632 lines
23 KiB
R
632 lines
23 KiB
R
|
## 21/04/2020 : shiny pour active learning étiquetage de textes
|
||
|
|
|
||
|
|
shinyUI(fluidPage(
|
||
|
|
title = "Active Tigger",
|
||
|
|
|
||
|
|
############################################################################
|
||
|
|
## Top panel: Title, strategy, model options
|
||
|
|
############################################################################
|
||
|
|
fluidRow(
|
||
|
|
column(
|
||
|
|
4,
|
||
|
|
fluidRow(
|
||
|
|
column(4, br(),
|
||
|
|
HTML('<img src="active_tigger.png" width="100%">')),
|
||
|
|
column(
|
||
|
|
8, br(),
|
||
|
|
p(strong("Project / Scheme")),
|
||
|
|
fluidRow(
|
||
|
|
column(2, HTML(paste0(
|
||
|
|
'<div title="New project">',
|
||
|
|
actionButton("createProject", "+"),
|
||
|
|
'</div>'))),
|
||
|
|
column(10, selectInput("selectProject", NULL, NULL, NULL))
|
||
|
|
),
|
||
|
|
fluidRow(
|
||
|
|
column(2, HTML(paste0(
|
||
|
|
'<div title="New scheme">',
|
||
|
|
actionButton("createScheme", "+"),
|
||
|
|
'</div>'))),
|
||
|
|
column(10, selectInput("selectScheme", NULL, NULL, NULL))
|
||
|
|
)
|
||
|
|
))
|
||
|
|
),
|
||
|
|
column(
|
||
|
|
2, br(),
|
||
|
|
|
||
|
|
HTML(paste0(
|
||
|
|
'<div title="Querying strategy">',
|
||
|
|
p(strong("Strategy")),
|
||
|
|
selectInput("strategy", NULL,
|
||
|
|
choices= c("Active" = "entropy",
|
||
|
|
"MaxProb" = "maxprob",
|
||
|
|
"Random" = "random",
|
||
|
|
"Deterministic" = "sequential"),
|
||
|
|
selected = "sequential"),
|
||
|
|
conditionalPanel("input.strategy == 'maxprob'",
|
||
|
|
selectInput("maxprobWhich", NULL, NULL)),
|
||
|
|
'</div>'))
|
||
|
|
),
|
||
|
|
column(
|
||
|
|
2, br(),
|
||
|
|
HTML(paste0(
|
||
|
|
'<div title="Query from which?">',
|
||
|
|
p(strong("On")),
|
||
|
|
selectInput("sampleChoice", NULL,
|
||
|
|
choices = c("Untagged" = "untagged",
|
||
|
|
"Tagged" = "tagged",
|
||
|
|
"All" = "all")),
|
||
|
|
conditionalPanel("input.sampleChoice == 'tagged'",
|
||
|
|
selectInput("taggedWhich", NULL, NULL)),
|
||
|
|
'</div>'))
|
||
|
|
),
|
||
|
|
|
||
|
|
column(4, p(br()),
|
||
|
|
verbatimTextOutput("trainDiagno"),
|
||
|
|
fluidRow(
|
||
|
|
column(8, HTML(paste0(
|
||
|
|
'<div title="Train prediction model 🤖">',
|
||
|
|
actionButton("modelTrain", "🤖 Train", width = "100%"),
|
||
|
|
'</div>'))),
|
||
|
|
column(4, HTML(paste0(
|
||
|
|
'<div title="🤖 Prediction model options">',
|
||
|
|
checkboxInput("showTrainOptions", "🔧"),
|
||
|
|
'</div>')))
|
||
|
|
))
|
||
|
|
),
|
||
|
|
|
||
|
|
|
||
|
|
## General training options
|
||
|
|
conditionalPanel(
|
||
|
|
"input.showTrainOptions",
|
||
|
|
hr(),
|
||
|
|
fluidRow(
|
||
|
|
column(
|
||
|
|
4,
|
||
|
|
conditionalPanel(
|
||
|
|
