API

scitex-decorators — decorator library extracted from scitex.decorators (standalone).

Includes type-conversion (numpy_fn, torch_fn, pandas_fn, xarray_fn, signal_fn), caching (cache_disk, cache_disk_async, cache_mem), batching, deprecation, timeout, and ordering helpers.

class scitex_decorators.AutoOrderDecorator(name)

Decorator that collects and applies decorators in predefined order.

exception scitex_decorators.ConversionWarning

scitex_decorators.batch_fn(func)

Return type:

Callable

scitex_decorators.batch_numpy_fn(func)

Combined decorator: numpy_fn → batch_fn.

Converts inputs to numpy arrays, then processes in batches. This is the recommended order for NumPy operations.

Return type:

Callable

Example

>>> @numpy_batch_fn
... def process_data(x, axis=None):
...     return np.mean(x, axis=axis)

scitex_decorators.batch_pandas_fn(func)

Combined decorator: pandas_fn → batch_fn.

Converts inputs to pandas DataFrames, then processes in batches. This is the recommended order for Pandas operations.

Return type:

Callable

Example

>>> @pandas_batch_fn
... def process_data(df):
...     return df.describe()

scitex_decorators.batch_torch_fn(func)

Combined decorator: torch_fn → batch_fn.

Converts inputs to torch tensors, then processes in batches. This is the recommended order for PyTorch operations.

Return type:

Callable

Example

>>> @torch_batch_fn
... def process_data(x, dim=None):
...     return x.mean(dim=dim)

scitex_decorators.cache_disk(func)

Disk caching decorator that uses joblib.Memory.

joblib is lazy-imported to keep import scitex.decorators (and the transitive import scitex.io chain) usable on venvs without joblib installed. Without lazy-import, the eager from joblib import Memory raised ModuleNotFoundError: No module named 'joblib' at import time and broke any caller of scitex.io that didn’t need caching at all (todo#442, same class as #441 / #279).

Usage:

@cache_disk
def expensive_function(x):
    return x ** 2

scitex_decorators.cache_disk_async(func)

Disk caching decorator for async functions.

joblib is lazy-imported inside the decorator body so that import scitex.decorators does not fail on venvs without joblib (todo#442, same class as #441 / #279). Without this, the eager top-level from joblib import Memory propagates ModuleNotFoundError up the scitex.io import chain and breaks unrelated callers.

Usage:

@cache_disk_async
async def expensive_async_function(x):
    await asyncio.sleep(1)
    return x ** 2

scitex_decorators.cache_mem(user_function)

scitex_decorators.deprecated(reason=None, forward_to=None)

A decorator to mark functions as deprecated. It will result in a warning being emitted when the function is used.

Parameters:

• reason (str) – A human-readable string explaining why this function was deprecated.

• forward_to (str) – Optional module path to forward calls to (e.g., “..session.start”). If provided, calls will be forwarded to the new function instead of executing the original deprecated function.

scitex_decorators.disable_auto_order()

Disable auto-ordering and restore original decorators.

scitex_decorators.enable_auto_order()

Enable auto-ordering for all decorators in the scitex_decorators module.

This replaces the standard decorators with auto-ordering versions.

Example

>>> import scitex_decorators
>>> scitex_decorators.enable_auto_order()
>>>
>>> # Now decorators will auto-order regardless of how they're written
>>> @scitex_decorators.batch_fn
>>> @scitex_decorators.torch_fn
>>> def my_func(x):
...     return x.mean()

scitex_decorators.is_cuda(*args, **kwargs)

Check if any input is a CUDA tensor.

Return type:

bool

scitex_decorators.is_nested_decorator()

Check if we’re in a nested decorator context.

scitex_decorators.is_torch(*args, **kwargs)

Check if any input is a PyTorch tensor.

Return type:

bool

scitex_decorators.not_implemented(func)

Decorator to mark methods as not implemented, issue a warning, and prevent their execution.

Parameters:

func (callable) – The function or method to decorate.

Returns:

A wrapper function that issues a warning and raises NotImplementedError when called.

Return type:

callable

scitex_decorators.numpy_batch_fn(func)

Combined decorator: numpy_fn → batch_fn.

Converts inputs to numpy arrays, then processes in batches. This is the recommended order for NumPy operations.

Return type:

Callable

Example

>>> @numpy_batch_fn
... def process_data(x, axis=None):
...     return np.mean(x, axis=axis)

scitex_decorators.numpy_fn(func)

Return type:

Callable

scitex_decorators.pandas_batch_fn(func)

Combined decorator: pandas_fn → batch_fn.

Converts inputs to pandas DataFrames, then processes in batches. This is the recommended order for Pandas operations.

Return type:

Callable

Example

>>> @pandas_batch_fn
... def process_data(df):
...     return df.describe()

scitex_decorators.pandas_fn(func)

Return type:

Callable

scitex_decorators.preserve_doc(loader_func)

Wrap the loader functions to preserve their docstrings

scitex_decorators.signal_fn(func)

Decorator for signal processing functions that converts only the first argument (signal) to torch tensor.

This decorator is designed for DSP functions where: - The first argument is the signal data that should be converted to torch tensor - Other arguments (like sampling frequency, bands, etc.) should remain as-is

Return type:

Callable

scitex_decorators.timeout(seconds=10, error_message='Timeout')

scitex_decorators.to_numpy(*args, return_fn=<function _return_if>, **kwargs)

Convert various data types to NumPy arrays.

Return type:

Any

scitex_decorators.to_torch(*args, return_fn=<function _return_if>, device=None, **kwargs)

Convert various data types to PyTorch tensors.

Return type:

Any

scitex_decorators.torch_batch_fn(func)

Combined decorator: torch_fn → batch_fn.

Converts inputs to torch tensors, then processes in batches. This is the recommended order for PyTorch operations.

Return type:

Callable

Example

>>> @torch_batch_fn
... def process_data(x, dim=None):
...     return x.mean(dim=dim)

scitex_decorators.torch_fn(func)

Decorator for PyTorch function compatibility.

Automatically converts inputs to PyTorch tensors and handles various data types gracefully. Preserves the original input type in the output.

Features

• Converts inputs to PyTorch tensors

• Preserves scalar parameters (int, float, bool, str)

• Preserves dimension tuples like dim=(0, 1)

• Handles nested lists/tuples gracefully

• Automatically converts axis to dim for torch functions

• Applies device=”cuda” if available

• Returns output in same type as input (numpy->numpy, pandas->pandas, etc.)

type func:

Callable

param func:

The function to decorate

type func:

Callable

returns:

The decorated function

rtype:

Callable

Examples

>>> @torch_fn
... def mean_squared(x, dim=None):
...     return (x ** 2).mean(dim=dim)
>>>
>>> # Works with numpy arrays
>>> result = mean_squared(np.array([1, 2, 3]))
>>>
>>> # Works with nested lists
>>> result = mean_squared([[1, 2], [3, 4]])
>>>
>>> # Preserves dimension tuples
>>> result = mean_squared(data, dim=(0, 1))

Notes

For optimal performance with batch processing, apply torch_fn before batch_fn: @batch_fn @torch_fn def my_function(x): …

Or use auto-ordering to handle this automatically.

scitex_decorators.wrap(func)

Basic function wrapper that preserves function metadata.

Usage:

@wrap
def my_function(x):
    return x + 1

# Or manually:
def my_function(x):
    return x + 1
wrapped_func = wrap(my_function)

This wrapper is useful as a template for creating more complex decorators or when you want to ensure function metadata is preserved.

scitex_decorators.xarray_fn(func)

Return type:

Callable