# NumPack
NumPack is a lightning-fast array manipulation engine that revolutionizes how you handle large-scale NumPy arrays. By combining Rust's raw performance with Python's ease of use, NumPack delivers up to 20x faster operations than traditional methods, while using minimal memory. Whether you're working with gigabyte-sized matrices or performing millions of array operations, NumPack makes it effortless with its zero-copy architecture and intelligent memory management.
Key highlights:
- 🚀 Up to 20x faster than traditional NumPy storage methods
- 💾 Zero-copy operations for minimal memory footprint
- 🔄 Seamless integration with existing NumPy workflows
- 🛠Battle-tested in production with arrays exceeding 1 billion rows
## Features
- **High Performance**: Optimized for both reading and writing large numerical arrays
- **Memory Mapping Support**: Efficient memory usage through memory mapping capabilities
- **Selective Loading**: Load only the arrays you need, when you need them
- **In-place Operations**: Support for in-place array modifications without full file rewrite
- **Parallel I/O**: Utilizes parallel processing for improved performance
- **Multiple Data Types**: Supports various numerical data types including:
- Boolean
- Unsigned integers (8-bit to 64-bit)
- Signed integers (8-bit to 64-bit)
- Floating point (32-bit and 64-bit)
## Installation
### From PyPI (Recommended)
#### Prerequisites
- Python >= 3.9
- NumPy >= 1.26.0
```bash
pip install numpack
```
### From Source
To build and install NumPack from source, you need to meet the following requirements:
#### Prerequisites
- Python >= 3.9
- Rust >= 1.70.0
- NumPy >= 1.26.0
- Appropriate C/C++ compiler (depending on your operating system)
- Linux: GCC or Clang
- macOS: Clang (via Xcode Command Line Tools)
- Windows: MSVC (via Visual Studio or Build Tools)
#### Build Steps
1. Clone the repository:
```bash
git clone https://github.com/BirchKwok/NumPack.git
cd NumPack
```
2. Install maturin (for building Rust and Python hybrid projects):
```bash
pip install maturin>=1.0,<2.0
```
3. Build and install:
```bash
# Install in development mode
maturin develop
# Or build wheel package
maturin build --release
pip install target/wheels/numpack-*.whl
```
#### Platform-Specific Notes
- **Linux Users**:
- Ensure python3-dev (Ubuntu/Debian) or python3-devel (Fedora/RHEL) is installed
- If using conda environment, make sure the appropriate compiler toolchain is installed
- **macOS Users**:
- Make sure Xcode Command Line Tools are installed: `xcode-select --install`
- Supports both Intel and Apple Silicon architectures
- **Windows Users**:
- Visual Studio or Visual Studio Build Tools required
- Ensure "Desktop development with C++" workload is installed
## Usage
### Basic Operations
```python
import numpy as np
from numpack import NumPack
# Create a NumPack instance
npk = NumPack("data_directory")
# Save arrays
arrays = {
'array1': np.random.rand(1000, 100).astype(np.float32),
'array2': np.random.rand(500, 200).astype(np.float32)
}
npk.save(arrays)
# Load arrays
# Normal mode
loaded = npk.load("array1")
# Memory mapping mode for large arrays
with npk.mmap_mode() as mmap_npk:
# Access specific arrays
array1 = mmap_npk.load('array1')
array2 = mmap_npk.load('array2')
```
### Advanced Operations
```python
# Replace specific rows
replacement = np.random.rand(10, 100).astype(np.float32)
npk.replace({'array1': replacement}, [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]) # Using list indices
npk.replace({'array1': replacement}, slice(0, 10)) # Using slice notation
# Append new arrays
new_arrays = {
'array3': np.random.rand(200, 100).astype(np.float32)
}
npk.append(new_arrays)
# Drop arrays or specific rows
npk.drop('array1') # Drop entire array
npk.drop(['array1', 'array2']) # Drop multiple arrays
npk.drop('array2', [0, 1, 2]) # Drop specific rows
# Random access operations
data = npk.getitem('array1', [0, 1, 2]) # Access specific rows
data = npk.getitem('array1', slice(0, 10)) # Access using slice
data = npk['array1'] # Dictionary-style access for entire array
# Metadata operations
shapes = npk.get_shape() # Get shapes of all arrays
shapes = npk.get_shape('array1') # Get shape of specific array
members = npk.get_member_list() # Get list of array names
mtime = npk.get_modify_time('array1') # Get modification time
metadata = npk.get_metadata() # Get complete metadata
# Stream loading for large arrays
for batch in npk.stream_load('array1', buffer_size=1000):
# Process 1000 rows at a time
process_batch(batch)
# Reset/clear storage
npk.reset() # Clear all arrays
# Iterate over all arrays
for array_name in npk:
data = npk[array_name]
print(f"{array_name} shape: {data.shape}")
```
### Lazy Loading and Buffer Operations
NumPack supports lazy loading and buffer operations, which are particularly useful for handling large-scale datasets. Using the `lazy=True` parameter enables data to be loaded only when actually needed, making it ideal for streaming processing or scenarios where only partial data access is required.
```python
from numpack import NumPack
import numpy as np
# Create NumPack instance and save large-scale data
npk = NumPack("test_data/", drop_if_exists=True)
a = np.random.random((1000000, 128)) # Create a large array
npk.save({"arr1": a})
# Lazy loading - keeps data in buffer
lazy_array = npk.load("arr1", lazy=True) # LazyArray Object
# Perform computations with lazy-loaded data
# Only required data is loaded into memory
similarity_scores = np.inner(a[0], npk.load("arr1", lazy=True))
```
### Memory Mapping Mode
For large arrays, memory mapping mode provides more efficient memory usage:
```python
# Using memory mapping mode
with npk.mmap_mode() as mmap_npk:
# Access specific arrays
array1 = mmap_npk.load('array1') # Array is not fully loaded into memory
array2 = mmap_npk.load('array2')
# Perform operations on memory-mapped arrays
result = array1[0:1000] + array2[0:1000]
```
## Performance
NumPack offers significant performance improvements compared to traditional NumPy storage methods, especially in data modification operations and random access. Below are detailed benchmark results:
### Benchmark Results
The following benchmarks were performed on an MacBook Pro (Apple Silicon) with arrays of size 1M x 10 and 500K x 5 (float32).
