Welcome to apu’s documentation!¶
apu Time¶
apu.datetime: anton python utils datetime module
-
class
apu.time.
DateTime
¶ Datetime contains the datetime utility functions
-
static
add_time
(datetime_: datetime.datetime, day: int = 0, hour: int = 0, minute: int = 0, seconds: int = 0) → datetime.datetime¶ - add time to a datetime object with respect to the
timezone
This function will keep the correct timezone
- Parameters
datetime (datetime.datetime) – datetime object
day (int) – days
hour (int) – hours
seconds (int) – seconds
- Returns
time in the correct timezone
- Return type
(datetime.datetime)
-
static
datetime2unix
(datetime_: datetime.datetime) → float¶ - convert a datetime object into
a unixtimestamp
- Parameters
datetime (dt.datetime) – datetime
- Returns
unix timestamp
- Return type
(float)
Examples: .. example_code:
>>> from apu.time.date_time import DateTime >>> DateTime.unixtime2date(datetime.now()) 1234567890.123
-
static
generate
(start: datetime.datetime, end: datetime.datetime, random_callback=<class 'random.Random'>) → datetime.datetime¶ generate a random date.
- Parameters
start (datetime.datetime) – start time
end (datetime.datetime) – end time
random_callback (random.Random) – random number generator
- Returns
datetime
- Return type
(datetime.datetime)
- Raises
ValueError – the start end oder is wrong
Examples: .. example_code:
>>> import random >>> import datetime as dt >>> from apu.time.date_time import DateTime >>> rand = random.Random() >>> rand.seed(0) >>> print(DateTime.generate(start=dt.datetime(2020,1,2), >>> end=dt.datetime(2020,2,3), >>> random_callback=rand)) 2020-01-29 00:30:57.535096
-
static
time_string
(datetime: datetime.datetime = datetime.datetime(2021, 2, 12, 18, 11, 14, 789860), form: str = '%y%m%d_%H%M%S') → str¶ create string date.
- Parameters
datetime (datetime.datetime) – date
form (str) – format string
- Returns
date as string
- Return type
(str)
Examples: .. example_code:
>>> import datetime as dt >>> from apu.time.date_time import DateTime >>> print(DateTime.time_string( ... datetime=dt.datetime(2020,1,2,6,4,53), ... form="%y%m%d_%H%M%S")) 200102_060453
-
static
unixtime2date
(unix: float) → datetime.datetime¶ convert a uniy timestamp to datetime object
- Parameters
unix (float) – unix timestamp
- Returns
datetime
- Return type
(datetime.datetime)
Examples: .. example_code:
>>> from apu.time.date_time import DateTime >>> DateTime.unixtime2date(1234567890.123) 2009-02-13 23:31:30.123000
-
static
-
apu.time.
time_it
(duration='s')¶ generate a random date.
- Parameters
start (datetime.datetime) – start time
end (datetime.datetime) – end time
random_callback (random.Random) – random number generator
- Returns
datetime
- Return type
(datetime.datetime)
- Raises
ValueError – the start end oder is wrong
Examples: .. example_code:
>>> import math >>> from apu.time.timer import time_it >>> @time_it() >>> def test(): >>> math.sqrt(1231) >>> test() Time taken by the function is [0.0003693103790283203] sec
apu Geography¶
apu.datetime: anton python utils geography module
-
apu.geo.
carree2pix
(coord: List[float], area: List[Tuple[float]], image_size: List[int], origin: str = 'upper') → tuple¶ Convert Carree lat/long to image poxel position
- Parameters
coord (List[float]) – (lat, lon) coordinate in image
area (List[Tuple[float]]) – ((u,v)_min, (u,v)_max) the area in the image to search in
image_size (List[int]) – image size in width and height
origin (str) – image convention. where is the image origin. ‘upper’ means the origin [0,0] is in the upper left corner ‘lower’ means that the image origin is in the lower left corner
- Returns
(u, v) pixel coordinate
- Return type
tuple
-
apu.geo.
km2pix
(height: float, extention: float, radius: float = 6378.137, distorsion_scaling: float = 1.0) → float¶ convert from kilometer in pixel
- Parameters
height (float) – image height
extention (float) – latitude extention of image in pixel
radius (float) – planet radius in kilometer default is the earth radius (6_378.137km)
distorsion_scaling (float) – Scaling factor for distortion between 0. and 1.
