Experiment_utils

Helper utils for track and manage Dl experimets with pytorch.

Very early stage - just draft for my utils.

Install

pip install experiment-utils

Editeble install: git clone https://github.com/ayasyrev/experiment-utils cd experiment-utils pip install -e .

How to use

Import Experiment:

from experiment_utils.experiment import *

After import you has p (stands for Parameters) and e (Experiment) objects.

Name the experiment, later it will be used in logs.

p.exp_name = 'test1'
p.exp_name

'test1'

e.p.exp_name

'test1'

Load learner

e.get_learner()

e.p.model

functools.partial(<function resnet18 at 0x7fc8d4dd08c0>, num_classes=10)

e.learn.model.fc

Linear(in_features=512, out_features=10, bias=True)

Short notation for learn - l

e.l.model.conv1

Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)

Now we can easy change some parameter anr start train with pipeline what yuo neg.

from experiment_utils.utils import train_fc, plot

p.pipeline = [train_fc, plot]

p.lr = 0.001

p.epochs = 10

e(repeat_times=2)