#### Storage Operations
| Operation | NumPack | NumPy NPZ | NumPy NPY |
|-----------|---------|-----------|-----------|
| Save | 0.014s (0.86x NPZ, 0.57x NPY) | 0.012s | 0.008s |
| Full Load | 0.008s (1.63x NPZ, 0.88x NPY) | 0.013s | 0.007s |
| Selective Load | 0.006s (1.50x NPZ, -) | 0.009s | - |
| Mmap Load | 0.006s (2.00x NPZ, 0.67x NPY) | 0.012s | 0.004s |
#### Data Modification Operations
| Operation | NumPack | NumPy NPZ | NumPy NPY |
|-----------|---------|-----------|-----------|
| Single Row Replace | 0.000s (19.00x NPZ, 12.00x NPY) | 0.019s | 0.012s |
| Continuous Rows (10K) | 0.001s (20.00x NPZ, 12.00x NPY) | 0.020s | 0.012s |
| Random Rows (10K) | 0.015s (1.33x NPZ, 0.87x NPY) | 0.020s | 0.013s |
| Large Data Replace (500K) | 0.019s (1.00x NPZ, 0.74x NPY) | 0.019s | 0.014s |
#### Drop Operations
| Operation (1M rows, float32) | NumPack | NumPy NPZ | NumPy NPY |
|-----------|---------|-----------|-----------|
| Drop Array | 0.001s (22.00x NPZ, 1.00x NPY) | 0.022s | 0.001s |
| Drop First Row | 0.014s (3.21x NPZ, 1.93x NPY) | 0.045s | 0.027s |
| Drop Last Row | 0.000s (∞x NPZ, ∞x NPY) | 0.045s | 0.027s |
| Drop Middle Row | 0.014s (3.21x NPZ, 1.93x NPY) | 0.045s | 0.027s |
| Drop Front Continuous (10K rows) | 0.016s (2.81x NPZ, 1.69x NPY) | 0.045s | 0.027s |
| Drop Middle Continuous (10K rows) | 0.016s (2.81x NPZ, 1.69x NPY) | 0.045s | 0.027s |
| Drop End Continuous (10K rows) | 0.001s (45.00x NPZ, 27.00x NPY) | 0.045s | 0.027s |
| Drop Random Rows (10K rows) | 0.018s (2.50x NPZ, 1.50x NPY) | 0.045s | 0.027s |
| Drop Near Non-continuous (10K rows) | 0.015s (3.00x NPZ, 1.80x NPY) | 0.045s | 0.027s |
#### Append Operations
| Operation | NumPack | NumPy NPZ | NumPy NPY |
|-----------|---------|-----------|-----------|
| Small Append (1K rows) | 0.000s (22.00x NPZ, 18.00x NPY) | 0.022s | 0.018s |
| Large Append (500K rows) | 0.003s (9.67x NPZ, 6.67x NPY) | 0.029s | 0.020s |
#### Random Access Performance (10K indices)
| Operation | NumPack | NumPy NPZ | NumPy NPY |
|-----------|---------|-----------|-----------|
| Random Access | 0.008s (2.00x NPZ, 1.38x NPY) | 0.016s | 0.011s |
#### File Size Comparison
| Format | Size | Ratio |
|--------|------|-------|
| NumPack | 47.68 MB | 1.0x |
| NPZ | 47.68 MB | 1.0x |
| NPY | 47.68 MB | 1.0x |
#### Large-scale Data Operations (>1B rows, Float32)
| Operation | NumPack | NumPy NPZ | NumPy NPY |
|-----------|---------|-----------|-----------|
| Replace | Zero-copy in-place modification | Memory exceeded | Memory exceeded |
| Drop | Zero-copy in-place deletion | Memory exceeded | Memory exceeded |
| Append | Zero-copy in-place addition | Memory exceeded | Memory exceeded |
| Random Access | Near-hardware I/O speed | Memory exceeded | Memory exceeded |
#### Matrix Computation Performance (1M rows x 128 columns, Float32)
| Operation | NumPack | NumPy NPZ | NumPy NPY | In-Memory |
|-----------|---------|-----------|-----------|-----------|
| Inner Product | 0.019s (6.58x NPZ, 1.00x NPY) | 0.125s | 0.019s | 0.011s |
| Other calculations are similar to the above case | ... | ... | ... | ... |
> **Key Advantage**: NumPack achieves the same performance as NumPy's NPY mmap (0.019s) for matrix computations, with several implementation advantages:
> - Uses Arc<Mmap> for reference counting, ensuring automatic resource cleanup
> - Implements MMAP_CACHE to avoid redundant data loading
> - Linux-specific optimizations with huge pages and sequential access hints
> - Supports parallel I/O operations for improved data throughput
> - Optimizes memory usage through Buffer Pool to reduce fragmentation
### Key Performance Highlights
1. **Data Modification**:
- Single row replacement: NumPack is **19x faster** than NPZ and **12x faster** than NPY
- Continuous rows: NumPack is **20x faster** than NPZ and **12x faster** than NPY
- Random rows: NumPack is **1.33x faster** than NPZ but **0.87x slower** than NPY
- Large data replacement: NumPack is comparable to NPZ but **0.74x slower** than NPY
2. **Drop Operations**:
- Drop array: NumPack is **22x faster** than NPZ and comparable to NPY
- Drop rows: NumPack is currently **0.61x slower** than NPZ and **0.41x slower** than NPY
- NumPack provides efficient in-place row deletion without full file rewrite
3. **Append Operations**:
- Small append (1K rows): NumPack is **22x faster** than NPZ and **18x faster** than NPY
- Large append (500K rows): NumPack is **9.67x faster** than NPZ and **6.67x faster** than NPY
- NumPack excels at both small and large append operations
4. **Loading Performance**:
- Full load: NumPack is **1.63x faster** than NPZ but **0.88x slower** than NPY
- Memory-mapped load: NumPack is **2.00x faster** than NPZ but **0.67x slower** than NPY
- Selective load: NumPack is **1.50x faster** than NPZ
5. **Random Access**:
- NumPack is **2.00x faster** than NPZ and **1.38x faster** than NPY for random index access
6. **Storage Efficiency**:
- All formats achieve identical compression ratios (47.68 MB)
- NumPack maintains high performance while keeping file sizes competitive
7. **Matrix Computation**:
- NumPack matches NPY mmap performance while providing better resource management
- **6.58x faster** than NPZ mmap for matrix operations
- Only 1.72x slower than pure in-memory computation
- Zero risk of file descriptor leaks or resource exhaustion
> Note: All benchmarks were performed with float32 arrays. Performance may vary depending on data types, array sizes, and system configurations. Numbers greater than 1.0x indicate faster performance, while numbers less than 1.0x indicate slower performance.