- Returns
conversionsfactor pix/m
- Return type
float
-
apu.geo.
m2pix
(height: float, extention: float, radius: float, distorsion_scaling: float = 1.0) → float¶ convert from meter in pixel
- Parameters
height (float) – image height
extention (float) – latitude extention of image in pixel
radius (float) – planet radius in metern
distorsion_scaling (float) – Scaling factor for distortion between 0. and 1.
- Returns
conversionsfactor pix/m
- Return type
float
-
apu.geo.
pix2carree
(pixel: List[float], area: List[Tuple[float]], image_size: List[int], origin: str = 'upper') → tuple¶ - Convert image pixel position to Carree lat/long
ASSAMTION: central median is 0 => (long [-180,180[)
- Parameters
pixel (List[float]) – (u,v) coordinate in image
area (List[Tuple[float]]) – ((u,v)_min, (u,v)_max) the area in the image to search in
image_size (List[int]) – image size in width and height
origin (str) – image convention. where is the image origin. ‘upper’ means the origin [0,0] is in the upper left corner ‘lower’ means that the image origin is in the lower left corner
- Returns
(lat, lon) coordinates in a Plate Carree image
- Return type
tuple
apu Setup¶
apu.setup: anton python utils setup module
-
class
apu.setup.
BuildProtoBuf
(dist)¶ find and execute protobuf related work
-
run
()¶ run the proto_builder
-
-
class
apu.setup.
CleanProtoBuf
(dist)¶ clean protobuf info
-
run
()¶ Delete generated files in the code tree.
-
-
class
apu.setup.
Module
¶ in this class all module containing functions are placed.
-
static
check
(module_name: str)¶ check if a module exists without importing it
- Parameters
module_name (str) – module name
- Returns
Module specification
- Return type
(ModuleSpec)
- Raises
ModuleNotFoundError – Module not found
AttributeError – given attribute error
-
static
import_module_from_spec
(module_name: str)¶ If the module is found utilizing check_module
- Parameters
module_name (str) – name of the module to import
- Returns
imported module
- Return type
(module)
- Raises
ModuleNotImportedError – cannot import module
ModuleNotFoundError – cannot find the module. install the module first
AttributeError – given attribute for the module check is not valid
-
static
-
apu.setup.
find_protoc
()¶ Locates protoc executable
-
apu.setup.
replace_line_in_file
(fpath, old_line_start, new_line)¶ find and delete string in a file
-
apu.setup.
setversion
(repopath, versionfile)¶ set the version number in a given file
apu Multiprocessing¶
apu.mp: anton python utils multiprocessing module
-
apu.mp.
parallel_for
(loop_callback: Callable[[Any], Any], parameters: List[Tuple[Any, …]], nb_threads: int = 8) → List[Any]¶ Execute a for loop body in parallel .. note:: Race-Conditions
Code executation in parallel can cause into an “race-condition” error.
- Parameters
loop_callback (Callable) – function callback running in the loop body
parameters (List[Tuple]) – element to execute in parallel
- Returns
list of values
- Return type
(List[Any])
Examples: .. example-code:
>>> x = lambda x: x ** 2 >>> parallel_for(x, [y for y in range(10)]) [0, 1, 4, 9, 16, 25, 36, 49, 64, 81]
-
apu.mp.
thread_funcrun
(func)¶ run function in thread
Examples: .. example_code:
>>> from apu.mp.thread_funcrun import thread_funcrun >>> @thread_funcrun ... def test(*args, **kwargs): ... for i in range(5): ... print(f'elem: {i}') elem: 0 elem: 1 elem: 2 elem: 3 elem: 4 Thread started for function "test"
-
apu.mp.
thread_n_funcrun
(number_of_threads=1)¶ run function in multiple threads
Examples: .. example_code:
>>> from apu.mp.thread_funcrun import thread_funcrun >>> @thread_n_funcrun(number_of_threads=3) ... def test(*args, **kwargs): ... pass Thread started for function "test" Thread started for function "test" Thread started for function "test"
apu Design Patterns¶
apu.dp: anton python utils design pattern module
-
class
apu.dp.
