Turning analysis
RAF-CIN
1 Purpose
- Predict when rats turn left on right using deeplabcut pose estimates
2 Report overview
Source code and data found at https://github.com/maRce10/cage-test_cin_rats_2021-
3 Detecting half-turns with DLCAnalyzerc
3.1 Analyze a single file
Code
Tracking <- ReadDLCDataFromCSV(file = "./data/raw/raf_deeplabcut_csvs/RAFII-01- AMPHDLC_resnet50_rafi_ratas_cinJan8shuffle1_90000.csv",
fps = 20)
names(Tracking$data)
head(Tracking$data$nose)
Tracking <- CleanTrackingData(Tracking, likelihoodcutoff = 0.95)
body_points <- c("nose", "ear_left", "ear_right", "body_center", "base_tail",
"tip_tail")
PlotPointData(Tracking, points = body_points)
Tracking <- CalculateMovement(Tracking, movement_cutoff = 5, integration_period = 5)
head(Tracking$data$body_center)
plots <- PlotDensityPaths(Tracking, points = body_points)
plots$body_center
Tracking <- OFTAnalysis(Tracking, points = body_points, movement_cutoff = 5,
integration_period = 5)
t(data.frame(Tracking$Report))3.2 Analyze multiple files
Code
input_folder <- "./data/raw/raf_deeplabcut_csvs/"
files <- list.files(input_folder)
files
pipeline <- function(path) {
Tracking <- ReadDLCDataFromCSV(path, fps = 20)
Tracking <- CleanTrackingData(Tracking, likelihoodcutoff = 0.95)
Tracking <- CalculateMovement(Tracking, movement_cutoff = 5, integration_period = 5)
Tracking <- OFTAnalysis(Tracking, points = "bodycentre", movement_cutoff = 5,
integration_period = 5)
return(Tracking)
}
TrackingAll <- RunPipeline(files, input_folder, FUN = pipeline)3.3 Train an ML classifier
Code
labeling.data <- read.table("./scripts/FSTLables_Oliver.csv", sep = ";",
header = T)
pointinfo <- read.table("./scripts/FST_pointinfo.csv", sep = ";",
header = T)
files <- unique(labeling.data$CSVname)
path <- "./scripts/DLC_Data/"
pipeline <- function(path) {
Tracking <- ReadDLCDataFromCSV(path, fps = 25)
Tracking <- CutTrackingData(Tracking, start = 300, end = 300)
Tracking <- CalibrateTrackingData(Tracking, "distance", in.metric = 20,
c("t", "b"))
Tracking <- CleanTrackingData(Tracking, likelihoodcutoff = 0.95)
Tracking <- AddLabelingData(Tracking, labeling.data[labeling.data$CSVname ==
Tracking$filename, ])
Tracking <- CalculateAccelerations(Tracking)
Tracking <- CreateAccelerationFeatures(Tracking)
Tracking <- CreateTrainingSet(Tracking, integration_period = 20)
Tracking <- AddPointInfo(Tracking, pointinfo)
Tracking <- FSTAnalysis(Tracking, cutoff_floating = 0.03, integration_period = 5,
points = "bodycentre", Object = "Mouse")
return(Tracking)
}Code
evaluation <- list()
for (i in names(TrackingAll)) {
# combine training data from all files except the one we
# want to evaluate
MLData <- CombineTrainingsData(TrackingAll[names(TrackingAll) !=
i], shuffle = TRUE)
# initialize model
model <- keras_model_sequential()
model %>%
layer_dense(units = 256, activation = "relu", input_shape = c(MLData$parameters$N_input),
kernel_regularizer = regularizer_l2(l = 0)) %>%
