Latent Dirichlet Allocation Using ‘tidyverse’ Conventions
tidylda implements an algorithm for Latent Dirichlet Allocation using style conventions from the tidyverse and tidymodels.
In addition this implementation of LDA allows you to:
Note that the seeding of topics and transfer learning are experimental for now. They are almost-surely useful but their behaviors have not been optimized or well-studied. Caveat emptor!
You can install the development version from GitHub with:
This package is still in its early stages of development. However, some basic functionality is below. Here, we will use the tidytext package to create a document term matrix, fit a topic model, predict topics of unseen documents, and update the model with those new documents.
tidylda uses the following naming conventions for topic models:
theta is a matrix whose rows are distributions of topics over documents, or P(topic|document)beta is a matrix whose rows are distributions of tokens over topics, or P(token|topic)lambda is a matrix whose rows are distributions of topics over tokens, or P(topic|token) lambda is useful for making predictions with a computationally-simple and efficient dot product and it may be interesting to analyze in its own right.alpha is the prior that tunes thetaeta is the prior that tunes betalibrary(tidytext)
library(tidyverse)
#> ── Attaching packages ─────────────────────────────────────── tidyverse 1.3.1 ──
#> ✓ ggplot2 3.3.5 ✓ purrr 0.3.4
#> ✓ tibble 3.1.5 ✓ dplyr 1.0.7
#> ✓ tidyr 1.1.4 ✓ stringr 1.4.0
#> ✓ readr 2.0.1 ✓ forcats 0.5.1
#> ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
#> x dplyr::filter() masks stats::filter()
#> x dplyr::lag() masks stats::lag()
library(tidylda)
#> tidylda is under active development. The API and behavior may change.
library(Matrix)
#>
#> Attaching package: 'Matrix'
#> The following objects are masked from 'package:tidyr':
#>
#> expand, pack, unpack
### Initial set up ---
# load some documents
docs <- nih_sample
# tokenize using tidytext's unnest_tokens
tidy_docs <- docs %>%
select(APPLICATION_ID, ABSTRACT_TEXT) %>%
unnest_tokens(output = word,
input = ABSTRACT_TEXT,
stopwords = stop_words$word,
token = "ngrams",
n_min = 1, n = 2) %>%
count(APPLICATION_ID, word) %>%
filter(n>1) #Filtering for words/bigrams per document, rather than per corpus
tidy_docs <- tidy_docs %>% # filter words that are just numbers
filter(! stringr::str_detect(tidy_docs$word, "^[0-9]+$"))
# append observation level data
colnames(tidy_docs)[1:2] <- c("document", "term")
# turn a tidy tbl into a sparse dgCMatrix
# note tidylda has support for several document term matrix formats
d <- tidy_docs %>%
cast_sparse(document, term, n)
# let's split the documents into two groups to demonstrate predictions and updates
d1 <- d[1:50, ]
d2 <- d[51:nrow(d), ]
# make sure we have different vocabulary for each data set to simulate the "real world"
# where you get new tokens coming in over time
d1 <- d1[, colSums(d1) > 0]
d2 <- d2[, colSums(d2) > 0]
### fit an intial model and inspect it ----
set.seed(123)
lda <- tidylda(
data = d1,
k = 10,
iterations = 200,
burnin = 175,
alpha = 0.1, # also accepts vector inputs
eta = 0.05, # also accepts vector or matrix inputs
optimize_alpha = FALSE, # experimental
calc_likelihood = TRUE,
calc_r2 = TRUE, # see https://arxiv.org/abs/1911.11061
return_data = FALSE
)
