domainlab.exp package

Submodules

domainlab.exp.exp_cuda_seed module

Random seed should be set from command line to ensure reproducibility: https://pytorch.org/docs/stable/notes/randomness.html https://discuss.pytorch.org/t/difference-between-torch-manual-seed-and-torch-cuda-manual-seed/13848/6

domainlab.exp.exp_cuda_seed.set_seed(seed)

domainlab.exp.exp_main module

experiment

class domainlab.exp.exp_main.Exp(args, task=None, model=None, observer=None, visitor=<class 'domainlab.exp.exp_utils.AggWriter'>)

Bases: object

Exp is combination of Task, Algorithm, and Configuration (including random seed)

clean_up()

to be called by a decorator

execute(num_epochs=None)

train model check performance by loading persisted model

domainlab.exp.exp_utils module

This module contains 3 classes inheriting:

ExpProtocolAggWriter(AggWriter(ExpModelPersistVisitor))

class domainlab.exp.exp_utils.AggWriter(host)

Bases: ExpModelPersistVisitor

1. aggregate results to text file.

2. all dependencies are in the constructor

confmat_to_file(confmat, confmat_filename)

Save confusion matrix as a figure

Parameters:

confmat – confusion matrix.

first_line(dict_cols)

generate header of the results aggregation file

get_cols()

call the same function to always get the same columns configuration

get_fpath(dirname='aggrsts')

for writing and reading, the same function is called to ensure name change in the future will not break the software

to_file(str_line)

Parameters:

str_line –

class domainlab.exp.exp_utils.ExpModelPersistVisitor(host)

Bases: object

This class couples with Task class attributes

clean_up()

load(suffix=None)

load pre-defined model name from disk the save function is the same class so to ensure load will ways work

mk_model_na(tag=None, dd_cut=19)

Parameters:

tag – for git commit hash for example

model_dir = 'saved_models'

model_suffix = '.model'

remove(suffix=None)

remove model after use

save(model, suffix=None)

Parameters:

model –

class domainlab.exp.exp_utils.ExpProtocolAggWriter(host)

Bases: AggWriter

AggWriter tailored to experimental protocol Output contains additionally index, exp task, te_d and params.

confmat_to_file(confmat, confmat_filename)

Save confusion matrix as a figure

Parameters:

confmat – confusion matrix.

get_cols()

columns

get_fpath(dirname=None)

filepath

Module contents