"input.showTrainOptions & input.use_regressors.includes('regex')",
|
||
|
|
uiOutput("panelExtraRegex")
|
||
|
|
)
|
||
|
|
),
|
||
|
|
column(
|
||
|
|
4,
|
||
|
|
p(strong("Predictors")),
|
||
|
|
selectizeInput("use_regressors", NULL, "",
|
||
|
|
multiple = TRUE, width = "100%"),
|
||
|
|
conditionalPanel(
|
||
|
|
"input.showTrainOptions & input.use_regressors.includes('extra')",
|
||
|
|
selectizeInput("use_ootregnum", "Extra predictors: continuous", "",
|
||
|
|
multiple = TRUE, width = "100%"),
|
||
|
|
selectizeInput("use_ootregcat", "Extra predictors: categorical", "",
|
||
|
|
multiple = TRUE, width = "100%")
|
||
|
|
),
|
||
|
|
|
||
|
|
conditionalPanel(
|
||
|
|
"input.showTrainOptions & input.use_regressors.includes('dfm')",
|
||
|
|
hr(),
|
||
|
|
fluidRow(
|
||
|
|
column(
|
||
|
|
6, numericInput("dfmMinTermfreq", "DFM : Min Termfreq",
|
||
|
|
min= 1, max= 1e3, value= 5, step= 1)),
|
||
|
|
column(
|
||
|
|
6, numericInput("dfmMinDocfreq", "DFM : Min Docfreq",
|
||
|
|
min= 1, max= 1e3, value= 5, step= 1))),
|
||
|
|
fluidRow(
|
||
|
|
column(
|
||
|
|
4, checkboxInput("dfmTfIdf", "Tf-Idf", TRUE)),
|
||
|
|
column(
|
||
|
|
4, selectInput("dfmTfScheme", label = NULL,
|
||
|
|
choices= c("logcount", "count", "prop", "propmax",
|
||
|
|
"boolean", "augmented", "logave"),
|
||
|
|
selected= "logcount")),
|
||
|
|
column(
|
||
|
|
4, selectInput("dfmDfScheme", label = NULL,
|
||
|
|
choices= c("inverse", "count",
|
||
|
|
"inversemax", "inverseprob", "unary"),
|
||
|
|
selected= "inverse"))
|
||
|
|
),
|
||
|
|
numericInput("dfmNgrams", "N-grams", value= 1,
|
||
|
|
min= 1, max= 10, step = 1)
|
||
|
|
)
|
||
|
|
),
|
||
|
|
column(
|
||
|
|
4,
|
||
|
|
HTML(paste0(
|
||
|
|
'<div title="Auto train after # tags (0=never)">',
|
||
|
|
fluidRow(
|
||
|
|
column(6, strong("Auto train every")),
|
||
|
|
column(6, numericInput("trainCountdown", NULL, 0, 0, 1e6, 1))),
|
||
|
|
'</div>'
|
||
|
|
)),
|
||
|
|
HTML(paste0(
|
||
|
|
'<div title="🤖 model (recommended: Liblinear)">',
|
||
|
|
fluidRow(
|
||
|
|
column(6, strong("Model")),
|
||
|
|
column(6, selectInput("predModel", NULL, selected = "linear",
|
||
|
|
choices = c("Naive Bayes" = "naive bayes",
|
||
|
|
"KNN" = "knn",
|
||
|
|
"Liblinear" = "linear",
|
||
|
|
"LASSO" = "lasso",
|
||
|
|
"Random Forest" = "random forest")))
|
||
|
|
),
|
||
|
|
'</div>'
|
||
|
|
)),
|
||
|
|
|
||
|
|
## Model-specific training options
|
||
|
|
### Random forest options
|
||
|
|
conditionalPanel(
|
||
|
|
"input.showTrainOptions & input.predModel == 'random forest'",
|
||
|
|
fluidRow(
|
||
|
|
numericInput("rfNumTrees", label = "Num. trees",
|
||
|
|
min = 1, max = 2e3, value = 500, step = 1),
|
||
|
|
numericInput("rfMtry", label = "mtry",
|
||
|
|
min = 0, max = 1e5, value = 0, step = 1),