layer_dropout(rate = 0.4) %>%
layer_dense(units = 128, activation = "relu", kernel_regularizer = regularizer_l2(l = 0)) %>%
layer_dropout(rate = 0.3) %>%
layer_dense(units = MLData$parameters$N_features, activation = "softmax")
# define optimizer
model %>%
compile(loss = "categorical_crossentropy", optimizer = optimizer_rmsprop(),
metrics = c("accuracy"))
# train model
history <- model %>%
fit(MLData$train_x, MLData$train_y, epochs = 5, batch_size = 32,
validation_split = 0)
# we now use this model to classify the behavior in the
# cross validation file
TrackingAll[[i]] <- ClassifyBehaviors(t = TrackingAll[[i]], model,
model_parameters = MLData$parameters)
}
PlotLabels(t = TrackingAll$FST_1.csv)
PlotLabels(TrackingAll$FST_2.csv)
EvaluateClassification(ts = TrackingAll)
PlotLabels.Multi.PDF(TrackingAll)3.4 Training a classifier and cross validating it
Code
labeling.data <- read.table("./scripts/FSTLables_Oliver.csv", sep = ";",
header = T)
pointinfo <- read.table("./scripts/FST_pointinfo.csv", sep = ";",
header = T)
files <- unique(labeling.data$CSVname)
path <- "./scripts/DLC_Data/"
pipeline <- function(path) {
Tracking <- ReadDLCDataFromCSV(path, fps = 25)
Tracking <- CutTrackingData(Tracking, start = 300, end = 300)
Tracking <- CalibrateTrackingData(Tracking, "distance", in.metric = 20,
c("t", "b"))
Tracking <- CleanTrackingData(Tracking, likelihoodcutoff = 0.95)
Tracking <- AddLabelingData(Tracking, labeling.data[labeling.data$CSVname ==
Tracking$filename, ])
Tracking <- CalculateAccelerations(Tracking)
Tracking <- CreateAccelerationFeatures(Tracking)
Tracking <- CreateTrainingSet(Tracking, integration_period = 20)
Tracking <- AddPointInfo(Tracking, pointinfo)
Tracking <- FSTAnalysis(Tracking, cutoff_floating = 0.03, integration_period = 5,
points = "bodycentre", Object = "Mouse")
return(Tracking)
}
TrackingAll <- RunPipeline(files, path, FUN = pipeline)Code
evaluation <- list()
for (i in names(TrackingAll)) {
# combine training data from all files except the one we
# want to evaluate
MLData <- CombineTrainingsData(TrackingAll[names(TrackingAll) !=
i], shuffle = TRUE)
# initialize model
model <- keras_model_sequential()
model %>%
layer_dense(units = 256, activation = "relu", input_shape = c(MLData$parameters$N_input),
kernel_regularizer = regularizer_l2(l = 0)) %>%
layer_dropout(rate = 0.4) %>%
layer_dense(units = 128, activation = "relu", kernel_regularizer = regularizer_l2(l = 0)) %>%
layer_dropout(rate = 0.3) %>%
layer_dense(units = MLData$parameters$N_features, activation = "softmax")
# define optimizer
model %>%
compile(loss = "categorical_crossentropy", optimizer = optimizer_rmsprop(),
metrics = c("accuracy"))
# train model
history <- model %>%
fit(MLData$train_x, MLData$train_y, epochs = 5, batch_size = 32,
validation_split = 0)
# we now use this model to classify the behavior in the
# cross validation file
TrackingAll[[i]] <- ClassifyBehaviors(TrackingAll[[i]], model,