## Contributing
Contributions are welcome! Please feel free to submit a Pull Request.
## License
This project is licensed under the Apache License, Version 2.0 - see the LICENSE file for details.
Copyright 2024 NumPack Contributors
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
Raw data
{
"_id": null,
"home_page": null,
"name": "numpack",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.9",
"maintainer_email": null,
"keywords": "numpy, array, storage, performance",
"author": "NumPack Contributors",
"author_email": null,
"download_url": null,
"platform": null,
"description": "# NumPack\n\nNumPack is a lightning-fast array manipulation engine that revolutionizes how you handle large-scale NumPy arrays. By combining Rust's raw performance with Python's ease of use, NumPack delivers up to 20x faster operations than traditional methods, while using minimal memory. Whether you're working with gigabyte-sized matrices or performing millions of array operations, NumPack makes it effortless with its zero-copy architecture and intelligent memory management.\n\nKey highlights:\n- \ud83d\ude80 Up to 20x faster than traditional NumPy storage methods\n- \ud83d\udcbe Zero-copy operations for minimal memory footprint\n- \ud83d\udd04 Seamless integration with existing NumPy workflows\n- \ud83d\udee0 Battle-tested in production with arrays exceeding 1 billion rows\n\n## Features\n\n- **High Performance**: Optimized for both reading and writing large numerical arrays\n- **Memory Mapping Support**: Efficient memory usage through memory mapping capabilities\n- **Selective Loading**: Load only the arrays you need, when you need them\n- **In-place Operations**: Support for in-place array modifications without full file rewrite\n- **Parallel I/O**: Utilizes parallel processing for improved performance\n- **Multiple Data Types**: Supports various numerical data types including:\n - Boolean\n - Unsigned integers (8-bit to 64-bit)\n - Signed integers (8-bit to 64-bit)\n - Floating point (32-bit and 64-bit)\n\n## Installation\n\n### From PyPI (Recommended)\n\n#### Prerequisites\n- Python >= 3.9\n- NumPy >= 1.26.0\n\n```bash\npip install numpack\n```\n\n### From Source\n\nTo build and install NumPack from source, you need to meet the following requirements:\n\n#### Prerequisites\n\n- Python >= 3.9\n- Rust >= 1.70.0\n- NumPy >= 1.26.0\n- Appropriate C/C++ compiler (depending on your operating system)\n - Linux: GCC or Clang\n - macOS: Clang (via Xcode Command Line Tools)\n - Windows: MSVC (via Visual Studio or Build Tools)\n\n#### Build Steps\n\n1. Clone the repository:\n```bash\ngit clone https://github.com/BirchKwok/NumPack.git\ncd NumPack\n```\n\n2. Install maturin (for building Rust and Python hybrid projects):\n```bash\npip install maturin>=1.0,<2.0\n```\n\n3. Build and install:\n```bash\n# Install in development mode\nmaturin develop\n\n# Or build wheel package\nmaturin build --release\npip install target/wheels/numpack-*.whl\n```\n\n#### Platform-Specific Notes\n\n- **Linux Users**:\n - Ensure python3-dev (Ubuntu/Debian) or python3-devel (Fedora/RHEL) is installed\n - If using conda environment, make sure the appropriate compiler toolchain is installed\n\n- **macOS Users**:\n - Make sure Xcode Command Line Tools are installed: `xcode-select --install`\n - Supports both Intel and Apple Silicon architectures\n\n- **Windows Users**:\n - Visual Studio or Visual Studio Build Tools required\n - Ensure \"Desktop development with C++\" workload is installed\n\n\n## Usage\n\n### Basic Operations\n\n```python\nimport numpy as np\nfrom numpack import NumPack\n\n# Create a NumPack instance\nnpk = NumPack(\"data_directory\")\n\n# Save arrays\narrays = {\n 'array1': np.random.rand(1000, 100).astype(np.float32),\n 'array2': np.random.rand(500, 200).astype(np.float32)\n}\nnpk.save(arrays)\n\n# Load arrays\n# Normal mode\nloaded = npk.load(\"array1\")\n\n# Memory mapping mode for large arrays\nwith npk.mmap_mode() as mmap_npk:\n # Access specific arrays\n array1 = mmap_npk.load('array1')\n array2 = mmap_npk.load('array2')\n```\n\n### Advanced Operations\n\n```python\n# Replace specific rows\nreplacement = np.random.rand(10, 100).astype(np.float32)\nnpk.replace({'array1': replacement}, [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]) # Using list indices\nnpk.replace({'array1': replacement}, slice(0, 10)) # Using slice notation\n\n# Append new arrays\nnew_arrays = {\n 'array3': np.random.rand(200, 100).astype(np.float32)\n}\nnpk.append(new_arrays)\n\n# Drop arrays or specific rows\nnpk.drop('array1') # Drop entire array\nnpk.drop(['array1', 'array2']) # Drop multiple arrays\nnpk.drop('array2', [0, 1, 2]) # Drop specific rows\n\n# Random access operations\ndata = npk.getitem('array1', [0, 1, 2]) # Access specific rows\ndata = npk.getitem('array1', slice(0, 10)) # Access using slice\ndata = npk['array1'] # Dictionary-style access for entire array\n\n# Metadata operations\nshapes = npk.get_shape() # Get shapes of all arrays\nshapes = npk.get_shape('array1') # Get shape of