library(kerasR) imdb <- load_imdb(num_words = 500, maxlen = 100) X_train <- pad_sequences(imdb$X_train[1:4000], maxlen = 100) Y_train <- imdb$Y_train[1:4000] X_test <- pad_sequences(imdb$X_train[4001:5736], maxlen = 100) Y_test <- imdb$Y_train[4001:5736]
mod <- Sequential() mod$add(Embedding(500, 32, input_length = 100, input_shape = c(100))) mod$add(Dropout(0.25)) mod$add(Flatten()) mod$add(Dense(256)) mod$add(Dropout(0.25)) mod$add(Activation('relu')) mod$add(Dense(1)) mod$add(Activation('sigmoid'))
keras_compile(mod, loss = 'binary_crossentropy', optimizer = RMSprop(lr = 0.00025)) keras_fit(mod, X_train, Y_train, batch_size = 32, epochs = 10, verbose = 1, validation_split = 0.1)
Y_test_hat <- keras_predict(mod, X_test) table(Y_test, round(Y_test_hat)) mean(Y_test == as.numeric(round(Y_test_hat)))
mod <- Sequential() mod$add(Embedding(500, 32, input_length = 100, input_shape = c(100))) mod$add(Dropout(0.25)) mod$add(LSTM(32)) mod$add(Dense(256)) mod$add(Dropout(0.25)) mod$add(Activation('relu')) mod$add(Dense(1)) mod$add(Activation('sigmoid'))
keras_compile(mod, loss = 'binary_crossentropy', optimizer = RMSprop(lr = 0.00025)) keras_fit(mod, X_train, Y_train, batch_size = 32, epochs = 10, verbose = 1, validation_split = 0.1)
Y_test_hat <- keras_predict(mod, X_test) mean(Y_test == as.numeric(round(Y_test_hat)))
The first saves the entire model, which is more than likely what most users would want, as a binary file. The second saves only the weights as a binary file; the actual model architecture would have to be created again in R. Finally, the last saves just a json description of the model. This is probably most helpful because it gives a human-readable description of your model architecture.
keras_save(mod, "full_model.h5") keras_save_weights(mod, "weights_model.h5") keras_model_to_json(mod, "model_architecture.json")
mod <- keras_load("full_model.h5") keras_load_weights(mod, tf) mod <- keras_model_to_json("model_architecture.json")
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.