`
MD5	`d5f0deb5773f1af431546b2cbf29411f`
BLAKE2b-256	`2e1ad1eabd7ee960e4427529895d4d671816dfb6fce0208885339f81ecd6c7d4`

Algorithm	Hash digest
SHA256	`3fe287898e391fe8d9d70c5737598de0e5cf58e9041a03f388774d9a9dcbf222`
MD5	`9ebc38ffd31f9b98dbebab12a7610e3a`
BLAKE2b-256	`7dd2b82815a3ab6b40747d47c0f7991f649289e398777d7fc581e4e4214d24b8`

Algorithm	Hash digest
SHA256	`8506740df5c731373abf5d0e2e1b3edeb37d8e273aec5ace4046857248514752`
MD5	`366de47b87ae7022aebd6da426540c71`
BLAKE2b-256	`b67f93fa37146ea69b35b9ec125215a04c0e089339a720c93fc4c03f69b40313`

Algorithm	Hash digest
SHA256	`f2eaee0298055af0ebb748f61199d35f671e0bf4d03205079d4f14fab86590d6`
MD5	`3807da40925a9966c34b3783b103e684`
BLAKE2b-256	`8c94cb6a003c56c6832f8528940f2dd3b766034456807f40f7a0a4070b701308`

Algorithm	Hash digest
SHA256	`8aeb32a32ea5cb108ee54cac953c53deb3e996e412a5f72555514003c2758f6c`
MD5	`066f511761da9d443fd1314c2b954792`
BLAKE2b-256	`94be2f7ab6b21ff331a848de8dfef862a28fc797b17411c74a02988ce335e5bd`

Algorithm	Hash digest
SHA256	`08131021a6b0323a517970de2d700b887cade6611c1e6e7783e1bc5540dbe164`
MD5	`cffbb9c0d34d1f89dd5dbdd392ab7976`
BLAKE2b-256	`2b00a33aeb47b352295cd01da946108e2a4f9df2ba32352365804ab7030efe73`

Algorithm	Hash digest
SHA256	`ec208215d546e6ab0b5a899aae7a543b2b590498e76bf77c5b132f9951dab269`
MD5	`d36b928268f21c8780584c75c7e65672`
BLAKE2b-256	`c89f896610c71f29fbc1b5dd13a88f4b938774af17c2596fa2087abaf71a8439`

Algorithm	Hash digest
SHA256	`23f552c601528b0afd371dab9b3d94fcf8539174e69fd72297e08883f336d8ae`
MD5	`77a87bda462e9f34f7c9481d3803ddb6`
BLAKE2b-256	`97909f72cd3cf0f367502c06773e5ab151cdf9b700f043fb57a48b9346664102`

Algorithm	Hash digest
SHA256	`bdfef465680cdfc2c25eb688fbcd0a77b5864ef366f2c08a406819dfd947793d`
MD5	`0baef6fab168f555f5caf0f1f6a5916d`
BLAKE2b-256	`d925791484afe1f62d98891b6881768e13c28e5e766ea0d8c3128fd24fe3b63d`

Algorithm	Hash digest
SHA256	`c0b00d6ad664118f46c81ad2491e4c0cc78c0fdecfdc9b74c8ed559c0e9f815b`
MD5	`17825bf93951d0d52e84f0a3bcf5a7a3`
BLAKE2b-256	`9e5791a23f0fe5ffca7fbf9f897cf26bc8243a6f3bd5126d1ef3c9681748adef`

Algorithm	Hash digest
SHA256	`ce5f34634bde5addf678cfc6e7d67ca79c69b7abb195c48f151dcf56c37b1871`
MD5	`f003d9a09fbda37535e04b088329236c`
BLAKE2b-256	`af8bffabc2aac579f5ddf85efd350da9d77503420344a01afd0267b00d128880`

Algorithm	Hash digest
SHA256	`b2d777f6eb04f1823106f22f0d3f5156e41b13f7d7b6090b4b25f0702d5d2e23`
MD5	`0e50335bfbd60026dc05a818f1e07d61`
BLAKE2b-256	`c00fd37b0380c98b931fedc60b24894a4f840e8cb0c80ecf3f7f3ca3f4829d6e`

Algorithm	Hash digest
SHA256	`cbd2ba889c451ba3cc8436aa9d4ec4e4c1ceef9fa987ac57165687489843dede`
MD5	`14f7bd9296f96c10dca84eb0384e8a0e`
BLAKE2b-256	`edc6a4e26a5f940e7740e296eeab56f7a2e3f5d7121c9ffc4e4549462f24c6e8`

NumCircBuf 1.2.0

NumCircBuf: High-performance numerical circular buffers for Python, featuring O(1) statistical accumulators.

