Your Browser is Out of Date

Nytro.ai uses technology that works best in other browsers.
For a full experience use one of the browsers below
Sentiment Analysis using Dell APEX File Storage for AWS and Amazon SageMaker
Executive summary Solution Architecture Solution Overview Solution Validation
Solution Validation
Step 1: Amazon SageMaker Studio
Step 2: Setup and import the IMDB dataset
Step 3: Create the Training, Testing, and Validation sets
Step 4: Select the BERT preprocessing module
Step 5: Pass data to the preprocessing module and BERT
Step 6: Pass data to the preprocessing module and BERT
Step 7: Training and Evaluation
Step 8: Saving and Re-loading the model.
Step 9: Predictions
Conclusion References

Step 7: Training and Evaluation

Home > APEX > Storage > White Papers > Sentiment Analysis using Dell APEX File Storage for AWS and Amazon SageMaker > Step 7: Training and Evaluation

Step 7: Training and Evaluation

  • Now we can proceed with training our model for three epochs and define some hyperparameters.

    For fine-tuning, we use the same optimizer that BERT was originally trained with, known as Adaptive Moment Estimation (ADAM). This optimizer minimizes the prediction loss and completes regularization by weight decay (not using moments). This is also known as AdamW.

    Training model with https://tfhub.dev/tensorflow/small_bert/bert_en_uncased_L-2_H-128_A-2/1

    Epoch 1/3

    625/625 [==============================] - 355s 558ms/step - loss: 0.6109 - binary_accuracy: 0.6386 - val_loss: 0.5041 - val_binary_accuracy: 0.7484

    Epoch 2/3

    625/625 [==============================] - 345s 552ms/step - loss: 0.4827 - binary_accuracy: 0.7576 - val_loss: 0.4535 - val_binary_accuracy: 0.7768

    Epoch 3/3

    625/625 [==============================] - 341s 546ms/step - loss: 0.4459 - binary_accuracy: 0.7831 - val_loss: 0.4527 - val_binary_accuracy: 0.7708

    The following output displays how the model performs. Two values are returned. Loss (a number which represents the error, where lower values are better), and accuracy.

    In the above plot, the red lines represent the training loss and accuracy, and the blue lines are the validation loss and accuracy. Both lines converge as expected.