Train Networks Using Deep Network Designer
The Deep Network
Designer app lets you build and train deep neural networks. Deep Network Designer
trainNetwork training using image data or
datastore objects. You can also export your untrained network for training at the command
line, for example, to train your network using custom training loops.
To train a network, follow these steps:
Select training options
You can build a network interactively using Deep Network Designer, or import a network from the workspace. You can also select a pretrained network from the Deep Network Designer start page for transfer learning. For more information, see Build Networks with Deep Network Designer.
To train a deep learning model, you must have a suitable network and training data. To
import image data from a folder containing a subfolder of images for each class, or from an
imageDatastore object, on the
Data tab, click Import Data > Import Image Classification Data. To import any datastore, on the Data tab, click Import Data > Import Custom Data. After import, Deep Network Designer displays a preview of the imported data
so that you can check that the data is as expected prior to training. For more information,
see Import Data into Deep Network Designer.
Select Training Options
Once you have your network and data, the next step is to select the training options.
On the Training tab, click Training Options.
If you do not know which training options to use, try training with the default settings
and then adjusting them to suit your network and data. For example, try adjusting the
initial learning rate, or train for longer by increasing the number of epochs. For
information about techniques for improving the accuracy of deep learning networks, see
Deep Learning Tips and Tricks. For more information
about the training options, see
After you select your training options, train the network by clicking Train. The Deep Network Designer app displays an animated plot of the training progress. The plot shows mini-batch loss and accuracy and additional information on the training progress. If you specified validation data, the plot also shows the validation loss and accuracy. The plot has a stop button in the top-right corner. Click the button to stop training and return the current state of the network. For more information on the training progress plot, see Monitor Deep Learning Training Progress.
You can train a variety of networks using Deep Network Designer. For example, image
classification or regression networks, sequence networks, numeric data networks,
semantic segmentation networks, and image-to-image regression networks. In Deep Network
Designer, you can train a network using the
on any data that you can express as a datastore object. The following examples show how
to build and train a network using Deep Network Designer.
Once training is complete, on the Training tab, click Export to export your trained network and results to the workspace. To save the training progress plot as an image, click Export Training Plot. You can learn how to build and train your network using command line functions by clicking Export > Generate Code for Training and examining the generated live script.
Deep Network Designer does not support training using custom training loops. To train
your network using a custom training loop, first export the network to the workspace and
convert it to a
dlnetwork object. You can then train
the network using the
dlnetwork object and a custom training loop. For
more information, see Train Network Using Custom Training Loop.
Once training is complete, click Export > Create Experiment to create a deep learning experiment in Experiment Manager. You can use Experiment Manager to sweep through a range of hyperparameter values or use Bayesian optimization to find optimal training options. For an example showing how to use Experiment Manager to tune the hyperparameters of a network trained in Deep Network Designer, see Generate Experiment Using Deep Network Designer.