TrainValSplits

Introduction

TrainValSplits widget in Supervisely is a tool that helps with the creation of training and validation datasets. The widget allows for easy splitting of the original dataset into training and validation sets based on customizable parameters such as percentage split or based on datasets or specific tag. TrainValSplits helps improve the performance of machine learning models by ensuring that they are trained on diverse and representative data.

Function signature

TrainValSplits(
    project_id=None,
    project_fs=None,
    random_splits=True,
    tags_splits=True,
    datasets_splits=True,
    widget_id=None,
    collections_splits=False,
)

default

Parameters

Parameters	Type	Description
project_id	int	Input Project ID
project_fs	str	Path to input Project on local host
random_splits	bool	Shuffle data and split with defined probability
tags_splits	bool	Images should have assigned train or val tag
datasets_splits	bool	Select one or several datasets for every split
widget_id	str	ID of the widget
collections_splits	bool	If True, allows to split data into train and validation sets based on selected collections within the project.

project_id

Determine input Project ID.

type: int

default value: None

splits = TrainValSplits(project_id=project_id)

project_fs

Determine path to input Project on local host.

type: str

default value: None

project_dir = os.path.join(sly.app.get_data_dir(), project_info.name)
sly.Project.download(api, project_id, project_dir)
project_fs = sly.Project(project_dir, sly.OpenMode.READ)

splits = TrainValSplits(project_fs=project_fs)

default

random_splits

Shuffle data and split with defined probability.

type: bool

default value: true

splits = TrainValSplits(
    project_id=project_id,
    random_splits=False,
)

random_splits

tags_splits

Images should have assigned train or val tag.

type: bool

default value: true

splits = TrainValSplits(
    project_id=project_id,
    tags_splits=False,
)

tags_splits

datasets_splits

Select one or several datasets for every split.

type: bool

default value: true

splits = TrainValSplits(
    project_id=project_id,
    datasets_splits=False,
)

datasets_splits

widget_id

ID of the widget.

type: str

default value: None

collections_splits

Allows to split data into train and validation sets based on the provided collections.

type: bool

default value: False

Methods and attributes

Attributes and Methods	Description
get_splits()	Return result train/val split.
disable()	Disable widget.
enable()	Enable widget.
set_project_id_for_collections()	Sets the project ID and updates available collections for train and validation splits.
get_train_collections_ids()	Returns a list of collection IDs selected for the training split.
get_val_collections_ids()	Returns a list of collection IDs selected for the validation split.
set_collections_splits()	Sets the collections (by their IDs) to be used for the training and validation splits when using the "collections" split method.
set_collections_splits_by_ids()	Set for the specified split (train or val) sets the provided collection IDs.

Mini App Example

You can find this example in our Github repository:

supervisely-ecosystem/ui-widgets-demos/controls/006_train_val_splits/src/main.py

Import libraries

import os

import supervisely as sly
from dotenv import load_dotenv
from supervisely.app.widgets import Card, Container, TrainValSplits, Button, Text

Init API client

First, we load environment variables with credentials and init API for communicating with Supervisely Instance:

load_dotenv("local.env")
load_dotenv(os.path.expanduser("~/supervisely.env"))

api = sly.Api()

Prepare project_id, download project and create project_fs

project_id = sly.env.project_id()
project_info = api.project.get_info_by_id(project_id)
project_dir = os.path.join(sly.app.get_data_dir(), project_info.name)
sly.fs.remove_dir(project_dir)
sly.Project.download(api, project_id, project_dir)
project_fs = sly.Project(project_dir, sly.OpenMode.READ)

Initialize TrainValSplits widget

splits = TrainValSplits(project_fs=project_fs)

Initialize Button and Text widget we will use

button = Button("Get splits")
text = Text("")
text.hide()

Create app layout

Prepare a layout for app using Card widget with the content parameter and place widget that we've just created in the Container widget.

card = Card(
    title="Train Val Splits",
    content=Container([splits, button, text], gap=5),
)
layout = Container(widgets=[card])

Create app using layout

Create an app object with layout parameter.

app = sly.Application(layout=layout)

Add function to control widgets from python code

@button.click
def get_splits():
    train_set, val_set = splits.get_splits()
    text.text = f"Train split: {len(train_set)} images, Val split: {len(val_set)} images"
    text.show()

layout