|
||
|
|
numericInput("rfSampleFrac", label = "Sample fraction",
|
||
|
|
min = 0, max = 1, value = 1, step = .01)
|
||
|
|
)
|
||
|
|
),
|
||
|
|
### Naive Bayes options
|
||
|
|
conditionalPanel(
|
||
|
|
"input.showTrainOptions & input.predModel == 'naive bayes'",
|
||
|
|
flowLayout(
|
||
|
|
numericInput("naiveSmooth", label = "Smooth",
|
||
|
|
min = 0, max = 2e3,
|
||
|
|
value = 1, step = 1e-3),
|
||
|
|
selectInput("naivePrior", "Prior",
|
||
|
|
c("uniform", "docfreq", "termfreq")),
|
||
|
|
selectInput("naiveDistri", "Distribution",
|
||
|
|
c("multinomial", "Bernoulli"))
|
||
|
|
)
|
||
|
|
),
|
||
|
|
### Lasso options
|
||
|
|
conditionalPanel(
|
||
|
|
"input.showTrainOptions & input.predModel == 'lasso'",
|
||
|
|
strong("Lasso penalty"),
|
||
|
|
fluidRow(
|
||
|
|
column(
|
||
|
|
6, numericInput("glmLambda", label = NULL, min = 0, max = 2e3,
|
||
|
|
value = 0, step = 1e-6)),
|
||
|
|
column(
|
||
|
|
6, actionButton("glmCV", label= "Find best (CV)")))
|
||
|
|
),
|
||
|
|
### Linear options
|
||
|
|
conditionalPanel(
|
||
|
|
"input.showTrainOptions & input.predModel == 'linear'",
|
||
|
|
strong("Liblinear Cost"),
|
||
|
|
fluidRow(
|
||
|
|
column(
|
||
|
|
6, numericInput("liblinCost", label= NULL, min= 0, max= 2e10,
|
||
|
|
value= 32, step= 1)),
|
||
|
|
column(
|
||
|
|
6, actionButton("liblinCV", label= "Find best (CV)")))
|
||
|
|
),
|
||
|
|
### KNN options
|
||
|
|
conditionalPanel(
|
||
|
|
"input.showTrainOptions & input.predModel == 'knn'",
|
||
|
|
flowLayout(
|
||
|
|
strong("N. Neighbours"),
|
||
|
|
numericInput("knnK", label = NULL, min = 1, max = 1e2,
|
||
|
|
value = 3, step = 1)
|
||
|
|
)
|
||
|
|
)
|
||
|
|
)
|
||
|
|
),
|
||
|
|
hr()
|
||
|
|
),
|
||
|
|
|
||
|
|
|
||
|
|
############################################################################
|
||
|
|
## Main panel set
|
||
|
|
############################################################################
|
||
|
|
tabsetPanel(
|
||
|
|
id = "mainPanelset",
|
||
|
|
selected = "Tagging",
|
||
|
|
|
||
|
|
########################################################################
|
||
|
|
## Project panel
|
||
|
|
########################################################################
|
||
|
|
tabPanel(
|
||
|
|
"Project",
|
||
|
|
br(),
|
||
|
|
tabsetPanel(
|
||
|
|
id = "tabsetProject",
|
||
|
|
selected = "Sample",
|
||
|
|
|
||
|
|
tabPanel(
|
||
|
|
"Settings",
|
||
|
|
br(),
|
||
|
|
actionButton("saveSystem", "Save changes"),
|
||
|
|
|
||
|
|
h4("Files"),
|
||
|
|
fluidRow(
|
||
|
|
column(2, p("Data directory")),
|
||
|
|
column(4, uiOutput("sys_datadir")),
|
||
|
|
column(6, p("Place (on the server) where the data and project are stored"))
|
||
|
|
),
|
||
|
|
fluidRow(
|
||