# did the model converge?
# there are actual test stats for this, but should look like "yes"
qplot(x = iteration, y = log_likelihood, data = lda$log_likelihood, geom = "line") +
ggtitle("Checking model convergence")
# look at the model overall
glance(lda)
#> # A tibble: 1 × 5
#> num_topics num_documents num_tokens iterations burnin
#> <int> <int> <int> <dbl> <dbl>
#> 1 10 50 1524 200 175
print(lda)
#> A Latent Dirichlet Allocation Model of 10 topics, 50 documents, and 1524 tokens:
#> tidylda(data = d1, k = 10, iterations = 200, burnin = 175, alpha = 0.1,
#> eta = 0.05, optimize_alpha = FALSE, calc_likelihood = TRUE,
#> calc_r2 = TRUE, return_data = FALSE)
#>
#> The model's R-squared is 0.2677
#> The 5 most prevalent topics are:
#> # A tibble: 10 × 4
#> topic prevalence coherence top_terms
#> <dbl> <dbl> <dbl> <chr>
#> 1 7 11.9 0.140 cancer, dcis, cells, breast, rb, ...
#> 2 1 11.0 0.22 cdk5, effects, v4, stiffening, wall, ...
#> 3 6 10.9 0.295 sud, risk, factors, drug, imaging, ...
#> 4 10 10.8 0.303 mitochondrial, plasticity, studies, redox, functio…
#> 5 3 10.5 0.248 sleep, cell, lung, cells, specific, ...
#> # … with 5 more rows
#>
#> The 5 most coherent topics are:
#> # A tibble: 10 × 4
#> topic prevalence coherence top_terms
#> <dbl> <dbl> <dbl> <chr>
#> 1 9 10.2 0.509 diabetes, numeracy, data, intervention, management…
#> 2 8 8.83 0.494 function, cmybp, injury, fragment, 29 kda, ...
#> 3 10 10.8 0.303 mitochondrial, plasticity, studies, redox, functio…
#> 4 6 10.9 0.295 sud, risk, factors, drug, imaging, ...
#> 5 5 8.06 0.258 research, disparities, program, cancer, mhirt, ...
#> # … with 5 more rows
# it comes with its own summary matrix that's printed out with print(), above
lda$summary
#> # A tibble: 10 × 4
#> topic prevalence coherence top_terms
#> <dbl> <dbl> <dbl> <chr>
#> 1 1 11.0 0.22 cdk5, effects, v4, stiffening, wall, ...
#> 2 2 8.07 0.18 research, responses, antibodies, natural, respons…
#> 3 3 10.5 0.248 sleep, cell, lung, cells, specific, ...
#> 4 4 9.76 0.145 cns, cells, brain, infection, based, ...
#> 5 5 8.06 0.258 research, disparities, program, cancer, mhirt, ...
#> 6 6 10.9 0.295 sud, risk, factors, drug, imaging, ...
#> 7 7 11.9 0.140 cancer, dcis, cells, breast, rb, ...
#> 8 8 8.83 0.494 function, cmybp, injury, fragment, 29 kda, ...
#> 9 9 10.2 0.509 diabetes, numeracy, data, intervention, managemen…
#> 10 10 10.8 0.303 mitochondrial, plasticity, studies, redox, functi…
# inspect the individual matrices
tidy_theta <- tidy(lda, matrix = "theta")
tidy_theta
#> # A tibble: 500 × 3
#> document topic theta
#> <chr> <dbl> <dbl>
#> 1 8574224 1 0.00238
#> 2 8574224 2 0.00238
#> 3 8574224 3 0.00238
#> 4 8574224 4 0.00238
#> 5 8574224 5 0.00238
#> 6 8574224 6 0.976
#> 7 8574224 7 0.00333
#> 8 8574224 8 0.00238
#> 9 8574224 9 0.00429
#> 10 8574224 10 0.00238
#> # … with 490 more rows
tidy_beta <- tidy(lda, matrix = "beta")
tidy_beta
#> # A tibble: 15,240 × 3
#> topic token beta
#> <dbl> <chr> <dbl>
#> 1 1 adolescence 0.0000673
#> 2 1 age 0.0000673
#> 3 1 application 0.0000673
#> 4 1 depressive 0.0000673
