Skip to content

Latest commit

 

History

History
131 lines (93 loc) · 2.29 KB

File metadata and controls

131 lines (93 loc) · 2.29 KB

Config

Project

All global project configurations are stored in config/project.yaml.

Example:

sub_project: "v01"
cv_library: "opencv"
device: "auto"
device_ids: "auto"
logs:
  history_train_batches: True
  history_train_epochs: True
  history_validation: True
  notification: True
on_wake: Null

sub_project

Specify the name of your current session. Results from the current session, such as weights, logs and history will be saved into a sub-directory with this name.

  • Data type: str
  • Default value: "version01"

Example:

name: "version01"

cv_library

Specify the computer vision library with which to read and manipulate image or video files.

  • Data type: str
  • Default value: "pil"
  • Valid entries:
    • "pil"
    • "opencv"

Use "pil" to specify the Python Imagin Library (a.k.a. PIL/Pillow) or "opencv" to specify the Open Source Computer Vision Library (OpenCV). (Note that OpenCV is typically ~ 50% faster.)

Example:

cv_library: "opencv"

device

Specify the device to use.

  • Data type: str
  • Default value: "auto"
  • Valid entries:
    • "auto"
    • "cuda:0"
    • "cpu"

Example:

device: "cuda:0"

The CPU or a CUDA device can be specified explicitly with "cpu" or "cuda:0" respectively. Or, if "auto" is given, the device will be set to "cuda:0" if available, otherwise "cpu".

device_ids

If using multiple GPUs, you may specify the specific devices to use. Alternatively, use "auto" to automatically find and use all available CUDA-enabled GPUs.

  • Data type: list of ints
  • Default value: "auto"
  • Valid entries:

Example:

device_ids: [0, 1, 2]

logs

Specify whether or not to write each type of log file.

Example:

logs:
  history_train_batches: True
  history_train_epochs: True
  history_validation: True
  notification: True
  • history_train_batches - training history for each batch.
  • history_train_epochs - training history for each epoch.
  • history_validation - validation history.
  • notification - log all inputs and outputs of the deeplodocus interface.

on_wake

Specify any start-up commands.

Example:

on_wake:
  - config.summary()
  - load()
  - train()
  - sleep()

Model

Optimizer

Metrics

Training

Data

Transform

History

Losses