Utilities
Submodules
kale.utils.distance module
Provide distance (i.e., similarity) calculation methods using various distance metrics.
- kale.utils.distance.calculate_distance(x1: Tensor, x2: Optional[Tensor] = None, eps: float = 1e-08, metric: DistanceMetric = DistanceMetric.COSINE) Tensor
Returns similarity between \(x_1\) and \(x_2\), computed along `dim`=1. This method calculates the similarity between each pair of data points in two input matrices.
Note that this implementation differs from the existing implementations in PyTorch, as they calculate the similarity between each row of one matrix with its corresponding row in the other matrix (i.e., pairwise distance between columns of input matrices).
- Parameters:
x1 (torch.Tensor) – The tensor input data.
x2 (torch.Tensor, optional) – The tensor input data. (default
None)eps (float, optional) – Small value to avoid division by zero. (default: 1e-8)
metric (DistanceMetric, optional) – The metric to compute distance between input matrices. (default:
DistanceMetric.COSINE)Returns –
torch.Tensor – The computed similarity tensor between \(x_1\) and \(x_2\).
kale.utils.download module
Data downloading and compressed data extraction functions, Based on https://github.com/pytorch/vision/blob/master/torchvision/datasets/utils.py https://github.com/pytorch/pytorch/blob/master/torch/hub.py
- kale.utils.download.download_file_by_url(url, output_directory, output_file_name, file_format=None)
Download file/compressed file by url.
- Parameters:
url (string) – URL of the object to download
output_directory (string, optional) – Full path where object will be saved Abosolute path recommended. Relative path also works.
output_file_name (string, optional) – File name which object will be saved as
file_format (string, optional) – File format For compressed file, support [“tar.xz”, “tar”, “tar.gz”, “tgz”, “gz”, “zip”]
- Example: (Grab the raw link from GitHub. Notice that using “raw” in the URL.)
>>> url = "https://github.com/pykale/data/raw/main/videos/video_test_data/ADL/annotations/labels_train_test/adl_P_04_train.pkl" >>> download_file_by_url(url, "data", "a.pkl", "pkl")
>>> url = "https://github.com/pykale/data/raw/main/videos/video_test_data.zip" >>> download_file_by_url(url, "data", "video_test_data.zip", "zip")
- kale.utils.download.download_file_gdrive(id, output_directory, output_file_name, file_format=None)
Download file/compressed file by Google Drive id.
- Parameters:
id (string) – Google Drive file id of the object to download
output_directory (string, optional) – Full path where object will be saved Abosolute path recommended. Relative path also works.
output_file_name (string, optional) – File name which object will be saved as
file_format (string, optional) – File format For compressed file, support [“tar.xz”, “tar”, “tar.gz”, “tgz”, “gz”, “zip”]
Example
>>> gdrive_id = "1U4D23R8u8MJX9KVKb92bZZX-tbpKWtga" >>> download_file_gdrive(gdrive_id, "data", "demo_datasets.zip", "zip")
>>> gdrive_id = "1SV7fmAnWj-6AU9X5BGOrvGMoh2Gu9Nih" >>> download_file_gdrive(gdrive_id, "data", "dummy_data.csv", "csv")
kale.utils.initialize_nn module
Provide methods for initializing neural network parameters (i.e., weights and biases).
- kale.utils.initialize_nn.xavier_init(module) None
Fills the weight of the input Tensor with values using a normal distribution.
- Parameters:
module (torch.Tensor) – The input module.
- kale.utils.initialize_nn.bias_init(module) None
Fills the bias of the input Tensor with zeros.
- Parameters:
module (torch.Tensor) – The input module.
kale.utils.logger module
Logging functions, based on https://github.com/HaozhiQi/ISONet/blob/master/isonet/utils/logger.py
- kale.utils.logger.out_file_core()
Creates an output file name concatenating a formatted date and uuid, but without an extension.
- Returns:
A string to be used in a file name.
- Return type:
string
- kale.utils.logger.construct_logger(name, save_dir, log_to_terminal=False)
Constructs a logger. Saves the output as a text file at a specified path. Also saves the output of git diff HEAD to the same folder. Takes option to log to terminal, which will print logging statements. Default is False.
The logger is configured to output messages at the DEBUG level, and it saves the output as a text file with a name based on the current timestamp and the specified name. It also saves the output of git diff HEAD to a file with the same name and the extension .gitdiff.patch.
- Parameters:
name (str) – The name of the logger, typically the name of the method being logged.
save_dir (str) – The directory where the log file and git diff file will be saved.
log_to_terminal (bool, optional) – Whether to also log messages to the terminal. Defaults to False.
- Returns:
The constructed logger.
- Return type:
logging.Logger
- Raises:
None. –
kale.utils.print module
Screen printing functions, from https://github.com/HaozhiQi/ISONet/blob/master/isonet/utils/misc.py
- kale.utils.print.tprint(*args)
Temporarily prints things on the screen so that it won’t be flooded
- kale.utils.print.pprint(*args)
Permanently prints things on the screen to have all info displayed
- kale.utils.print.pprint_without_newline(*args)
Permanently prints things on the screen, separated by space rather than newline
kale.utils.seed module
Setting seed for reproducibility
- kale.utils.seed.set_seed(seed=1000)
Sets the seed for generating random numbers to get (as) reproducible (as possible) results.
The CuDNN options are set according to the official PyTorch guidance on reproducibility: https://pytorch.org/docs/stable/notes/randomness.html. Another references are https://discuss.pytorch.org/t/difference-between-torch-manual-seed-and-torch-cuda-manual-seed/13848/6 https://pytorch.org/docs/stable/cuda.html#torch.cuda.manual_seed https://github.com/open-mmlab/mmcv/blob/master/mmcv/runner/utils.py#L58
- Parameters:
seed (int, optional) – The desired seed. Defaults to 1000.