MLData$parameters)
}4 DLCAnalyzer on CIN data
4.1 Organizing data
Code
csvs <- list.files(path = "./data/processed/solomon_turn_annotations",
pattern = "csv$", full.names = TRUE)
out <- warbleR:::pblapply_wrblr_int(X = csvs, cl = 10, function(x) {
# print()
y <- read.csv(x, header = FALSE, skip = 1)
y$ms <- sapply(strsplit(y[, 2], ";"), "[", 1)
y$behav <- sapply(strsplit(y[, 2], ";"), "[", 2)
y$type <- ifelse(y$behav == "Inmovilidad", "left", ifelse(y$behav ==
"Digging", "right", NA))
y$CSVname <- basename(x)
y$time <- y$V1 + as.numeric(y$ms)/100
y$V2 <- y$V1 <- y$ms <- y$behav <- NULL
y$start <- ifelse(is.na(y$type), y$time - 0.04, y$time - 0.055)
y$end <- ifelse(is.na(y$type), y$time + 0.04, y$time + 0.055)
y$sound.files <- y$CSVname
y$selec <- 1:nrow(y)
mo <- merge_overlaps(y, pb = FALSE)
mo$from <- mo$start + 0.055
mo$to <- mo$end - 0.055
mo$start <- mo$end <- mo$selec <- mo$sound.files <- mo$time <- NULL
mo <- mo[!is.na(mo$type), ]
return(mo)
})
turn_annot <- do.call(rbind, out)
dlc_files <- list.files(path = "./data/raw/raf_deeplabcut_csvs", pattern = "\\.csv$")
turn_annot$Experimento <- ifelse(grepl("RAFII", turn_annot$CSVname),
"RAFII", "RAFI")
turn_annot$Animal <- substring(0, 2, text = gsub("RAFI-|RAFII-", "",
turn_annot$CSVname))
turn_annot$dia.num <- NA
turn_annot$dia.num[grep("DÍA 1.csv", turn_annot$CSVname)] <- 1
turn_annot$dia.num[grep("-1.csv", turn_annot$CSVname)] <- -1
turn_annot$dia.num[grep("7.csv", turn_annot$CSVname)] <- 7
turn_annot$dia.num[grep("21.csv|APO", turn_annot$CSVname)] <- 21
turn_annot$dia.num[grep("AMP", turn_annot$CSVname)] <- 14
turn_annot[!duplicated(turn_annot$CSVname), ]
turn_annot$id <- paste0(turn_annot$Experimento, "-", turn_annot$Animal,
" DÍA ", turn_annot$dia.num)
turn_annot$id[turn_annot$id == "RAFII-13 DÍA 21"] <- "RAFII-13 APO"
turn_annot$id[turn_annot$id == "RAFII-1 DÍA 14"] <- "RAFII-01 PRE AMP"
turn_annot$id[turn_annot$id == "RAFII-8 DÍA 1"] <- "RAFII-08 DÍA 1"
turn_annot$id[turn_annot$id == "RAFII-15 DÍA 14"] <- "RAFII-15 AMP"
turn_annot$id[turn_annot$id == "RAFII-4 DÍA 21"] <- "RAFII-04 APO"
turn_annot$id[turn_annot$id == "RAFII-9 DÍA 1"] <- "RAFII-09 DÍA 1"
turn_annot$id[turn_annot$id == "RAFII-11 DÍA 21"] <- "RAFII-11 APO"
turn_annot$id[turn_annot$id == "RAFI-10 DÍA 21"] <- "RAFI-10_pre_APO"
turn_annot$id[turn_annot$id == "RAFII-2 DÍA 1"] <- "RAFII-02 DÍA 1"
turn_annot$id[turn_annot$id == "RAFII-3 DÍA 14"] <- "RAFII-03 AMP"
turn_annot$id[turn_annot$id == "RAFII-6 DÍA 14"] <- "RAFII-06 PRE AMPH"
turn_annot$id[turn_annot$id == "RAFII-7 DÍA 21"] <- "RAFII-07 PRE APO"
turn_annot$id[turn_annot$id == "RAFII-14 DÍA 14"] <- "RAFII-14 PRE AMPH"
turn_annot$id[turn_annot$id == "RAFII-16 DÍA 21"] <- "RAFII-16 APO"
turn_annot$id[turn_annot$id == "RAFII-01 DÍA 14"] <- "RAFII-01 AMPH"
turn_annot$id[turn_annot$id == "RAFII-03 DÍA 14"] <- "RAFII-03 AMPH"
turn_annot$id[turn_annot$id == "RAFII-04 DÍA 21"] <- "RAFII-04 APO"