specific array\nmembers = npk.get_member_list() # Get list of array names\nmtime = npk.get_modify_time('array1') # Get modification time\nmetadata = npk.get_metadata() # Get complete metadata\n\n# Stream loading for large arrays\nfor batch in npk.stream_load('array1', buffer_size=1000):\n # Process 1000 rows at a time\n process_batch(batch)\n\n# Reset/clear storage\nnpk.reset() # Clear all arrays\n\n# Iterate over all arrays\nfor array_name in npk:\n data = npk[array_name]\n print(f\"{array_name} shape: {data.shape}\")\n```\n\n### Lazy Loading and Buffer Operations\n\nNumPack supports lazy loading and buffer operations, which are particularly useful for handling large-scale datasets. Using the `lazy=True` parameter enables data to be loaded only when actually needed, making it ideal for streaming processing or scenarios where only partial data access is required.\n\n```python\nfrom numpack import NumPack\nimport numpy as np\n\n# Create NumPack instance and save large-scale data\nnpk = NumPack(\"test_data/\", drop_if_exists=True)\na = np.random.random((1000000, 128)) # Create a large array\nnpk.save({\"arr1\": a})\n\n# Lazy loading - keeps data in buffer\nlazy_array = npk.load(\"arr1\", lazy=True) # LazyArray Object\n\n# Perform computations with lazy-loaded data\n# Only required data is loaded into memory\nsimilarity_scores = np.inner(a[0], npk.load(\"arr1\", lazy=True))\n```\n\n### Memory Mapping Mode\n\nFor large arrays, memory mapping mode provides more efficient memory usage:\n\n```python\n# Using memory mapping mode\nwith npk.mmap_mode() as mmap_npk:\n # Access specific arrays\n array1 = mmap_npk.load('array1') # Array is not fully loaded into memory\n array2 = mmap_npk.load('array2')\n \n # Perform operations on memory-mapped arrays\n result = array1[0:1000] + array2[0:1000]\n```\n\n## Performance\n\nNumPack offers significant performance improvements compared to traditional NumPy storage methods, especially in data modification operations and random access. Below are detailed benchmark results:\n\n### Benchmark Results\n\nThe following benchmarks were performed on an MacBook Pro (Apple Silicon) with arrays of size 1M x 10 and 500K x 5 (float32).\n\n#### Storage Operations\n\n| Operation | NumPack | NumPy NPZ | NumPy NPY |\n|-----------|---------|-----------|-----------|\n| Save | 0.014s (0.86x NPZ, 0.57x NPY) | 0.012s | 0.008s |\n| Full Load | 0.008s (1.63x NPZ, 0.88x NPY) | 0.013s | 0.007s |\n| Selective Load | 0.006s (1.50x NPZ, -) | 0.009s | - |\n| Mmap Load | 0.006s (2.00x NPZ, 0.67x NPY) | 0.012s | 0.004s |\n\n#### Data Modification Operations\n\n| Operation | NumPack | NumPy NPZ | NumPy NPY |\n|-----------|---------|-----------|-----------|\n| Single Row Replace | 0.000s (19.00x NPZ, 12.00x NPY) | 0.019s | 0.012s |\n| Continuous Rows (10K) | 0.001s (20.00x NPZ, 12.00x NPY) | 0.020s | 0.012s |\n| Random Rows (10K) | 0.015s (1.33x NPZ, 0.87x NPY) | 0.020s | 0.013s |\n| Large Data Replace (500K) | 0.019s (1.00x NPZ, 0.74x NPY) | 0.019s | 0.014s |\n\n#### Drop Operations\n\n| Operation (1M rows, float32) | NumPack | NumPy NPZ | NumPy NPY |\n|-----------|---------|-----------|-----------|\n| Drop Array | 0.001s (22.00x NPZ, 1.00x NPY) | 0.022s | 0.001s |\n| Drop First Row | 0.014s (3.21x NPZ, 1.93x NPY) | 0.045s | 0.027s |\n| Drop Last Row | 0.000s (\u221ex NPZ, \u221ex NPY) | 0.045s | 0.027s |\n| Drop Middle Row | 0.014s (3.21x NPZ, 1.93x NPY) | 0.045s | 0.027s |\n| Drop Front Continuous (10K rows) | 0.016s (2.81x NPZ, 1.69x NPY) | 0.045s | 0.027s |\n| Drop Middle Continuous (10K rows) | 0.016s (2.81x NPZ, 1.69x NPY) | 0.045s | 0.027s |\n| Drop End Continuous (10K rows) | 0.001s (45.00x NPZ, 27.00x NPY) | 0.045s | 0.027s |\n| Drop Random Rows (10K rows) | 0.018s (2.50x NPZ, 1.50x NPY) | 0.045s | 0.027s |\n| Drop Near Non-continuous (10K rows) | 0.015s (3.00x NPZ, 1.80x NPY) | 0.045s | 0.027s |\n\n#### Append Operations\n\n| Operation | NumPack | NumPy NPZ | NumPy NPY |\n|-----------|---------|-----------|-----------|\n| Small Append (1K rows) | 0.000s (22.00x NPZ, 18.00x NPY) | 0.022s | 0.018s |\n| Large Append (500K rows) | 0.003s (9.67x NPZ, 6.67x NPY) | 0.029s | 0.020s |\n\n#### Random Access Performance (10K indices)\n\n| Operation | NumPack | NumPy NPZ | NumPy NPY |\n|-----------|---------|-----------|-----------|\n| Random Access | 0.008s (2.00x NPZ, 1.38x NPY) | 0.016s | 0.011s |\n\n#### File Size Comparison\n\n| Format | Size | Ratio |\n|--------|------|-------|\n| NumPack | 47.68 MB | 1.0x |\n| NPZ | 47.68 MB | 1.0x |\n| NPY | 47.68 MB | 1.0x |\n\n#### Large-scale Data Operations (>1B rows, Float32)\n\n| Operation | NumPack | NumPy NPZ | NumPy NPY |\n|-----------|---------|-----------|-----------|\n| Replace | Zero-copy in-place modification | Memory exceeded | Memory exceeded |\n| Drop | Zero-copy in-place deletion | Memory exceeded | Memory exceeded |\n| Append | Zero-copy in-place addition | Memory exceeded | Memory exceeded |\n| Random Access | Near-hardware I/O speed | Memory exceeded | Memory exceeded |\n\n#### Matrix