# Loading all the libraries library(keras) library(tensorflow) library(tfdatasets) library(stringr) library(dplyr) # We pick a random .wav file and decode it using tf$audio$decode_wav. # This will give us access to two tensors: the samples themselves, and the sampling rate. audio <- tf$contrib$framework$python$ops$audio_ops files <- fs::dir_ls(path = "../input/train_curated/", recursive = TRUE, glob = "*.wav") cast_variables <- function(df, variable){ x <- as.character(unique(df[[variable]])) x <- gsub(" ", "", toString(x)) #so it can split on strings like "A1,A2" and "A1, A2" x <- unlist(strsplit(x, ",")) x <- as.character(x) new_columns <- unique(sort(x)) for (i in seq_along(new_columns)){ df$temp <- NA df$temp <- ifelse(grepl(new_columns[i], df[[variable]]), 1, 0) colnames(df)[colnames(df) == "temp"] <- new_columns[i] } return(df) } train <- read.csv('../input/train_curated.csv', stringsAsFactors = F) new_df <- cast_variables(train, 'labels') new_df$fname <- paste('../input/train_curated/',train$fname, sep = "") # Sett head(new_df) labels <- c('Accelerating_and_revving_and_vroom','Accordion','Acoustic_guitar', 'Applause', 'Bark' ,'Bass_drum', 'Bass_guitar' ,'Bathtub_(filling_or_washing)', 'Bicycle_bell' ,'Burping_and_eructation', 'Bus' ,'Buzz' , 'Car_passing_by' ,'Cheering' ,'Chewing_and_mastication', 'Child_speech_and_kid_speaking' , 'Chink_and_clink', 'Chirp_and_tweet', 'Church_bell', 'Clapping', 'Computer_keyboard', 'Crackle' ,'Cricket', 'Crowd', 'Cupboard_open_or_close', 'Cutlery_and_silverware' ,'Dishes_and_pots_and_pans', 'Drawer_open_or_close', 'Drip' ,'Electric_guitar', 'Fart' ,'Female_singing', 'Female_speech_and_woman_speaking', 'Fill_(with_liquid)', 'Finger_snapping', 'Frying_(food)' ,'Gasp' , 'Glockenspiel', 'Gong', 'Gurgling', 'Harmonica', 'Hi-hat', 'Hiss' ,'Keys_jangling', 'Knock', 'Male_singing', 'Male_speech_and_man_speaking', 'Marimba_and_xylophone', 'Mechanical_fan', 'Meow', 'Microwave_oven', 'Motorcycle' ,'Printer' ,'Purr', 'Race_car_and_auto_racing' ,'Raindrop', 'Run' , 'Scissors', 'Screaming', 'Shatter', 'Sigh', 'Sink_(filling_or_washing)', 'Skateboard' ,'Slam' ,'Sneeze', 'Squeak', 'Stream', 'Strum', 'Tap' ,'Tick-tock', 'Toilet_flush', 'Traffic_noise_and_roadway_noise', 'Trickle_and_dribble', 'Walk_and_footsteps', 'Water_tap_and_faucet' ,'Waves_and_surf' ,'Whispering', 'Writing' ,'Yell' ,'Zipper_(clothing)') batch_size <- 256 buffer_size <- nrow(new_df) window_size_ms <- 30 window_stride_ms <- 10 data_generator <- function(df, window_size_ms, window_stride_ms, shuffle = FALSE) { sampling_rate <- 16000 window_size <- as.integer(sampling_rate*window_size_ms/1000) stride <- as.integer(sampling_rate*window_stride_ms/1000) ds <- tensor_slices_dataset(df) if(shuffle) ds <- ds %>% dataset_shuffle(buffer_size = 100) ds <- ds %>% dataset_map(function(obs) { # decoding wav files audio_binary <- tf$read_file(tf$reshape(obs$fname, shape = list())) wav <- audio$decode_wav(audio_binary, desired_channels = 1) # create the spectrogram spectrogram <- audio$audio_spectrogram( wav$audio, window_size = window_size, stride = stride, magnitude_squared = TRUE ) # Custom brightness brightness <- 100L mul = tf$multiply(spectrogram, brightness) # Normalize pixels min_const = tf$constant(255) minimum = tf$minimum(mul, min_const) # Expand dims so we get the proper shape expand_dims = tf$expand_dims(minimum, -1L) # Resize the spectrogram to input size of the model resize = tf$image$resize_bilinear(expand_dims, c(128L, 128L)) # Remove the trailing dimension squeeze = tf$squeeze(resize, 0L) response <- as.matrix(obs[labels]) list(squeeze, response) }) %>% dataset_repeat() ds <- ds %>% dataset_padded_batch(batch_size, list(shape(128, 128, 1), shape(80))) return(ds) } id_train <- sample(nrow(new_df), size = 0.7 * nrow(new_df)) n_train <- length(id_train) n_test <- nrow(new_df) - n_train ds_train <- data_generator(new_df[id_train, ], window_size_ms = window_size_ms, window_stride_ms = window_stride_ms) ds_val <- data_generator(new_df[-id_train, ], window_size_ms = window_size_ms, window_stride_ms = window_stride_ms) model <- keras_model_sequential() model %>% layer_conv_2d(input_shape = c(128, 128, 1), filters = 32, kernel_size = c(3,3), activation = 'relu') %>% layer_max_pooling_2d(pool_size = c(2, 2)) %>% layer_conv_2d(filters = 64, kernel_size = c(3,3), activation = 'relu') %>% layer_max_pooling_2d(pool_size = c(2, 2)) %>% layer_conv_2d(filters = 128, kernel_size = c(3,3), activation = 'relu') %>% layer_max_pooling_2d(pool_size = c(2, 2)) %>% layer_conv_2d(filters = 256, kernel_size = c(3,3), activation = 'relu') %>% layer_max_pooling_2d(pool_size = c(2, 2)) %>% layer_dropout(rate = 0.25) %>% layer_flatten() %>% layer_dense(units = 256, activation = 'relu') %>% layer_dropout(rate = 0.2) %>% layer_dense(units = 80, activation = 'softmax') model %>% compile( loss = loss_binary_crossentropy, optimizer = optimizer_adam(), metrics = c('accuracy') ) model %>% summary() early_stopping <- callback_early_stopping(monitor = 'val_loss', patience = 2) model %>% fit_generator( generator = ds_train, steps_per_epoch = 0.7*nrow(new_df)/batch_size, epochs = 50, verbose = 1, validation_data = ds_val, validation_steps = 0.3*nrow(new_df)/batch_size, callbacks = c(early_stopping) ) test <- read.csv('../input/sample_submission.csv', stringsAsFactors = F) cnames <- data.frame(fname = test$fname) test$fname <- paste('../input/test/',test$fname, sep = "") colnames(test) <- c('fname', labels) ds_test <- data_generator(test, window_size_ms = window_size_ms, window_stride_ms = window_stride_ms) n_steps <- nrow(test)/batch_size +1 predictions <- predict_generator(object = model, ds_test, steps = n_steps ) predictions <- as.data.frame(predictions[1:1120,]) predictions <- cbind(cnames, predictions) colnames(predictions) <- c('fname', labels) head(predictions) write.csv(predictions, 'submission.csv', row.names=FALSE)