[GitHub] hetong007 commented on a change in pull request #12670: [MXNET-833] [R] Char-level RNN tutorial fix

[GitBox <gi...@apache.org>]

hetong007 commented on a change in pull request #12670: [MXNET-833] [R] Char-level RNN tutorial fix
URL: https://github.com/apache/incubator-mxnet/pull/12670#discussion_r220741246
 
 

 ##########
 File path: R-package/vignettes/CharRnnModel.Rmd
 ##########
 @@ -36,237 +32,262 @@ download.data <- function(data_dir) {
 }
 ```
 
-Make dictionary from text.
-
-```{r}
-make.dict <- function(text, max.vocab=10000) {
-    text <- strsplit(text, '')
-    dic <- list()
-    idx <- 1
-    for (c in text[[1]]) {
-        if (!(c %in% names(dic))) {
-            dic[[c]] <- idx
-            idx <- idx + 1
-        }
-    }
-    if (length(dic) == max.vocab - 1)
-        dic[["UNKNOWN"]] <- idx
-    cat(paste0("Total unique char: ", length(dic), "\n"))
-    return (dic)
-}
-```
-
-Transfer text into data feature.
-
-```{r}
-make.data <- function(file.path, seq.len=32, max.vocab=10000, dic=NULL) {
-    fi <- file(file.path, "r")
-    text <- paste(readLines(fi), collapse="\n")
-    close(fi)
-
-    if (is.null(dic))
-        dic <- make.dict(text, max.vocab)
-    lookup.table <- list()
-    for (c in names(dic)) {
-        idx <- dic[[c]]
-        lookup.table[[idx]] <- c 
-    }
-
-    char.lst <- strsplit(text, '')[[1]]
-    num.seq <- as.integer(length(char.lst) / seq.len)
-    char.lst <- char.lst[1:(num.seq * seq.len)]
-    data <- array(0, dim=c(seq.len, num.seq))
-    idx <- 1
-    for (i in 1:num.seq) {
-        for (j in 1:seq.len) {
-            if (char.lst[idx] %in% names(dic))
-                data[j, i] <- dic[[ char.lst[idx] ]]-1
-            else {
-                data[j, i] <- dic[["UNKNOWN"]]-1
-            }
-            idx <- idx + 1
-        }
-    }
-    return (list(data=data, dic=dic, lookup.table=lookup.table))
+Next we transform the test into feature vectors that is fed into the RNN model. The `make_data` function reads the dataset, cleans it of any non-alphanumeric characters, splits it into individual characters and groups it into sequences of length `seq.len`.
+
+
+```R
+make_data <- function(path, seq.len=32, dic=NULL) {
+  
+  text_vec <- read_file(file = path)
+  text_vec <- stri_enc_toascii(str = text_vec)
+  text_vec <- str_replace_all(string = text_vec, pattern = "[^[:print:]]", replacement = "")
+  text_vec <- strsplit(text_vec, '') %>% unlist
+  
+  if (is.null(dic)) {
+    char_keep <- sort(unique(text_vec))
+  } else char_keep <- names(dic)[!dic == 0]
+  
+  # Remove terms not part of dictionnary
+  text_vec <- text_vec[text_vec %in% char_keep]
+  
+  # Build dictionnary
+  dic <- 1:length(char_keep)
+  names(dic) <- char_keep
+  
+  # reverse dictionnary
+  rev_dic <- names(dic)
+  names(rev_dic) <- dic
+  
+  # Adjust by -1 to have a 1-lag for labels
+  num.seq <- as.integer(floor((length(text_vec)-1)/seq.len))
 
 Review comment:
   `a %/% b` gives you integer division.

----------------------------------------------------------------
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