Computation Performance (1M rows x 128 columns, Float32)\n\n| Operation | NumPack | NumPy NPZ | NumPy NPY | In-Memory |\n|-----------|---------|-----------|-----------|-----------|\n| Inner Product | 0.019s (6.58x NPZ, 1.00x NPY) | 0.125s | 0.019s | 0.011s |\n| Other calculations are similar to the above case | ... | ... | ... | ... |\n\n> **Key Advantage**: NumPack achieves the same performance as NumPy's NPY mmap (0.019s) for matrix computations, with several implementation advantages:\n> - Uses Arc<Mmap> for reference counting, ensuring automatic resource cleanup\n> - Implements MMAP_CACHE to avoid redundant data loading\n> - Linux-specific optimizations with huge pages and sequential access hints\n> - Supports parallel I/O operations for improved data throughput\n> - Optimizes memory usage through Buffer Pool to reduce fragmentation\n\n### Key Performance Highlights\n\n1. **Data Modification**:\n - Single row replacement: NumPack is **19x faster** than NPZ and **12x faster** than NPY\n - Continuous rows: NumPack is **20x faster** than NPZ and **12x faster** than NPY\n - Random rows: NumPack is **1.33x faster** than NPZ but **0.87x slower** than NPY\n - Large data replacement: NumPack is comparable to NPZ but **0.74x slower** than NPY\n\n2. **Drop Operations**:\n - Drop array: NumPack is **22x faster** than NPZ and comparable to NPY\n - Drop rows: NumPack is currently **0.61x slower** than NPZ and **0.41x slower** than NPY\n - NumPack provides efficient in-place row deletion without full file rewrite\n\n3. **Append Operations**:\n - Small append (1K rows): NumPack is **22x faster** than NPZ and **18x faster** than NPY\n - Large append (500K rows): NumPack is **9.67x faster** than NPZ and **6.67x faster** than NPY\n - NumPack excels at both small and large append operations\n\n4. **Loading Performance**:\n - Full load: NumPack is **1.63x faster** than NPZ but **0.88x slower** than NPY\n - Memory-mapped load: NumPack is **2.00x faster** than NPZ but **0.67x slower** than NPY\n - Selective load: NumPack is **1.50x faster** than NPZ\n\n5. **Random Access**:\n - NumPack is **2.00x faster** than NPZ and **1.38x faster** than NPY for random index access\n\n6. **Storage Efficiency**:\n - All formats achieve identical compression ratios (47.68 MB)\n - NumPack maintains high performance while keeping file sizes competitive\n\n7. **Matrix Computation**:\n - NumPack matches NPY mmap performance while providing better resource management\n - **6.58x faster** than NPZ mmap for matrix operations\n - Only 1.72x slower than pure in-memory computation\n - Zero risk of file descriptor leaks or resource exhaustion\n\n> Note: All benchmarks were performed with float32 arrays. Performance may vary depending on data types, array sizes, and system configurations. Numbers greater than 1.0x indicate faster performance, while numbers less than 1.0x indicate slower performance.\n\n## Contributing\n\nContributions are welcome! Please feel free to submit a Pull Request.\n\n## License\n\nThis project is licensed under the Apache License, Version 2.0 - see the LICENSE file for details.\n\nCopyright 2024 NumPack Contributors\n\nLicensed under the Apache License, Version 2.0 (the \"License\");\nyou may not use this file except in compliance with the License.\nYou may obtain a copy of the License at\n\n http://www.apache.org/licenses/LICENSE-2.0\n\nUnless required by applicable law or agreed to in writing, software\ndistributed under the License is distributed on an \"AS IS\" BASIS,\nWITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\nSee the License for the specific language governing permissions and\nlimitations under the License.\n\n",
"bugtrack_url": null,
"license": null,
"summary": "A high-performance array storage and manipulation library",
"version": "0.1.6",
"project_urls": null,
"split_keywords": [
"numpy",
" array",
" storage",
" performance"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "78b227d298799b6166053b0441223475bb195f9196a02c5f6c77f85feed6720f",
"md5": "6b163f93569277d2722a2bc262321ed7",
"sha256": "f0ab9f79c63dfef6dd9c36df267bd35220f57b6b1fdfd83acfaee854c88bb541"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp310-cp310-macosx_10_12_x86_64.whl",
"has_sig": false,
"md5_digest": "6b163f93569277d2722a2bc262321ed7",
"packagetype": "bdist_wheel",
"python_version": "cp310",
"requires_python": ">=3.9",
"size": 565662,
"upload_time": "2025-01-20T03:56:29",
"upload_time_iso_8601": "2025-01-20T03:56:29.695312Z",
"url": "https://files.pythonhosted.org/packages/78/b2/27d298799b6166053b0441223475bb195f9196a02c5f6c77f85feed6720f/numpack-0.1.6-cp310-cp310-macosx_10_12_x86_64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "20635f5720639c85c7a05401e6593925cb8c4ac03143b41797a380306473914d",
"md5": "1216d7c7929184786078fb0323854444",
"sha256": "0014b4debd1d7ce0ecf167be1f01a520a62a998cc2aba2a021699f969505487a"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp310-cp310-macosx_11_0_arm64.whl",
"has_sig": false,
"md5_digest": "1216d7c7929184786078fb0323854444",
