Unexpected outcome in text analysis using R with gutenbergr and cv.glmnet - r

Question

hope that all of you are fine,
I'm trying to replicate a text analysis exercise ir R with the gutembergr library, and trying to run a machine learning model with the glmnet library.
The problem is that the code actually runs, but it returns the wrong outcome.
I leave the full code, which is not very extensive and is fully replicable.
First part of the code, runs fine
##### Carga de librerias sajtas #####
library(tidyverse)
library(tidytext)
library(udpipe)
library(gutenbergr)
library(rsample)
library(glmnet)
library(yardstick)
##### 1. Cargar los libros del gutenberg ####
twist_tale <- gutenberg_metadata %>%
filter(
title %in% c("A Tale of Two Cities", "Oliver Twist"),
has_text,
language == "en") %>%
pull(gutenberg_id) %>%
gutenberg_download(meta_fields = "title")
##### 2. Quitarle los blancos #####
twist_tale <- twist_tale %>%
filter(text != "")
View(twist_tale)
##### 3. Crear la varialbe lógica ######
twist_tale <- twist_tale %>%
mutate(
es_two_cities = case_when(
title == "A Tale of Two Cities" ~ 1L,
title == "Oliver Twist" ~ 0L
),
line_id = row_number()
)%>%
view()
##### 3.1. Guardar los resultados y contar las filas ####
twist_tale %>%
count(title)
##### 4. Preparar el dataset y modelar #####
dl <- udpipe_download_model(language = "english")
english_model <- udpipe_load_model(dl$file_model)
text <- twist_tale %>%
select(doc_id = line_id, text)
twist_tale_preprocesado <- udpipe(text, english_model, parallel.cores = 4L)
##### 4b. crear un algoritmo de entrenamiento #####
set.seed(1234L)
twist_tale_split <- initial_split(twist_tale)
twist_tale_training <- training(twist_tale_split)
twist_tale_testing <- testing(twist_tale_split)
##### 4a. Ponerlo en minúsculas y quitarle la mugre ####
sparse_train_data <- twist_tale_preprocesado %>%
mutate(lemma = str_to_lower(lemma)) %>%
anti_join(stop_words, by = c("lemma" = "word")) %>%
filter(upos %in% c("PUNCT", "SYM", "X", "NUM")) %>%
mutate(doc_id = as.integer(doc_id)) %>%
anti_join(twist_tale_testing, by = c("doc_id" = "line_id")) %>%
count(doc_id, lemma) %>%
cast_sparse(doc_id, lemma, n)
Until here, everything is fine. The problem arises running the cv.glmnet() model, since it doesn't return any error or warning message, but the outcome is a row vector, and I believe that it must be a matrix or tibble.
# The code that presents the problem
##### 5a. Excluir los resultados irrelevantes y guardarlos ####
y <- twist_tale_training %>%
filter(line_id %in% rownames(sparse_train_data)) %>%
pull(es_two_cities)
##### 6a. Calcular la regresión logística regularizada ####
model <- cv.glmnet(sparse_train_data, y, family = "binomial",
keep = T, trace.it=1) # THE PROBLEM IS ACTUALLY HERE
coeficientes <- model$glmnet.fit %>%
tidy() %>%
filter(lambda == model$lambda.1se)
coeficientes %>%
group_by(estimate > 0) %>%
slice_max(estimate, n = 5) %>%
ungroup()
coeficientes %>%
group_by(estimate > 0) %>%
slice_max(estimate, n = 5) %>%
ungroup() %>%
ggplot() +
geom_col(aes(x = fct_reorder(term, estimate), y = estimate, fill = estimate > 0)) +
coord_flip()
I Apologize for the extent of the question but was for showing the full context.

Answer 1

Haven't run your code but probably you need to make sure the x matrix and y response vector are in the same order. Something alongside the following will do that (replacing step 5a. and the beginning of your step 6)
library(udpipe)
y <- setNames(twist_tale_training$es_two_cities, twist_tale_training$line_id)
traindata <- dtm_align(x = sparse_train_data, y = y)
model <- cv.glmnet(x = traindata$x, y = traindata$y, family = "binomial", keep = T, trace.it = 1)