#> 5 1 disorder 0.0000673
#> 6 1 emotionality 0.0000673
#> 7 1 information 0.00276
#> 8 1 mdd 0.0000673
#> 9 1 onset 0.0000673
#> 10 1 onset mdd 0.0000673
#> # … with 15,230 more rows
tidy_lambda <- tidy(lda, matrix = "lambda")
tidy_lambda
#> # A tibble: 15,240 × 3
#> topic token lambda
#> <dbl> <chr> <dbl>
#> 1 1 adolescence 0.00780
#> 2 1 age 0.00912
#> 3 1 application 0.00780
#> 4 1 depressive 0.0201
#> 5 1 disorder 0.0201
#> 6 1 emotionality 0.0201
#> 7 1 information 0.273
#> 8 1 mdd 0.0111
#> 9 1 onset 0.00773
#> 10 1 onset mdd 0.0201
#> # … with 15,230 more rows
# append observation-level data
augmented_docs <- augment(lda, data = tidy_docs)
augmented_docs
#> # A tibble: 4,566 × 3
#> document term topic
#> <chr> <chr> <int>
#> 1 8574224 adolescence 6
#> 2 8574224 age 6
#> 3 8574224 application 6
#> 4 8574224 depressive 6
#> 5 8574224 disorder 6
#> 6 8574224 emotionality 6
#> 7 8574224 information 7
#> 8 8574224 mdd 6
#> 9 8574224 onset 6
#> 10 8574224 onset mdd 6
#> # … with 4,556 more rows
### predictions on held out data ---
# two methods: gibbs is cleaner and more techically correct in the bayesian sense
p_gibbs <- predict(lda, new_data = d2[1, ], iterations = 100, burnin = 75)
# dot is faster, less prone to error (e.g. underflow), noisier, and frequentist
p_dot <- predict(lda, new_data = d2[1, ], method = "dot")
# pull both together into a plot to compare
tibble(topic = 1:ncol(p_gibbs), gibbs = p_gibbs[1,], dot = p_dot[1, ]) %>%
pivot_longer(cols = gibbs:dot, names_to = "type") %>%
ggplot() +
geom_bar(mapping = aes(x = topic, y = value, group = type, fill = type),
stat = "identity", position="dodge") +
scale_x_continuous(breaks = 1:10, labels = 1:10) +
ggtitle("Gibbs predictions vs. dot product predictions")
### Augment as an implicit prediction using the 'dot' method ----
# Aggregating over terms results in a distribution of topics over documents
# roughly equivalent to using the "dot" method of predictions.
augment_predict <-
augment(lda, tidy_docs, "prob") %>%
group_by(document) %>%
select(-c(document, term)) %>%
summarise_all(function(x) sum(x, na.rm = T))
#> Adding missing grouping variables: `document`
# reformat for easy plotting
augment_predict <-
as_tibble(t(augment_predict[, -1]))
#> Warning: The `x` argument of `as_tibble.matrix()` must have unique column names if `.name_repair` is omitted as of tibble 2.0.0.
#> Using compatibility `.name_repair`.
#> This warning is displayed once every 8 hours.
#> Call `lifecycle::last_lifecycle_warnings()` to see where this warning was generated.
colnames(augment_predict) <- unique(tidy_docs$document)
augment_predict$topic <- 1:nrow(augment_predict) %>% as.factor()
compare_mat <-
augment_predict %>%
select(
topic,
augment = matches(rownames(d2)[1])
) %>%
mutate(
augment = augment / sum(augment), # normalize to sum to 1
dot = p_dot[1, ]
) %>%
pivot_longer(cols = c(augment, dot))
ggplot(compare_mat) +
geom_bar(aes(y = value, x = topic, group = name, fill = name),
stat = "identity", position = "dodge") +
labs(title = "Prediction using 'augment' vs 'predict(..., method = \"dot\")'")