Table of Contents

Overview

Features

Installation

Verification

Common Buffer API

The Buffer View (ViewProtocol)

Usage Example

Main Buffers

1. OverwriteCircBuffer

Constructor

Overwrite Return Policy

Supported input types

Mathematical Metrics

2. BlockingCircBuffer

Constructor

Writing to the buffer

Reading from the buffer

Usage Example

Utility Buffers

Common Utility Buffer API

1. RunningMeanSqBuffer

Constructor

Operation Focus

Recalculation Threshold

Usage Example

2. RunningMeanBuffer

Constructor

Usage Example

3. IntegratedGatedBuffer

Constructor

Threshold Parameters

Usage Example

Exception Handling

Exceptions

Usage Example

Warnings

Usage Example

Performance

Benchmark Results

Documentation

Changelog

License

Support

Citation

Acknowledgements

Project links

Key dates

1 maintainer

Credits

GitHub Statistics

License

Requires

Provides Extra

Tags

Classifiers

Download files

Source Distribution

Built Distributions

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Algorithm	Hash digest
SHA256	`ed9c283bd46c139f950c98d321782a04669c3dbac8ed1e061a90e9d9626a10a3`
MD5	`19bcdf3ec605ae1710afd86527763eab`
BLAKE2b-256	`f8609f204d36a012dd034344d74eb1b009eeaf97cb0f8385ff99e1b48a51bc63`

Algorithm	Hash digest
SHA256	`da1dc4037331a371bd479cf4c2b4d8fec7aa2df6f6a559d829fe0f2fea21aac6`
MD5	`45eea8c4b8093c9c5af6d599fdb458f1`
BLAKE2b-256	`4618d1497c35bd3abdf927d6d73f6b0ffed8ddfcfcfe370ec8cd6433ad69397d`

Algorithm	Hash digest
SHA256	`72da3de5a5172313169d7fd8e11debd3bb70bc0da9f36ca1f2a0c95a83fe6b80`
MD5	`f75bbc5d8df48da5a7953c9ccdba6ec1`
BLAKE2b-256	`6bf446e3c2e1f3a29a4849985bef01d33721cef469398d43d208d47ec3841a21`

Algorithm	Hash digest
SHA256	`95d736de6d11bd0afde578fade167ae16814ee5d990368ad28871106a979b430`
MD5	`c4754e1883a15cfe93818ada02220bf5`
BLAKE2b-256	`b499ac8f55aa4c3695edb94b5c403989a077f080540ebd5be217695250acca76`

Algorithm	Hash digest
SHA256	`81bd139cf596a72039238f8a12d23efa3a6c3406c78feb2039b48633aee13ba6`
MD5	`58fc5a5810b466cd960b8847351005c0`
BLAKE2b-256	`2cdf72a807d522b1dccdd9c7db2e6461c50a0c3f9c01617261ee532b58272d6e`

Algorithm	Hash digest
SHA256	`dab4ad597dfd6f76d7892d18b8a6e7529dc83a40cee67d988135f7798bf6da39`
MD5	`29f7ebce71b6fee460f6301582b7cca0`
BLAKE2b-256	`9192050a8054ee2cb77b09c4653c226a003a681fab113ed19b7229ae04415371`

Algorithm	Hash digest
SHA256	`2063417934c1f55c2db00717cfe244a8ccd4f7a1b01894bc9c2a1cc073b9a924`
MD5	`4118eacb671c2bad1bbd0956b96b1279`
BLAKE2b-256	`b37813c091e5d372c23202792b77fa99d2020f3555daa7490df917894ce43ecb`

Algorithm	Hash digest
SHA256	`956ec2dbb45d4b79f311dae7d3e6a22926902eb0eb5fad9610c1206065da2042`
MD5	`1e65ef6d606b8c7d1241865b4dad6957`
BLAKE2b-256	`e8caba18877c5f44271ba071d1eb2849a46a98ca0efea727a821f40c8e4d736f`