AlphabeticalOrderCollection
(collection: List[Any] = ())¶ Collection handling a collection with iterator
Examples: .. example-code:
>>> liste = AlphabeticalOrderCollection() >>> liste.add_item(0) >>> liste.add_item(1) >>> liste.add_item(2) >>> print("
- “.join(liste))
0 1 2 >>> print(“
- “.join(liste.get_reverse_iterator()), end=””)
2 1 0
-
add_item
(item: Any) → None¶ add an element to the collection
-
item
¶ an item for the collection
- Type
Any
-
-
get_reverse_iterator
() → apu.dp.iterator.AlphabeticalOrderIterator¶ return an iterator object
- Returns
the iterator in reversed order
- Return type
-
class
apu.dp.
AlphabeticalOrderIterator
(collection: List[Any] = (), reverse_order: bool = False)¶ Alphabetical ordered collection. this class stores the traversal positon at all time
-
class
apu.dp.
Blackboard
¶ a shared memory structure to handover information in different objects TODO: Memory info
-
clear
()¶ clean the memory for all reference keys
-
clear_callbacks
(key: str)¶ clear the callback
- Parameters
key (str) – reference key
- Raises
NonExistingKey – key does not exists
-
close
()¶ delete the meta information and close the memory
-
drop
(key: str) → bool¶ drop memory with pointer
- Parameters
key (str) – pointer to memory
- Returns
Ture if the drop was successful
- Return type
(bool)
- Raises
NonExistingKey – key to memory is unknown
-
get
(key: str) → Any¶ get the memory
- Parameters
key (str) – reference key
- Returns
memory value
- Return type
(Any)
- Raises
NonExistingKey – key does not exists
-
keys
(in_list: bool = False) → List[str]¶ get a list of all reference keys
- Parameters
in_list (bool) – as a python list
- Returns
an iterator or python list object
- Return type
List[str]
-
load
(dir_path: str = './.blackboard', safe: bool = True)¶ load the data from blackboard file
- Parameters
dir_path (str) – filepath
safe (bool) – load safe
- Raises
UnsafeLoading – unsafe loading
NonExistingDirectory – path does not exists
-
register_callback
(key: str, callback: Callable[[Any], Any]) → int¶ register a callback on the memory values
- Parameters
key (str) – reference key
callback (Callable[[Any], Any]) – callback function
- Returns
python id Object
- Return type
int
- Raises
NonExistingKey – the key does not exists
-
remove_callback
(key: str, callback: Callable[[Any], Any]) → int¶ remove the callback from the memory
- Parameters
key (str) – reference key
callback (Callable[[Any], Any]) – delete this callback
- Returns
python object id
- Return type
int
- Raises
NonExistingKey – key does not exists
-
save
(dir_path: str = './.blackboard')¶ save the blackboard
- Parameters
dir_path (str) – file path
-
set
(key: str, value: Any, read_only: bool = False) → bool¶ set the memory related to the reference key
- Parameters
key (str) – reference
value (Any) – memory value
read_only (bool) – store the data read only?
- Returns
True if the memory could be placed
- Return type
(bool)
- Raises
KeyNotString – the reference key is not a string
ExistingKey – the reference key is already set
-
update
(key: str, value: Any) → bool¶ change the memory
- Parameters
key (str) – reference key
value (Any) – Memory value
- Returns
true if the data are places successfully
- Return type
(bool)
- Raises
NonExistingKey – key does not exists
NotEditable – the memory is read only
-
-
class
apu.dp.
MetaInfo
(read_only: bool = True)¶ additional information and data manipulation utilizing callbacks
-
property
callback
¶ get the callbacks
- Returns
list of all callbacks
- Return type
(List[Callable[[Any], Any]])
-
clean
()¶ delete all callbacks
-
del_callback
(callback: Callable[[Any], Any])¶ get one callback if the callback is in the list
- Parameters
callback (Callable[[Any], Any]) – callable structure
-
property
-
class
apu.dp.
Null
(*args, **kargs)¶ Null Object implementation
This class ignores all parameters parsed to the object on creation. The created instances do nothing and mark an object nor nothing
-
class
apu.dp.
Singleton
¶ singleton metaclass
Examples: .. example_code:
>>> from apu.dp.singleton import Singleton >>> class Duck(metaclass=Singleton): >>> pass >>> Duck() is Duck() True
-
apu.dp.
singleton
(cls)¶ singleton declerator
Examples: .. example_code:
>>> from apu.dp.singleton import singleton >>> @singleton >>> class Duck(): >>> pass >>> Duck() is Duck() True
apu Exceptions¶
apu.exception: anton python utils excaptions module
-
exception
apu.exception.