|
|
column(2, p("Data filename")),
|
||
|
|
column(4, uiOutput("sys_datafile")),
|
||
|
|
column(6, p("Main file, containing id and text columns"))
|
||
|
|
),
|
||
|
|
|
||
|
|
h4("Variables"),
|
||
|
|
fluidRow(
|
||
|
|
column(2, p("ID")),
|
||
|
|
column(4, uiOutput("sys_var_id")),
|
||
|
|
column(6, p("Name of the id variable, unique identifier of each text"))
|
||
|
|
),
|
||
|
|
fluidRow(
|
||
|
|
column(2, p("Text")),
|
||
|
|
column(4, uiOutput("sys_var_text")),
|
||
|
|
column(6, p("Name of the text variables: if more than one, texts are concatenated in the specified order"))
|
||
|
|
),
|
||
|
|
fluidRow(
|
||
|
|
column(2, p("Tags")),
|
||
|
|
column(4, uiOutput("sys_var_tag")),
|
||
|
|
column(6, p("Names of scheme variables"))
|
||
|
|
),
|
||
|
|
fluidRow(
|
||
|
|
column(2, p("Comments")),
|
||
|
|
column(4, uiOutput("sys_var_comm_ui")),
|
||
|
|
column(6, p("Name of the comments variable"))
|
||
|
|
),
|
||
|
|
fluidRow(
|
||
|
|
column(2, p("Context")),
|
||
|
|
column(4, uiOutput("sys_var_context_ui")),
|
||
|
|
column(6, p("Names of variables not used in the models, but may be displayed during tagging"))
|
||
|
|
),
|
||
|
|
|
||
|
|
h4("System"),
|
||
|
|
fluidRow(
|
||
|
|
column(2, checkboxInput("sys_use_python", "Python backend", FALSE)),
|
||
|
|
column(4, conditionalPanel(
|
||
|
|
"input.sys_use_python",
|
||
|
|
textInput("sys_which_python", NULL, value = "python3",
|
||
|
|
placeholder = "(custom python path)"))),
|
||
|
|
column(6, conditionalPanel(
|
||
|
|
"input.sys_use_python",
|
||
|
|
p("This must be a working python3 environment, with the required modules installed (see documentation)")))
|
||
|
|
),
|
||
|
|
|
||
|
|
conditionalPanel("input.sys_use_python", list(
|
||
|
|
fluidRow(
|
||
|
|
column(2, checkboxInput("sys_use_spacy", "SpaCy tokenization", FALSE)),
|
||
|
|
column(4, conditionalPanel("input.sys_use_spacy", textInput(
|
||
|
|
"sys_use_spacy_model", NULL, NULL, placeholder = "(spacy model name)"))),
|
||
|
|
column(6, p("Name of the spacy tokenizer model, used in DTM and word embeddings"))
|
||
|
|
),
|
||
|
|
conditionalPanel("input.sys_use_spacy", fluidRow(
|
||
|
|
column(2),
|
||
|
|
column(9, uiOutput("sys_spacyDlUI")))
|
||
|
|
),
|
||
|
|
fluidRow(
|
||
|
|
column(2, checkboxInput("sys_use_ft", "FastText word embeddings", FALSE)),
|
||
|
|
column(4, conditionalPanel("input.sys_use_ft", textInput(
|
||
|
|
"sys_use_ft_model", NULL, NULL, placeholder = "(fasttext model path)"))),
|
||
|
|
column(6, p("Path to the local fasttext model binary"))
|
||
|
|
),
|
||
|
|
conditionalPanel("input.sys_use_ft", fluidRow(
|
||
|
|
column(2),
|
||
|
|
column(9, uiOutput("sys_ftDlUI")))
|
||
|
|
),
|
||
|
|
fluidRow(
|
||
|
|
