Deep Learning In R
1 like10,849 views
This document provides an introduction to deep learning in R. It defines artificial intelligence, machine learning, and deep learning. Deep learning uses multi-layered neural networks to learn data representations without requiring manual feature extraction. It notes that large datasets, powerful computers/GPUs, and new techniques for handling gradients have enabled recent advances in deep learning. Several R libraries for deep learning are listed, including Keras. An example sentiment analysis using Keras is shown, demonstrating defining and training a model on movie reviews to predict sentiment.
![> predictions <- predict_classes(model, x_test)
> table(y_test, predictions)
0 1
0 10773 1727
1 2911 9589
> mean(y_test == predictions)
[1] 0.81448
Predicting Sentiment](https://image.slidesharecdn.com/deeplearning-180525193038/85/Deep-Learning-In-R-16-320.jpg)
Deep Learning In R
- 1. Whoa, That’s Deep Man An introduction to deep learning in R Martin Eastwood
- 2. ARTIFICIAL INTELLIGENCE Broad field of study dedicated to simulating human intelligence using machines MACHINE LEARNING Algorithms that can learn without being explicitly programmed DEEP LEARNING Use of multi-layered neural networks to learn data representations What is Deep learning?
- 3. Manual Feature Extraction / transformation ML Classifier algorithm It’s a cat! Neural network layer Neural network layer Neural network layer Neural network layer Machine Learning Vs Deep Learning Low-level features mid-level features higher-level features High-level features
- 4. No more feature engineering* * Maybe just a little bit…
- 5. But aren’t neural networks old, what’s changed? We now have much larger data sets for training models Commodity computers and GPUs are much more powerful New ways of handling exploding and vanishing gradients
- 8. Google Pixel Ear Buds
- 9. Deep Learning Libraries in R •MXNet •darch •deepnet •h20 •deepwater
- 11. > devtools::install_github("rstudio/keras") > library(keras) > install_keras()
- 12. Sentiment Analysis Using Keras Sentiment Review <chr> <chr> 1 I don't know why I like this movie so well, but I never get tired of watching it. 0 You'd better choose Paul Verhoeven's even if you have watched it. 1 This is the definitive movie version of Hamlet. 0 Long, boring, blasphemous. Never have I been so glad to see ending credits roll. 0 Ming The Merciless does a little Bardwork and a movie most foul! 0 I wouldn't rent this one even on dollar rental night.
- 13. > model <- keras_model_sequential() > model %>% layer_embedding(input_dim=max_features, output_dim=128) %>% layer_lstm(units=64, dropout=0.25, recurrent_dropout=0.25) %>% layer_dense(units=1, activation='sigmoid') Define the Keras Model
- 14. model %>% compile(loss=‘binary_crossentropy', optimizer='adam', metrics='accuracy') model %>% fit(x_train, y_train, batch_size=batch_size, epochs=10, validation_data=list(x_test, y_test)) Train the Keras Model
- 15. Training History
- 16. > predictions <- predict_classes(model, x_test) > table(y_test, predictions) 0 1 0 10773 1727 1 2911 9589 > mean(y_test == predictions) [1] 0.81448 Predicting Sentiment
- 17. • Keras now provides access to Tensorflow, CNTK and Theano through R • Can run models using either GPU / CPU • Lots of pre-trained Keras models available Conclusions
- 18. Questions