TensorBoard helps in visualizing and analyzing our TensorFlow based AI/ML model runs and graphs.
We will understand, how the initial setup is done, so that model visualization can be performed using
The first step is to create a log file for storing data for the purpose of TensorBoard visualization. In this case we are storing this in google storage
import datetime log_dir = "gs://ai-mod-6-pathology-dataset/AI-MODULE-6/logsdn" + datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir=log_dir)
tf.keras.callback specifies where the logs should be stored
callbacks_list = [lr_schedule,tensorboard_callback]
history = model.fit(train_dataset, epochs=EPOCHS, callbacks=callbacks_list, steps_per_epoch=STEPS_PER_EPOCH, validation_data=valid_dataset)
Callback is then passed to model.fit during the model training. After training the model with the above code, the logs will be available in the designated folder, namely logsdn20230528-094952
After transferring the data from google storage to the local system, we can launch TensorBoard with
when we launch the tensorboard, a web link is provided as
copy paste this in a browser to visualize the results.
Categorical accuracy and loss of the DenseNet model are shown here. From the picture it is clear that there are more analysis options with TensorBoard such as Training accuracy/loss and Validation accuracy/loss, learning rate. If there is bias in the model, it can be quickly identified and corrective measures can be taken.
DenseNet model accuracy and loss - TensorBoard
EfficientNet model accuracy and loss - TensorBoard
Additional Reference Material