column(2, checkboxInput("sys_use_sb", "SBERT sentence embeddings", FALSE)),
|
||
|
|
column(4, conditionalPanel("input.sys_use_sb", textInput(
|
||
|
|
"sys_use_sb_model", NULL, NULL,
|
||
|
|
placeholder = "(custom sentence_transformers model)"))),
|
||
|
|
column(6, p("(GPU recommended) Name or path of the sentence-transformers model"))
|
||
|
|
),
|
||
|
|
|
||
|
|
conditionalPanel("input.sys_use_python", list(
|
||
|
|
checkboxInput("sys_use_gpu", "GPU support (CUDA, for SBERT and BERT)", FALSE),
|
||
|
|
|
||
|
|
br(),
|
||
|
|
wellPanel(
|
||
|
|
h4("Model picker"),
|
||
|
|
fluidRow(
|
||
|
|
column(2, p("Language")),
|
||
|
|
column(4, uiOutput("sys_ex_lang_ui")),
|
||
|
|
column(6, p("Used to preset tokenization and embedding models"))
|
||
|
|
),
|
||
|
|
fluidRow(
|
||
|
|
column(2),
|
||
|
|
column(4, strong("Recommended model")),
|
||
|
|
column(6, strong("Download instructions"))
|
||
|
|
),
|
||
|
|
fluidRow(
|
||
|
|
column(2, p("SpaCy tokenization")),
|
||
|
|
column(4, uiOutput("sys_ex_spacy")),
|
||
|
|
column(6, uiOutput("sys_ex_spacy_dl"))
|
||
|
|
),
|
||
|
|
fluidRow(
|
||
|
|
column(2, p("FastText word embeddings")),
|
||
|
|
column(4, uiOutput("sys_ex_ft")),
|
||
|
|
column(6, uiOutput("sys_ex_ft_dl"))
|
||
|
|
),
|
||
|
|
fluidRow(
|
||
|
|
column(2, p("SBERT sentence embeddings")),
|
||
|
|
column(4, uiOutput("sys_ex_sb")),
|
||
|
|
column(6, p("(Auto download by python module)"))
|
||
|
|
)
|
||
|
|
)
|
||
|
|
))
|
||
|
|
))
|
||
|
|
),
|
||
|
|
|
||
|
|
tabPanel(
|
||
|
|
"Sample",
|
||
|
|
br(),
|
||
|
|
fluidRow(
|
||
|
|
column(
|
||
|
|
4,
|
||
|
|
wellPanel(
|
||
|
|
fluidRow(
|
||
|
|
column(8, h4("Sample")),
|
||
|
|
column(4, actionButton("dataImport", "Import", width = "100%"))),
|
||
|
|
fluidRow(
|
||
|
|
column(6, numericInput("dataNrows", "N. rows", 500, 10, 1e4, 1)),
|
||
|
|
column(6, numericInput("dataSkipRows", "Skip rows", 0, 0, step = 1))
|
||
|
|
)
|
||
|
|
)
|
||
|
|
),
|
||
|
|
column(8, uiOutput("dataMessage"), uiOutput("panelData"))
|
||
|
|
)
|
||
|
|
),
|
||
|
|
|
||
|
|
tabPanel(
|
||
|
|
"Scheme",
|
||
|
|
br(),
|
||
|
|
fluidRow(
|
||
|
|
# column(4, uiOutput("panelScheme")),
|
||
|
|
column(
|
||
|
|
4,
|
||
|
|
wellPanel(
|
||
|
|
h4("Current scheme"),
|
||
|
|
fluidRow(
|
||
|
|
column(2, HTML(paste0(
|
||
|
|
"<div title='Delete scheme'>",
|
||
|
|
actionButton("schemeDelete", "🗑", width = "100%"),
|
||
|
|
"</div>"))),
|
||
|
|
column(6, uiOutput("printScheme")),
|
||
|
|
column(4, HTML(paste0(
|
||
|
|
"<div title='Save scheme description'>",
|
||
|
|
actionButton("schemeDescrSave", "Save", width = "100%"),
|
||
|
|
"</div>")))
|
||
|
|
),
|
||
|
|
br(),
|
||
|
|
textAreaInput("schemeDescr", NULL, width = "100%", rows = 10,
|
||
|
|
placeholder = "Write scheme description here"),