"packagetype": "bdist_wheel",
"python_version": "cp310",
"requires_python": ">=3.9",
"size": 547520,
"upload_time": "2025-01-20T03:56:31",
"upload_time_iso_8601": "2025-01-20T03:56:31.675545Z",
"url": "https://files.pythonhosted.org/packages/20/63/5f5720639c85c7a05401e6593925cb8c4ac03143b41797a380306473914d/numpack-0.1.6-cp310-cp310-macosx_11_0_arm64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "57de7f6e1f661e81b4676aac691c90fcc2982b3281661d15bc90fb49dd4da236",
"md5": "63c39b98bce99cb9f1a98aa27252a0d9",
"sha256": "57eb7008beb22023ea8bf0cf46a17b4de23efa1c0075c8aeb86c9c74ef05135b"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl",
"has_sig": false,
"md5_digest": "63c39b98bce99cb9f1a98aa27252a0d9",
"packagetype": "bdist_wheel",
"python_version": "cp310",
"requires_python": ">=3.9",
"size": 599114,
"upload_time": "2025-01-20T03:56:32",
"upload_time_iso_8601": "2025-01-20T03:56:32.749920Z",
"url": "https://files.pythonhosted.org/packages/57/de/7f6e1f661e81b4676aac691c90fcc2982b3281661d15bc90fb49dd4da236/numpack-0.1.6-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "6986e2e669986b288024959b52d10367d090f787117d603528d5b52ca326bb72",
"md5": "695018b91afd5b121ab68c880690cb49",
"sha256": "f64d499ed7272d221e514a42f8fd18d11322907b6b6d542d5ec0e0920baf6f51"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp310-cp310-win_amd64.whl",
"has_sig": false,
"md5_digest": "695018b91afd5b121ab68c880690cb49",
"packagetype": "bdist_wheel",
"python_version": "cp310",
"requires_python": ">=3.9",
"size": 413143,
"upload_time": "2025-01-20T03:56:34",
"upload_time_iso_8601": "2025-01-20T03:56:34.502929Z",
"url": "https://files.pythonhosted.org/packages/69/86/e2e669986b288024959b52d10367d090f787117d603528d5b52ca326bb72/numpack-0.1.6-cp310-cp310-win_amd64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "003afadc5dc0ea2e32ff2e1ff0a3ea3a5d44ab45beb87b9ef8c8539e3576c2b1",
"md5": "b86b9e213de0ee0b0f388ee54e1af0b1",
"sha256": "144360407f2c76b6b9e5dc657856058f03d06fe37b98473411ee300d1ee7594e"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp311-cp311-macosx_10_12_x86_64.whl",
"has_sig": false,
"md5_digest": "b86b9e213de0ee0b0f388ee54e1af0b1",
"packagetype": "bdist_wheel",
"python_version": "cp311",
"requires_python": ">=3.9",
"size": 565662,
"upload_time": "2025-01-20T03:56:35",
"upload_time_iso_8601": "2025-01-20T03:56:35.667029Z",
"url": "https://files.pythonhosted.org/packages/00/3a/fadc5dc0ea2e32ff2e1ff0a3ea3a5d44ab45beb87b9ef8c8539e3576c2b1/numpack-0.1.6-cp311-cp311-macosx_10_12_x86_64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "6a0114a0d57af5b1271b32e7fbf65572fa281e989e7c8595a95a113a9a66f57a",
"md5": "6eb81ecb81e16e6cf90d34521a719aef",
"sha256": "32100f4c213a325cb98816e7d4f8473e5c8b8e426af32d40452f0dc041c4103c"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp311-cp311-macosx_11_0_arm64.whl",
"has_sig": false,
"md5_digest": "6eb81ecb81e16e6cf90d34521a719aef",
"packagetype": "bdist_wheel",
"python_version": "cp311",
"requires_python": ">=3.9",
"size": 547661,
"upload_time": "2025-01-20T03:56:37",
"upload_time_iso_8601": "2025-01-20T03:56:37.410301Z",
"url": "https://files.pythonhosted.org/packages/6a/01/14a0d57af5b1271b32e7fbf65572fa281e989e7c8595a95a113a9a66f57a/numpack-0.1.6-cp311-cp311-macosx_11_0_arm64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "555d2a46637f12a18d39fe60e19a0862fd4d29f5d0c481f71cf16f3c05003877",
"md5": "b1dff5ff7b8663c335899b22b1937d38",
"sha256": "a97d10135f1a91d4c7ac29b7e196a3b14aab15fe5ac67187b2be8fb27e5c2092"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl",
"has_sig": false,
"md5_digest": "b1dff5ff7b8663c335899b22b1937d38",
"packagetype": "bdist_wheel",
"python_version": "cp311",
"requires_python": ">=3.9",
"size": 599093,
"upload_time": "2025-01-20T03:56:39",
"upload_time_iso_8601": "2025-01-20T03:56:39.288273Z",
"url": "https://files.pythonhosted.org/packages/55/5d/2a46637f12a18d39fe60e19a0862fd4d29f5d0c481f71cf16f3c05003877/numpack-0.1.6-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "66af9fa5f4f599efb9dd5daa3e7d1fdf205c33f94d4bce09df2c9ee2b3213db3",
"md5": "7bd4bc11fec3403baf7e9dd9e168ea2c",
"sha256": "8f9e87f08b425231282466ef6a16b0ecb4210f203eb09d226ad6812fc8422e4a"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp311-cp311-win_amd64.whl",
"has_sig": false,
"md5_digest": "7bd4bc11fec3403baf7e9dd9e168ea2c",
"packagetype": "bdist_wheel",
"python_version": "cp311",
"requires_python": ">=3.9",
"size": 413156,
"upload_time": "2025-01-20T03:56:40",
"upload_time_iso_8601": "2025-01-20T03:56:40.849246Z",
"url": "https://files.pythonhosted.org/packages/66/af/9fa5f4f599efb9dd5daa3e7d1fdf205c33f94d4bce09df2c9ee2b3213db3/numpack-0.1.6-cp311-cp311-win_amd64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "aad2f403a349df1ed263122d460c1947337471b03e45f45686179a63a7a800d6",
"md5": "0131b4c519fe9cd932586ae8854ba85a",
"sha256": "cc76f80aeaf08128322d70bc89736d88bc9a961a7689826b3589a5b2d285710c"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp312-cp312-macosx_10_12_x86_64.whl",
"has_sig": false,