ExistingKey
¶ the key allready exists
-
exception
apu.exception.
KeyNotString
¶ key type is not String
-
exception
apu.exception.
ModuleNotImportedError
¶ Extends ModuleNotFoundError. Because a Module can also available but not imported
-
exception
apu.exception.
NonExistingDirectory
¶ The requested directory do not exists
-
exception
apu.exception.
NonExistingKey
¶ the key does not exists
-
exception
apu.exception.
NotEditable
¶ you try to edit a not editable object
-
exception
apu.exception.
ReadWrongFile
¶ read the wrong file error
-
exception
apu.exception.
UnsafeLoading
¶ unsafe loading. are you sure
-
exception
apu.exception.
UnsupportedDataType
¶ requested datatype is not supported
apu Input/Output¶
apu.io anton python utils input output module
-
class
apu.io.
FileFormat
(path: str, kwargs: Optional[Dict] = None, data: Optional[Any] = None)¶ base class. so each object implements the same functions
-
property
access_time
¶ date of last access
-
compair
(filepath: str, method='sha1')¶ compair two files utilizing the fingerprint
-
property
creation_time
¶ date of creation
-
fingerprint
(method: str = 'sha1')¶ build file fingerprint
-
meta
() → Any¶ aditional meta informations
-
property
modification_time
¶ date of modification
-
abstract
read
()¶ read information from file into data buffer
-
classmethod
suffix
()¶ file extentions
-
abstract
write
(sink: str, create: bool = True) → None¶ write buffer into file
-
property
-
class
apu.io.
Path
(*args, **kwargs)¶ path changes the type on different os
-
expand
()¶ expand all together and resolve all
-
expand_vars
()¶ expand vars
-
-
apu.io.
download
(source: str, sink: Optional[str] = None) → str¶ download from page
-
apu.io.
load
(data: str) → Any¶ serialize the data
- Parameters
data (str) – string to deserialize
- Returns
deserialized data
- Return type
(Any)
- Raises
UnsupportedDataType – the data are not deserializeable
-
apu.io.
reconstruct
(value: Any) → str¶ serialize the data
- Parameters
value (Any) – object to serialize
- Returns
serialized data
- Return type
(str)
- Raises
UnsupportedDataType – the data are not serializeable
-
apu.io.
urlread
(url: str, encoding: str = 'utf8') → str¶ read a page
file format input output realization with meta info
-
class
apu.io.__fileformat.
CSV
(path: str, kwargs: Optional[Dict] = None, data: Optional[Any] = None)¶ read write csv files
-
read
()¶ read a csv file
-
classmethod
suffix
()¶ file extentions
-
write
()¶ write a csv file
-
-
class
apu.io.__fileformat.
DILL
(path: str, kwargs: Optional[Dict] = None, data: Optional[Any] = None)¶ working with a dill file
-
read
()¶ read dill
-
classmethod
suffix
()¶ file extentions
-
write
()¶ “write dill
-
-
class
apu.io.__fileformat.
H5
(path: str, kwargs: Optional[Dict] = None, data: Optional[Any] = None)¶ h5 fileformat handler
-
read
()¶ read information from file into data buffer
-
recursively_load_dict_contents_from_group
(h5_file, path)¶ Load contents of an HDF5 group. If further groups are encountered, treat them like dicts and continue to load them recursively.
-
recursively_save_dict_contents_to_group
(h5file: h5py._hl.files.File, path: str, dic: dict)¶ save recursivly
-
classmethod
suffix
()¶ file extentions
-
write
()¶ write buffer into file
-
-
class
apu.io.__fileformat.
JSON
(path: str, kwargs: Optional[Dict] = None, data: Optional[Any] = None)¶ json file
-
read
()¶ read json file
-
classmethod
suffix
()¶ json suffix
-
write
()¶ write json files
-
-
class
apu.io.__fileformat.
JSONL
(path: str, kwargs: Optional[Dict] = None, data: Optional[Any] = None)¶ json line files
-
read
()¶ read json line files
-
classmethod
suffix
()¶ json line suffix
-
write
()¶ read json line
-
-
class
apu.io.__fileformat.
MAT
(path: str, kwargs: Optional[Dict] = None, data: Optional[Any] = None)¶ handle mat files
-
read
()¶ read mat file
-
classmethod
suffix
()¶ matlab suffix
-
write
()¶ write mat file
-
-
class
apu.io.__fileformat.