|
||
|
|
hr()
|
||
|
|
)
|
||
|
|
),
|
||
|
|
column(8, uiOutput("panelRetag"))
|
||
|
|
)
|
||
|
|
)
|
||
|
|
)
|
||
|
|
),
|
||
|
|
|
||
|
|
|
||
|
|
########################################################################
|
||
|
|
## Text / visualization panel
|
||
|
|
########################################################################
|
||
|
|
tabPanel(
|
||
|
|
"Tagging",
|
||
|
|
fluidRow(
|
||
|
|
column(
|
||
|
|
3,
|
||
|
|
br(),
|
||
|
|
fluidRow(
|
||
|
|
column(8, textInput("regexFilter", label = NULL,
|
||
|
|
placeholder = "(Regex filter)")),
|
||
|
|
column(4, checkboxInput("regexCaseSens", "Case"))),
|
||
|
|
|
||
|
|
wellPanel(
|
||
|
|
## Tagging buttons
|
||
|
|
fluidRow(
|
||
|
|
column(8, textInput("newLab", label = NULL,
|
||
|
|
placeholder = "(New label)")),
|
||
|
|
column(4, actionButton("currentAction", "Create"))
|
||
|
|
),
|
||
|
|
|
||
|
|
# fluidRow(uiOutput("oracleButtons")),
|
||
|
|
uiOutput("oracleButtons"),
|
||
|
|
|
||
|
|
br(),
|
||
|
|
textInput("currentComment", NULL, "", width = "100%",
|
||
|
|
placeholder = "(Comment)"),
|
||
|
|
br(),
|
||
|
|
uiOutput("makeOracleConfirm")
|
||
|
|
),
|
||
|
|
|
||
|
|
# fluidRow(
|
||
|
|
# column(6, checkboxInput("showContext", "Context")),
|
||
|
|
# column(6, actionButton("oops", strong("Oops")))
|
||
|
|
# ),
|
||
|
|
checkboxInput("showContext", "Context"),
|
||
|
|
conditionalPanel("input.showContext", htmlOutput("currentContext"))
|
||
|
|
),
|
||
|
|
column(
|
||
|
|
9,
|
||
|
|
fluidRow(
|
||
|
|
column(2, checkboxInput("panelText", "Text", TRUE)),
|
||
|
|
column(2, checkboxInput("panelVisu", "Visualization", FALSE),
|
||
|
|
offset = 8)
|
||
|
|
),
|
||
|
|
uiOutput("textVisuCols") # Handled in server.R for adaptive columns
|
||
|
|
)
|
||
|
|
)
|
||
|
|
),
|
||
|
|
|
||
|
|
########################################################################
|
||
|
|
## History panel
|
||
|
|
########################################################################
|
||
|
|
tabPanel(
|
||
|
|
"History",
|
||
|
|
br(),
|
||
|
|
actionButton("histSave", "Save changes"),
|
||
|
|
br(),
|
||
|
|
br(),
|
||
|
|
DT::dataTableOutput("histDTable")
|
||
|
|
),
|
||
|
|
|
||
|
|
|
||
|
|
########################################################################
|
||
|
|
## Stats panel
|
||
|
|
########################################################################
|
||
|
|
tabPanel(
|
||
|
|
"Stats",
|
||
|
|
br(),
|
||
|
|
fluidRow(
|
||
|
|
column(
|
||
|
|
3,
|
||
|
|
h3("Counts"),
|
||
|
|
tableOutput("statsTagTable")
|
||
|
|
),
|
||
|
|
column(
|
||
|
|
9,
|
||
|
|
h3("10-CV diagnostics"),
|
||
|
|
actionButton("statsCVgo", "Compute 10-CV"),
|
||
|
|
br(),
|
||
|
|
verbatimTextOutput("statsCVoutput"),