"md5_digest": "0131b4c519fe9cd932586ae8854ba85a",
"packagetype": "bdist_wheel",
"python_version": "cp312",
"requires_python": ">=3.9",
"size": 565899,
"upload_time": "2025-01-20T03:56:42",
"upload_time_iso_8601": "2025-01-20T03:56:42.579236Z",
"url": "https://files.pythonhosted.org/packages/aa/d2/f403a349df1ed263122d460c1947337471b03e45f45686179a63a7a800d6/numpack-0.1.6-cp312-cp312-macosx_10_12_x86_64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "1e4136468b518381ee9b78651b9396068dba9c56c8df1b9737837cd3da60b78a",
"md5": "caf6f295aa47da7a1b30adf5b64cf014",
"sha256": "e1bd2d41464841c51d4b50e7a4704c9c70ab348e502612e1a5cbafd821876378"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp312-cp312-macosx_11_0_arm64.whl",
"has_sig": false,
"md5_digest": "caf6f295aa47da7a1b30adf5b64cf014",
"packagetype": "bdist_wheel",
"python_version": "cp312",
"requires_python": ">=3.9",
"size": 548348,
"upload_time": "2025-01-20T03:56:44",
"upload_time_iso_8601": "2025-01-20T03:56:44.427863Z",
"url": "https://files.pythonhosted.org/packages/1e/41/36468b518381ee9b78651b9396068dba9c56c8df1b9737837cd3da60b78a/numpack-0.1.6-cp312-cp312-macosx_11_0_arm64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "265990002a15969e92bd2c65e57275c68913b0050329c2da3f0f409eff5d5214",
"md5": "9e215b7eb4b1533d87795ac99b832a20",
"sha256": "47b40ba6199f6fe90b6967bab78df3ddfdb54f9b1a9a6a637db8070607316525"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl",
"has_sig": false,
"md5_digest": "9e215b7eb4b1533d87795ac99b832a20",
"packagetype": "bdist_wheel",
"python_version": "cp312",
"requires_python": ">=3.9",
"size": 599481,
"upload_time": "2025-01-20T03:56:46",
"upload_time_iso_8601": "2025-01-20T03:56:46.280962Z",
"url": "https://files.pythonhosted.org/packages/26/59/90002a15969e92bd2c65e57275c68913b0050329c2da3f0f409eff5d5214/numpack-0.1.6-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "7eac9af06d68f14e1b628921736ebb33da5e9fc1f621893776b626b6025e84b3",
"md5": "d363e27d08c2a25d9266b3fc1b733903",
"sha256": "f0d6623b45aaa9ef6d3d653e8d56faa955d182ab2aab49617ea01de636f5159a"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp312-cp312-win_amd64.whl",
"has_sig": false,
"md5_digest": "d363e27d08c2a25d9266b3fc1b733903",
"packagetype": "bdist_wheel",
"python_version": "cp312",
"requires_python": ">=3.9",
"size": 413011,
"upload_time": "2025-01-20T03:56:48",
"upload_time_iso_8601": "2025-01-20T03:56:48.043847Z",
"url": "https://files.pythonhosted.org/packages/7e/ac/9af06d68f14e1b628921736ebb33da5e9fc1f621893776b626b6025e84b3/numpack-0.1.6-cp312-cp312-win_amd64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "e9811ee0c685f0db7310d0ca2f93620122161fd082b6aacc4ddb05744e64c97a",
"md5": "bb57dc28ecdf0d3cca97ff0e7940d0ca",
"sha256": "2fdce8098fd04ad58628793f0651cff68545326fb7be87760cbeb842690f3985"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp313-cp313-macosx_10_12_x86_64.whl",
"has_sig": false,
"md5_digest": "bb57dc28ecdf0d3cca97ff0e7940d0ca",
"packagetype": "bdist_wheel",
"python_version": "cp313",
"requires_python": ">=3.9",
"size": 565897,
"upload_time": "2025-01-20T03:56:49",
"upload_time_iso_8601": "2025-01-20T03:56:49.085734Z",
"url": "https://files.pythonhosted.org/packages/e9/81/1ee0c685f0db7310d0ca2f93620122161fd082b6aacc4ddb05744e64c97a/numpack-0.1.6-cp313-cp313-macosx_10_12_x86_64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "2b4a8d96007f7630781cbe7e7f712396d6bc489ddf8105db1df13523893a4a66",
"md5": "6d07e8b66d5e3302ef51c981655b08f7",
"sha256": "e37433150f6040e7466ad3451b582dabba03622bd2beb387560802ce3e104ed1"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp313-cp313-macosx_11_0_arm64.whl",
"has_sig": false,
"md5_digest": "6d07e8b66d5e3302ef51c981655b08f7",
"packagetype": "bdist_wheel",
"python_version": "cp313",
"requires_python": ">=3.9",
"size": 548349,
"upload_time": "2025-01-20T03:56:50",
"upload_time_iso_8601": "2025-01-20T03:56:50.607994Z",
"url": "https://files.pythonhosted.org/packages/2b/4a/8d96007f7630781cbe7e7f712396d6bc489ddf8105db1df13523893a4a66/numpack-0.1.6-cp313-cp313-macosx_11_0_arm64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "3ef91a2878d70e9e06e959890f6f91264f950c9e1db6fb3c88704e334bcee574",
"md5": "4edd2f77c30bae6b96e1f3c90cf15353",
"sha256": "0d53eceae5cf0d5c3de6850f31ec7de7ca9b48a7dff6bd1e8222617ddbc6e7ee"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl",
"has_sig": false,
"md5_digest": "4edd2f77c30bae6b96e1f3c90cf15353",
"packagetype": "bdist_wheel",
"python_version": "cp313",
"requires_python": ">=3.9",
"size": 599481,
"upload_time": "2025-01-20T03:56:51",
"upload_time_iso_8601": "2025-01-20T03:56:51.778492Z",
"url": "https://files.pythonhosted.org/packages/3e/f9/1a2878d70e9e06e959890f6f91264f950c9e1db6fb3c88704e334bcee574/numpack-0.1.6-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "98c19748497b8b62f32d7c1096420a12f072d8158cb5926200e48b31b50a005a",
"md5": "ae8cb538bc384933f95eb9432206f6d7",
"sha256": "9a361f09a9ad8f70d8757fd872c443024f1f81fc91524a1e3998297499b35c41"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp313-cp313-win_amd64.whl",