NPY
(path: str, kwargs: Optional[Dict] = None, data: Optional[Any] = None)¶ handle npy files
-
read
()¶ read npy files
-
classmethod
suffix
()¶ numpy suffix
-
write
()¶ write npy files
-
-
class
apu.io.__fileformat.
NPZ
(path: str, kwargs: Optional[Dict] = None, data: Optional[Any] = None)¶ handle npz files
-
classmethod
suffix
()¶ compressed numpy suffix
-
write
()¶ write npz files
-
classmethod
apu Datastructures¶
apu.datastructures anton python utils datastructures
-
class
apu.datastructures.
CircularBuffer
(size=1)¶ ringbuffer based on deque
-
property
average
¶ calculate the avarage of all the componentes in the ringbuffer
- Returns
the avarage
- Return type
float
-
property
-
class
apu.datastructures.
Dictionary
¶ shared Memory dictionary
-
property
all
¶ get the whole dict
- Returns
the shared memory dictionary
- Return type
(dict)
-
delete
(key: str) → None¶ delete a key value pair
- Parameters
key (str) – key pointing to memory position
-
exists
(key: str) → bool¶ - check if the key is in the list of
pointers in the memory
- Parameters
key (str) – pointer to value
- Returns
true if the key is a valid pointer
- Return type
(bool)
-
flush
()¶ clean the memory
-
get
(key: str) → Any¶ get an dictionary element by key
- Parameters
key (str) – key pointing to the memory related to it
- Returns
the value or a Null object
- Return type
(Any)
-
keys
() → List[str]¶ get all keys of the data storage
- Returns
return a list of strings
- Return type
(List[str])
-
set
(key: str, value: Any) → bool¶ set a key value pair into the dictionary
- Parameters
key (str) – key value has to be string
value (Any) – any value is allowed
- Returns
successful added
- Return type
(bool)
-
property
-
class
apu.datastructures.
DictionaryWrapper
(**kwargs)¶ Dictionary class wrapper class. This is used for using Dictionary as a memory.
-
close
()¶ close the memory and delete all items
-
delete
(key: str) → bool¶ delete the value related to the pointer
- Parameters
key (str) – point to memory
- Returns
successfully deleted
- Return type
(bool)
-
get
(key: str) → Any¶ get the memory utilizing a reference
- Parameters
key (str) – key to memory
- Returns
memory or Null-Object
- Return type
(Any)
-
has
(key: str) → bool¶ does points contain the key
- Parameters
key (str) – pointer wo memory
- Returns
True if the pointers contains the key
- Return type
(bool)
-
set
(key: str, value: Any) → bool¶ store data to memory
- Parameters
key (str) – pointer to memory
value (Any) – memory to store
- Returns
True if the memory are set successfull
- Return type
(bool)
-
setup
()¶ setup the Wrapper with a dictionary object
-
-
class
apu.datastructures.
EnhancedList
(*args: Iterable[T])¶ extends list of a gerneric type
-
reject_indices
(indices: List[int])¶ remove the elements utilizing a list
- Parameters
indices – List[int]
- Returns
EnhancedList
- Return type
list without rejected elements
-
-
class
apu.datastructures.
MemoryWrapper
(**kwargs)¶ wrap the memory and setup steps. future change to additional wrapper
-
abstract
close
()¶ set the Pointer to Null
-
abstract
delete
(key: str) → bool¶ delete the memory object
- Parameters
key (str) – pointer to memory object
- Returns
true if successfull deleted
- Return type
(bool)
-
abstract
get
(key: str) → Any¶ get the memory value by string
- Parameters
key (str) – pointer to memory object
- Returns
memory object
- Return type
(Any)
-
abstract
has
(key: str) → bool¶ has pointer to memory?