|
||
|
|
DT::dataTableOutput("statsCVtable")
|
||
|
|
)
|
||
|
|
),
|
||
|
|
hr(),
|
||
|
|
h3("Gold Standard")
|
||
|
|
),
|
||
|
|
|
||
|
|
########################################################################
|
||
|
|
## BERT panel
|
||
|
|
########################################################################
|
||
|
|
tabPanel(
|
||
|
|
"BERT",
|
||
|
|
|
||
|
|
fluidRow(
|
||
|
|
column(
|
||
|
|
3,
|
||
|
|
br(),
|
||
|
|
h3("Train new BERT"),
|
||
|
|
fluidRow(
|
||
|
|
column(6, actionButton("bertTrain", "Train BERT", width = "100%")),
|
||
|
|
column(6, checkboxInput("bertOptions", "Options"))),
|
||
|
|
fluidRow(
|
||
|
|
column(6, textInput(
|
||
|
|
"bertSaveName", NULL, placeholder = "(save name)")),
|
||
|
|
column(6, actionButton("bertSave", "Save", width = "100%"))),
|
||
|
|
actionLink("bertLast", "Last trained model"),
|
||
|
|
h3("Saved models"),
|
||
|
|
uiOutput("bertSaved")
|
||
|
|
),
|
||
|
|
column(
|
||
|
|
9,
|
||
|
|
br(),
|
||
|
|
conditionalPanel(
|
||
|
|
"input.bertOptions",
|
||
|
|
fluidRow(
|
||
|
|
column(6, selectInput(
|
||
|
|
"bertModel", "Model",
|
||
|
|
c("(Fr) CamemBERT-base" = "camembert/camembert-base",
|
||
|
|
"(Fr) CamemBERT-large" = "camembert/camembert-large",
|
||
|
|
"(Fr) FlauBERT-small" = "flaubert/flaubert_small_cased",
|
||
|
|
"(Fr) FlauBERT-base" = "flaubert/flaubert_base_cased",
|
||
|
|
"(Fr) FlauBERT-large" = "flaubert/flaubert_large_cased",
|
||
|
|
"(En) DistilBERT-base" = "distilbert-base-cased",
|
||
|
|
"(En) RoBERTa-base" = "roberta-base",
|
||
|
|
"(En) DeBERTa-base" = "microsoft/deberta-base",
|
||
|
|
"(Multi) DistilmBERT-base" = "distilbert-base-multilingual-cased",
|
||
|
|
"(Multi) MiniLM" = "microsoft/Multilingual-MiniLM-L12-H384",
|
||
|
|
"(Multi) XLM-RoBERTa-base" = "xlm-roberta-base"))),
|
||
|
|
column(6)
|
||
|
|
),
|
||
|
|
fluidRow(
|
||
|
|
column(3, numericInput("bertEpochs", "Epochs", 3, 1, 20, 1)),
|
||
|
|
column(3, numericInput("bertLrate", "Learning rate", 2e-5, 1e-6, 1, 1e-6)),
|
||
|
|
column(3, numericInput("bertWdecay", "Weight decay", 0.01, 0, 10, 1e-6)),
|
||
|
|
column(3)
|
||
|
|
),
|
||
|
|
fluidRow(
|
||
|
|
column(3, numericInput("bertBatchsize", "Batch size", 4, 1, 32, 1)),
|
||
|
|
column(3, numericInput("bertGradacc", "Gradient accum.", 4, 1, 32, 1)),
|
||
|
|
column(3, br(), checkboxInput("bertAdapt", "Adapt token length to batch", TRUE)),
|
||
|
|
column(3)
|
||
|
|
),
|
||
|
|
fluidRow(
|
||
|
|
column(3, numericInput("bertValidFrac", "Validation fraction", .2, 0, .9)),
|
||
|
|
column(3, numericInput("bertValidSeed", "Validation seed", 1234, 1, 9e8)),
|
||
|
|
column(3, numericInput("bertNeval", "N. validation evals", 10, 1, 100, 1)),
|
||
|
|
column(3, br(), checkboxInput("bertBest", "Keep best", TRUE))
