Tips and Tricks

Best practices with data

It is strongly advised to setup your input/output data files hierarchy relative to a root folder, so that when you deploy your application, experts can simply specify a different root data path and things will work out great. For instance:

import os
from onecode import file_input, Project, Mode

project = Project()
project.mode = Mode.EXECUTE

file =  os.path.join('region', 'model.h5')
x = file_input('x', file)

will automatically append the path region/model.h5 to your Project data path.

print(f'Are path equals? {x == os.path.join(project.data_root, file)}')

Output

Are path equals? True

From there, any experts can have a data folder with the same hierarchy anywhere on their disk and launch the script by simply changing the data path at runtime:

ONECODE_PROJECT_DATA=/path/to/my/data python main.py

Without changing the code, the OneCode project can execute properly on different machines.

On the other hand, as soon as you specify an absolute path for your input files, the Project data path is ignored. There could be some special cases for doing that, but most of the time you should use relative paths.

Note

Input elements such as csv_reader and file_input make the path relative to Project().data_root.

Output elements such as csv_output, file_output, image_output and text_output make the path relative to {Project().data_root}/{Project.current_flow}/outputs.

Tip

How is the Project data path determined? The data path is initialized according to the following rules ordered by priority:

1. to ONECODE_PROJECT_DATA if provided in the Environment variables
2. to the data folder located in the same directory from where the project is run if existing (typically the OneCode project data folder)
3. to the current working directory for all other cases

Using runtime Expressions in elements

For Streamlit mode, some input element parameters can be dynamic, like for instance, changing the number of file_input widget based on a slider value that is changed on the fly by an expert.

All input elements can accept runtime expressions regarding the count and optional parameters. Expressions are in Python and may refer to defined elements values by wrapping around their key a $ sign. For instance, getting one more checkbox instances than an input number:

x = number_input('x', 1, min=0)
y = checkbox('y', [False, False], count='$x$ + 1')

Or disabling a file_input if the value of a slider is greater than 0.5:

x = slider('x', 0.4)
y = file_input('y', None, optional='$x$ > 0.5')

Some elements may have specific parameters that can accept expressions, like dropdown. For instance a dropdown can propose a selection from the columns of a selected CSV file:

df = csv_reader("csv", "/path/to/file.csv")

widget = dropdown(
    key="Dynamic Dropdown",
    value=None,
    options='$csv$.columns',
    optional=True
)

For a real use-case example, look at the DeepLearning project:

• the columns in the dropdown menu are read from the CSV input file
• the number of neural net dense layers adjust based on a number input field.