- Parameters
key (str) – request key
- Returns
True if the pointer is in memory
- Return type
(bool)
-
load
(file_path: str) → bool¶ load the data from file
- Parameters
file_path (str) – the path to the fail containing the memory
- Returns
True if the storage was successful
- Return type
(bool)
-
save
(file_path: str) → bool¶ store the memory to file
- Parameters
file_path (str) – the path to the fail containing the memory
- Returns
True if the storage was successful
- Return type
(bool)
-
abstract
set
(key: str, value: Any) → bool¶ set a key to the memory value :param key: string typed key :param value: Any memory object
- Returns
bool
- Return type
true if successfull stored
-
abstract
setup
()¶ memory setup
-
static
transform_pickle_to_value
(data: str) → Any¶ deserialize object
- Parameters
data (str) – serialized value
- Returns
deserialized Data
- Return type
(Any)
-
static
transform_value_to_pickle
(value: Any) → str¶ serialize an object
- Parameters
value (Any) – the memory
- Returns
serialized object
- Return type
(str)
-
abstract
apu machine Lerning¶
apu.ml: anton python utils machine learning
apu.ml.init: anton python utils machine learning initializers
-
apu.ml.init.
glorot_normal
(tensor: torch.Tensor, gain: float = 1.0) → torch.Tensor¶ Fills the input Tensor with values according to the method described in Understanding the difficulty of training deep feedforward neural networks - Glorot, X. & Bengio, Y. (2010), using a uniform distribution. The resulting tensor will have values sampled from \(\mathcal{U}(-a, a)\) where
\[a = \text{gain} \times \sqrt{\frac{6}{\text{fan\_in} + \text{fan\_out}}}\]Also known as Glorot initialization.
- Parameters
tensor – an n-dimensional torch.Tensor
gain – an optional scaling factor
Examples
>>> w = torch.empty(3, 5) >>> nn.init.xavier_uniform_(w, gain=nn.init.calculate_gain('relu'))
-
apu.ml.init.
glorot_unified
(tensor: torch.Tensor, gain: float = 1.0) → torch.Tensor¶ Fills the input Tensor with values according to the method described in Understanding the difficulty of training deep feedforward neural networks - Glorot, X. & Bengio, Y. (2010), using a normal distribution. The resulting tensor will have values sampled from \(\mathcal{N}(0, \text{std}^2)\) where
\[\text{std} = \text{gain} \times \sqrt{\frac{2}{\text{fan\_in} + \text{fan\_out}}}\]Also known as Glorot initialization.
- Parameters
tensor – an n-dimensional torch.Tensor
gain – an optional scaling factor
Examples
>>> w = torch.empty(3, 5) >>> nn.init.xavier_normal_(w)
-
apu.ml.init.
he_normal
(tensor, a=0, mode='fan_in', nonlinearity='leaky_relu')¶ Fills the input Tensor with values according to the method described in Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification - He, K. et al. (2015), using a normal distribution. The resulting tensor will have values sampled from \(\mathcal{N}(0, \text{std}^2)\) where
\[\text{std} = \frac{\text{gain}}{\sqrt{\text{fan\_mode}}}\]Also known as He initialization.
- Parameters
tensor – an n-dimensional torch.Tensor
a – the negative slope of the rectifier used after this layer (only used with
'leaky_relu'
)mode – either
'fan_in'
(default) or'fan_out'
. Choosing'fan_in'
preserves the magnitude of the variance of the weights in the forward pass. Choosing'fan_out'
preserves the magnitudes in the backwards pass.nonlinearity – the non-linear function (nn.functional name), recommended to use only with
'relu'
or'leaky_relu'
(default).
Examples
>>> w = torch.empty(3, 5) >>> nn.init.kaiming_normal_(w, mode='fan_out', nonlinearity='relu')
-
apu.ml.init.
he_uniform
(tensor, a=0, mode='fan_in', nonlinearity='leaky_relu')¶ Fills the input Tensor with values according to the method described in Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification - He, K. et al. (2015), using a uniform distribution. The resulting tensor will have values sampled from \(\mathcal{U}(-\text{bound}, \text{bound})\) where
\[\text{bound} = \text{gain} \times \sqrt{\frac{3}{\text{fan\_mode}}}\]Also known as He initialization.
- Parameters
tensor – an n-dimensional torch.Tensor
a – the negative slope of the rectifier used after this layer (only used with
'leaky_relu'
)mode – either
'fan_in'
(default) or'fan_out'
. Choosing'fan_in'
preserves the magnitude of the variance of the weights in the forward pass. Choosing'fan_out'
preserves the magnitudes in the backwards pass.nonlinearity – the non-linear function (nn.functional name), recommended to use only with
'relu'
or'leaky_relu'
(default).
Examples
>>> w = torch.empty(3, 5) >>> nn.init.kaiming_uniform_(w, mode='fan_in', nonlinearity='relu')