bnosac / etm Goto Github PK

• on torch > 0.5 with fixed seeds
• to proof that we saw already that the implementation gives the same results as https://github.com/bnosac-dev/ETM

• using log directory 'd:/RCompile/CRANguest/R-devel/ETM.Rcheck'
• using R Under development (unstable) (2021-08-13 r80752)
• using platform: x86_64-w64-mingw32 (64-bit)
• using session charset: ISO8859-1
• checking for file 'ETM/DESCRIPTION' ... OK
• checking extension type ... Package
• this is package 'ETM' version '0.1.0'
• package encoding: UTF-8
• checking CRAN incoming feasibility ... ERROR

New submission

Conflicting package names (submitted: ETM, existing: etm [https://CRAN.R-project.org])

Conflicting package names (submitted: ETM, existing: etm [CRAN archive])

• as a generic
• maybe include in R package word2vec
• passing to uwot::umap / alongside uwot::umap_transform

explain KL / NELBO / Loss in docs based on the paper

• - showing to show evolution of loss (plot(model, type = "loss", ...)
• - showing words emitted by each topic (plot(model, type = "terminology", ...)
  ================> this will simplify workflow

• topic diversity
• topic coherence

Remarks can on better structuring of license / copyright can be put in this thread. Issue created as remarked from Adji at adjidieng/ETM-R#3
Some notes on how copyright is referenced currently in the package

• Description file indicates

  • R part at the R folder is from Jan Wijffels and indicated at https://github.com/bnosac/ETM/blob/master/DESCRIPTION#L7
  • The Python code at inst/orig is from Adji B. Dieng and colleagues https://github.com/bnosac/ETM/blob/master/DESCRIPTION#L9:L11

• The file LICENSE.note provides the license as indicated at https://github.com/adjidieng/ETM/blob/master/LICENSE according to the CRAN recommendation at https://cran.r-project.org/doc/manuals/r-release/R-exts.html#Licensing and refers to the clone in the inst/orig/ETM folder of the source code at https://github.com/adjidieng/ETM. The R package does not use this code when executing the model, code is only there as a reference.

• The file LICENSE uses the template is indicated by the documentation of the R-project https://www.r-project.org/Licenses/MIT. At that location we can add as well Adji B. Dieng, Francisco J. R. Ruiz, David M. Blei @adjidieng agrees with this.

• We could as well add the citation as https://github.com/adjidieng/ETM-R/blob/main/inst/CITATION

Hello! When I first tried your ETM package in R using the Belgian parliament data, it worked. However, when I was testing it on my (small) data, after running the ETM() function and optimizer, I encountered this error:

Error in seq_len(nrow(x)) : argument must be coercible to non-negative integer

Here is my code:

library(topicmodels.etm)
library(doc2vec)
library(word2vec)
gcash_data <- read.csv("GCash_200_Reviews_PlayStore_RepeatScroll20_Wait5s_TimeOut60s_AJAx.csv")
names(gcash_data) <- c("UserName", "Date", "Likes", "Review", "Rating")
gcash_r5 <- filter(gcash_data, Rating == "5")
head(gcash_r5)
str(gcash_r5)

x      <- data.frame(doc_id           = gcash_r5$UserName, 
                     text             = gcash_r5$Review, 
                     stringsAsFactors = FALSE)
x$text <- txt_clean_word2vec(x$text)

w2v        <- word2vec(x = x$text, dim = 25, type = "skip-gram", iter = 10, min_count = 5, threads = 2)
embeddings <- as.matrix(w2v)
predict(w2v, newdata = c("app", "convenient"), type = "nearest", top_n = 4)

library(udpipe)
dtm   <- strsplit.data.frame(x, group = "doc_id", term = "text", split = " ")
dtm   <- document_term_frequencies(dtm)
dtm   <- document_term_matrix(dtm)
dtm   <- dtm_remove_tfidf(dtm, prob = 0.50)

vocab        <- intersect(rownames(embeddings), colnames(dtm))
embeddings   <- dtm_conform(embeddings, rows = vocab)
dtm          <- dtm_conform(dtm,     columns = vocab)
dim(dtm)
dim(embeddings)

set.seed(1234)
torch_manual_seed(4321)
model     <- ETM(k = 5, dim = 100, embeddings = embeddings)
optimizer <- optim_adam(params = model$parameters, lr = 0.005, weight_decay = 0.0000012)
loss      <- model$fit(data = dtm, optimizer = optimizer, epoch = 20, batch_size = 5)

As you may see in the code, for the ETM function, I changed args to k=5 topics. For model$fit, I changed args to batch_size =5.

After running the last line above with "model$fit", the following error occurs:
Error in seq_len(nrow(x)) : argument must be coercible to non-negative integer

Is this because I am trying to run a small dataset? How may I solve this?

Thank you in advance! :)

to_device