"has_sig": false,
"md5_digest": "ae8cb538bc384933f95eb9432206f6d7",
"packagetype": "bdist_wheel",
"python_version": "cp313",
"requires_python": ">=3.9",
"size": 413004,
"upload_time": "2025-01-20T03:56:53",
"upload_time_iso_8601": "2025-01-20T03:56:53.641286Z",
"url": "https://files.pythonhosted.org/packages/98/c1/9748497b8b62f32d7c1096420a12f072d8158cb5926200e48b31b50a005a/numpack-0.1.6-cp313-cp313-win_amd64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "ab4076f39cc442fb8d075db712790be470a0064994c40f361e0a4ab3b88776fc",
"md5": "9d00ac86e5c91d09c5c449355b380a7f",
"sha256": "f771e9bfb96bc499a09129dc91aeb2f96e63724178569766d92b4734ff4dab20"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp39-cp39-macosx_10_12_x86_64.whl",
"has_sig": false,
"md5_digest": "9d00ac86e5c91d09c5c449355b380a7f",
"packagetype": "bdist_wheel",
"python_version": "cp39",
"requires_python": ">=3.9",
"size": 566259,
"upload_time": "2025-01-20T03:56:54",
"upload_time_iso_8601": "2025-01-20T03:56:54.779580Z",
"url": "https://files.pythonhosted.org/packages/ab/40/76f39cc442fb8d075db712790be470a0064994c40f361e0a4ab3b88776fc/numpack-0.1.6-cp39-cp39-macosx_10_12_x86_64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "f807a18ca38a6cad2b88e9c5bbdfa750dd3968a9cd759fd21d605996a9b1b1ab",
"md5": "0ecd15293ba1f8b3ab029478c5a96a93",
"sha256": "cb88f6a3c9fbaf1d4273c03e8485b5d4e9afcd44d1a77769f1dc4d2a2915545b"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp39-cp39-macosx_11_0_arm64.whl",
"has_sig": false,
"md5_digest": "0ecd15293ba1f8b3ab029478c5a96a93",
"packagetype": "bdist_wheel",
"python_version": "cp39",
"requires_python": ">=3.9",
"size": 547712,
"upload_time": "2025-01-20T03:56:56",
"upload_time_iso_8601": "2025-01-20T03:56:56.570993Z",
"url": "https://files.pythonhosted.org/packages/f8/07/a18ca38a6cad2b88e9c5bbdfa750dd3968a9cd759fd21d605996a9b1b1ab/numpack-0.1.6-cp39-cp39-macosx_11_0_arm64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "3d85f4a10bbf4e431679b529cff2a14979f8c67bfa95f3e0858fa2c61a2e0df1",
"md5": "2b2dda8ef1baa8ade7fc5ec5ae5698a5",
"sha256": "a1a4a24b1896c8088b525f631a13fd8d628fd844959673b5165f5df4588820a2"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl",
"has_sig": false,
"md5_digest": "2b2dda8ef1baa8ade7fc5ec5ae5698a5",
"packagetype": "bdist_wheel",
"python_version": "cp39",
"requires_python": ">=3.9",
"size": 599468,
"upload_time": "2025-01-20T03:56:58",
"upload_time_iso_8601": "2025-01-20T03:56:58.357124Z",
"url": "https://files.pythonhosted.org/packages/3d/85/f4a10bbf4e431679b529cff2a14979f8c67bfa95f3e0858fa2c61a2e0df1/numpack-0.1.6-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "14f7883ba1e395f682d53fd3a482665743bcd8d6d3c03bd34c82779e7bee9dae",
"md5": "232bc55db2d4c8eda2c88860ab9ad2c8",
"sha256": "dac314bd38e285d5a84d0e0cc975632851ca2bc143a76e0e8f8988a702275641"
},
"downloads": -1,
"filename": "numpack-0.1.6-cp39-cp39-win_amd64.whl",
"has_sig": false,
"md5_digest": "232bc55db2d4c8eda2c88860ab9ad2c8",
"packagetype": "bdist_wheel",
"python_version": "cp39",
"requires_python": ">=3.9",
"size": 413485,
"upload_time": "2025-01-20T03:56:59",
"upload_time_iso_8601": "2025-01-20T03:56:59.471489Z",
"url": "https://files.pythonhosted.org/packages/14/f7/883ba1e395f682d53fd3a482665743bcd8d6d3c03bd34c82779e7bee9dae/numpack-0.1.6-cp39-cp39-win_amd64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "87e5a27baad34a9de856bc5a666e41e0b2345202f74e02b06d2287f150a7cf3a",
"md5": "91e99f9409dda5023eac264a2355e6e9",
"sha256": "dd871ac02c3572bd801c4c9693f81d01be0279a431b702f3c409ef9c89076628"
},
"downloads": -1,
"filename": "numpack-0.1.6-pp310-pypy310_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl",
"has_sig": false,
"md5_digest": "91e99f9409dda5023eac264a2355e6e9",
"packagetype": "bdist_wheel",
"python_version": "pp310",
"requires_python": ">=3.9",
"size": 600561,
"upload_time": "2025-01-20T03:57:01",
"upload_time_iso_8601": "2025-01-20T03:57:01.431934Z",
"url": "https://files.pythonhosted.org/packages/87/e5/a27baad34a9de856bc5a666e41e0b2345202f74e02b06d2287f150a7cf3a/numpack-0.1.6-pp310-pypy310_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "8d76e000ae235a09ca020d2bcd741074efb92d5ca635cade234f3d5a1de0d272",
"md5": "761331d413159a5f62267a6483728bd6",
"sha256": "d5b810c54720c2212e089ea6f6f95f28f60ffc46d0eb802b8040e46a79c6ba89"
},
"downloads": -1,
"filename": "numpack-0.1.6-pp39-pypy39_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl",
"has_sig": false,
"md5_digest": "761331d413159a5f62267a6483728bd6",
"packagetype": "bdist_wheel",
"python_version": "pp39",
"requires_python": ">=3.9",
"size": 600430,
"upload_time": "2025-01-20T03:57:02",
"upload_time_iso_8601": "2025-01-20T03:57:02.617308Z",
"url": "https://files.pythonhosted.org/packages/8d/76/e000ae235a09ca020d2bcd741074efb92d5ca635cade234f3d5a1de0d272/numpack-0.1.6-pp39-pypy39_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2025-01-20 03:56:29",
"github": false,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"lcname": "numpack"
}
JSON