|
||
|
|
),
|
||
|
|
fluidRow(
|
||
|
|
column(3, numericInput("bertMinOccur", "Min. class occurences", 1, 1, 1e4, 1)),
|
||
|
|
column(3, br(), checkboxInput("bertBalance", "Balance classes", FALSE)),
|
||
|
|
column(3),
|
||
|
|
column(3)
|
||
|
|
)
|
||
|
|
),
|
||
|
|
|
||
|
|
|
||
|
|
fluidRow(
|
||
|
|
column(
|
||
|
|
6,
|
||
|
|
# flowLayout(
|
||
|
|
# actionButton(
|
||
|
|
# "bertGoPred", "Infer on current data", width = "100%"),
|
||
|
|
# actionButton(
|
||
|
|
# "bertDelete", "Delete saved model", width = "100%")),
|
||
|
|
verbatimTextOutput("bertMsg")),
|
||
|
|
column(6, plotOutput("bertValPlot", height = 200))),
|
||
|
|
|
||
|
|
verbatimTextOutput("bertMsgHyperpar"),
|
||
|
|
|
||
|
|
DT::dataTableOutput("bertValstats")
|
||
|
|
)
|
||
|
|
)
|
||
|
|
),
|
||
|
|
|
||
|
|
########################################################################
|
||
|
|
## Export panel
|
||
|
|
########################################################################
|
||
|
|
tabPanel(
|
||
|
|
"Export",
|
||
|
|
h4("Export tagged data"),
|
||
|
|
p("Download the tags and predicted probabilities from the complete model, on the current data sample."),
|
||
|
|
# downloadButton("downloadCsv", "Save csv"),
|
||
|
|
flowLayout(
|
||
|
|
selectInput(
|
||
|
|
"dlTagSelect", NULL, c("tags", "comments", "predictions"),
|
||
|
|
c("tags", "comments", "predictions"), multiple = TRUE),
|
||
|
|
selectInput("dlTagFormat", NULL, c("csv", "feather"), "csv"),
|
||
|
|
downloadButton("dlTagSave", NULL, title = "Save tags")
|
||
|
|
),
|
||
|
|
|
||
|
|
hr(),
|
||
|
|
h4("Export embeddings"),
|
||
|
|
p("Download the embeddings (incl. from visualization if present), on the current data sample."),
|
||
|
|
flowLayout(
|
||
|
|
selectInput(
|
||
|
|
"dlEmbedSelect", NULL, c("FastText" = "ft", "SBERT" = "sb"),
|
||
|
|
selected = "sb", multiple = TRUE),
|
||
|
|
selectInput("dlEmbedFormat", NULL, c("csv", "feather"), "feather"),
|
||
|
|
downloadButton("dlEmbedSave", NULL, title = "Save embeddings")
|
||
|
|
),
|
||
|
|
|
||
|
|
hr(),
|
||
|
|
h4("Export BERT predictions"),
|
||
|
|
p("Download the predicted probabilities from the chosen BERT model, on the complete dataset."),
|
||
|
|
flowLayout(
|
||
|
|
selectInput("dlBPSelect", NULL, NULL, NULL),
|
||
|
|
selectInput("dlBPFormat", NULL, c("csv", "feather"), "feather"),
|
||
|
|
actionButton("dlBPInfer", "Predict"),
|
||
|
|
verbatimTextOutput("dlBPMsg"),
|
||
|
|
uiOutput("dlBPDlButton")
|
||
|
|
),
|
||
|
|
|
||
|
|
hr(),
|
||
|
|
h4("Export BERT models")
|
||
|
|
)
|
||
|
|
|
||
|
|
),
|
||
|
|
br(), br(), br(), br(), br(), br(), br(), br(), br(), br(), br(), br(), br(),
|
||
|
|
br(), br(), br(), br(), br(), br()
|
||
|
|
))
|