# Not shown: aggregating over documents results in recovering the "tidy" lambda.
### updating the model ----
# now that you have new documents, maybe you want to fold them into the model?
lda2 <- refit(
object = lda,
new_data = d, # save me the trouble of manually-combining these by just using d
iterations = 200,
burnin = 175,
calc_likelihood = TRUE,
calc_r2 = TRUE
)
# we can do similar analyses
# did the model converge?
qplot(x = iteration, y = log_likelihood, data = lda2$log_likelihood, geom = "line") +
ggtitle("Checking model convergence")
# look at the model overall
glance(lda2)
#> # A tibble: 1 × 5
#> num_topics num_documents num_tokens iterations burnin
#> <int> <int> <int> <dbl> <dbl>
#> 1 10 99 2962 200 175
print(lda2)
#> A Latent Dirichlet Allocation Model of 10 topics, 99 documents, and 2962 tokens:
#> refit.tidylda(object = lda, new_data = d, iterations = 200, burnin = 175,
#> calc_likelihood = TRUE, calc_r2 = TRUE)
#>
#> The model's R-squared is 0.1478
#> The 5 most prevalent topics are:
#> # A tibble: 10 × 4
#> topic prevalence coherence top_terms
#> <dbl> <dbl> <dbl> <chr>
#> 1 5 14.0 0.148 research, program, cancer, disparities, health, ...
#> 2 9 12.2 0.0532 data, health, diabetes, hiv, intervention, ...
#> 3 2 11.5 0.0739 core, research, response, microbiome, responses, .…
#> 4 6 11.0 0.163 risk, factors, clinical, sud, drug, ...
#> 5 7 9.92 0.217 cancer, tumor, treatment, cells, breast, ...
#> # … with 5 more rows
#>
#> The 5 most coherent topics are:
#> # A tibble: 10 × 4
#> topic prevalence coherence top_terms
#> <dbl> <dbl> <dbl> <chr>
#> 1 8 7.74 0.423 injury, function, cmybp, fragment, 29 kda, ...
#> 2 10 6.09 0.320 mitochondrial, plasticity, studies, redox, functio…
#> 3 7 9.92 0.217 cancer, tumor, treatment, cells, breast, ...
#> 4 4 9.2 0.171 cns, cell, cells, inflammation, mast, ...
#> 5 6 11.0 0.163 risk, factors, clinical, sud, drug, ...
#> # … with 5 more rows
# how does that compare to the old model?
print(lda)
#> A Latent Dirichlet Allocation Model of 10 topics, 50 documents, and 1524 tokens:
#> tidylda(data = d1, k = 10, iterations = 200, burnin = 175, alpha = 0.1,
#> eta = 0.05, optimize_alpha = FALSE, calc_likelihood = TRUE,
#> calc_r2 = TRUE, return_data = FALSE)
#>
#> The model's R-squared is 0.2677
#> The 5 most prevalent topics are:
#> # A tibble: 10 × 4
#> topic prevalence coherence top_terms
#> <dbl> <dbl> <dbl> <chr>
#> 1 7 11.9 0.140 cancer, dcis, cells, breast, rb, ...
#> 2 1 11.0 0.22 cdk5, effects, v4, stiffening, wall, ...
#> 3 6 10.9 0.295 sud, risk, factors, drug, imaging, ...
#> 4 10 10.8 0.303 mitochondrial, plasticity, studies, redox, functio…
#> 5 3 10.5 0.248 sleep, cell, lung, cells, specific, ...
#> # … with 5 more rows
#>
#> The 5 most coherent topics are:
#> # A tibble: 10 × 4
#> topic prevalence coherence top_terms
#> <dbl> <dbl> <dbl> <chr>
#> 1 9 10.2 0.509 diabetes, numeracy, data, intervention, management…
#> 2 8 8.83 0.494 function, cmybp, injury, fragment, 29 kda, ...
#> 3 10 10.8 0.303 mitochondrial, plasticity, studies, redox, functio…
#> 4 6 10.9 0.295 sud, risk, factors, drug, imaging, ...
#> 5 5 8.06 0.258 research, disparities, program, cancer, mhirt, ...
#> # … with 5 more rowsI plan to have more analyses and a fuller accounting of the options of the various functions when I write the vignettes.
See the “Issues” tab on GitHub to see planned features as well as bug fixes.
If you have any suggestions for additional functionality, changes to functionality, changes to arguments or other aspects of the API please let me know by opening an issue on GitHub or sending me an email: jones.thos.w at gmail.com.