You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@mxnet.apache.org by GitBox <gi...@apache.org> on 2018/08/21 23:03:59 UTC

[GitHub] ankkhedia edited a comment on issue #12002: How do I properly dimensionalize my array and tune `rnn.graph.unroll` to make the LSTM work for this multidimensional sequence

ankkhedia edited a comment on issue #12002: How do I properly dimensionalize my array and tune `rnn.graph.unroll` to make the LSTM work for this multidimensional sequence
URL: https://github.com/apache/incubator-mxnet/issues/12002#issuecomment-414849456
 
 
   Hi @alexmosc 
   Please find a small example training on random multidimensional data. I trained a small RNN for 5 epochs. This is just an example with random values and I didn't check for accuracy but just wanted a demonstration of how training could be done.
   
   library("readr")
   library("dplyr")
   library("plotly")
   library("mxnet")
   seq_len = 100
   samples = 192
   
   seeds <- runif(samples, min = -pi, max = pi)
   pts <- sapply(seeds, function(x) sin(x + pi/12 * (0:(seq_len))))
   
   x <- pts[1:seq_len, ]
   x <- array(x, dim = c(3, seq_len, samples/3))
   y1 <- runif(32, 0.0, 1.0)
   y2 <- runif(32, 0.0, 1.0)
   batch.size = 8
   ( I have assumed 3 dimensional time series of seq_len 100 and having 64 samples )
   train.data <- mx.io.arrayiter(data = x[,,1:32, drop = F], label = y1, 
                                 batch.size = batch.size, shuffle = FALSE)
   
   eval.data <- mx.io.arrayiter(data = x[,,-(1:32), drop = F], label = y2, 
                                 batch.size = batch.size, shuffle = FALSE)
   
   ctx <- mx.gpu()
   
   initializer <- mx.init.Xavier(rnd_type = "gaussian", 
                                 factor_type = "avg", 
                                 magnitude = 3)
   
   optimizer <- mx.opt.create("adadelta", rho = 0.9, eps = 1e-5, wd = 1e-6,
                              clip_gradient = 1, rescale.grad = 1/batch.size)
   
   
   logger <- mx.metric.logger()
   epoch.end.callback <- mx.callback.log.train.metric(period = 1, logger = logger)
   batch.end.callback <- mx.callback.log.train.metric(period = 50)
   
   symbol<- rnn.graph(config = "seq-to-one", cell_type = "lstm", 
                               num_rnn_layer = 1, num_embed = NULL, num_hidden = 4, 
                               num_decode = 1, input_size = NULL, dropout = 0.5, 
                               ignore_label = -1, loss_output = "softmax",
                               output_last_state = F, masking = T)
   
   system.time(
     model <- mx.model.buckets(symbol = symbol,
                               train.data = train.data, 
                               eval.data = eval.data, 
                               num.round = 5, ctx = ctx, verbose = TRUE,
                               metric = mx.metric.accuracy, 
                               initializer = initializer, optimizer = optimizer, 
                               batch.end.callback = NULL, 
                               epoch.end.callback = epoch.end.callback)
   )
   
   
   Please let me know if this helps your use case.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services