turn_annot$id[turn_annot$id == "RAFII-06 DÍA 14"] <- "RAFII-06 AMP"
turn_annot$id[turn_annot$id == "RAFII-07 DÍA 21"] <- "RAFII-07 APO"
turn_annot$dlc.file <- sapply(turn_annot$id, function(x) {
y <- grep(x, dlc_files, value = TRUE)[1]
if (length(y) == 0)
y <- NA
if (length(y) > 1)
y <- length(y)
return(y)
})
turn_annot$file <- turn_annot$CSVname
turn_annot$CSVname <- turn_annot$dlc.file
turn_annot$Experimenter <- "Jaime_Fornaguera"
turn_annot$ID <- turn_annot$id
turn_annot$dlc.file <- turn_annot$id <- NULL
turn_annot <- turn_annot[, c("file", "Experimenter", "from", "to",
"type", "ID", "CSVname")]
head(turn_annot)
write.csv(turn_annot, "./data/raw/turning_annotations_fixed.csv")Code
labeling.data <- read.csv("./data/raw/turning_annotations_fixed.csv")
pointinfo <- data.frame(PointName = c("nose", "ear_left", "ear_right",
"body_center", "base_tail", "tip_tail"), PointType = "Mouse")
files <- unique(labeling.data$CSVname)
path <- "./data/raw/raf_deeplabcut_csvs/"
pipeline <- function(path) {
Tracking <- ReadDLCDataFromCSV(path, fps = 20)
# Tracking <- ReadDLCDataFromCSV(file.path(path, files[1]),
# fps = 20) keep 5 s to 5 min frames (20 fps)
Tracking <- CutTrackingData(Tracking, keep.frames = 100:6000)
# print(length(Tracking$frames)) Tracking <-
# CalibrateTrackingData(Tracking, 'distance',in.metric = 20,
# c('t','b'))
Tracking <- CleanTrackingData(Tracking, likelihoodcutoff = 0.95)
Tracking <- AddLabelingData(Tracking, labeling.data[labeling.data$CSVname ==
Tracking$filename, ])
Tracking <- CalculateMovement(Tracking, movement_cutoff = 5, integration_period = 5)
Tracking <- CalculateAccelerations(Tracking)
Tracking <- CreateAccelerationFeatures(Tracking)
Tracking <- create_speed_features(Tracking) # my own function
Tracking <- create_distance_features(Tracking, point.pairs = c("nose/body_center",
"nose/tip_tail", "nose/base_tail", "ear_right/tip_tail", "ear_right/base_tail",
"ear_left/base_tail", "ear_left/tip_tail", "body_center/tip_tail",
"body_center/base_tail")) # my own function
Tracking <- CreateSkeletonData_OFT(Tracking) # modified
Tracking <- create_angle_features(Tracking) # my own function
Tracking <- CreateTrainingSet(Tracking, integration_period = 40)
Tracking <- AddPointInfo(Tracking, pointinfo)
sapply(Tracking, function(x) if (is.list(x))
sapply(x, head) else head(x))
return(Tracking)
}
TrackingAll <- RunPipeline(files, path, FUN = pipeline)
saveRDS(TrackingAll, "./data/processed/tracking_data_dlc_analyzer_raf.RDS")4.2 Tensorflow training
4.2.1 Without cross-validation
Code
TrackingAll <- readRDS("./data/processed/tracking_data_dlc_analyzer_raf.RDS")
# evaluation <- list() for(i in names(TrackingAll)){ combine
# training data from all files
MLData <- CombineTrainingsData(TrackingAll, shuffle = FALSE)
saveRDS(MLData, "./data/processed/training_data_for_ML.RDS")
# subsample data to balance it out
counts <- apply(MLData$train_y, 2, sum)
mean(counts[-2])
subsample <- c(which(MLData$train_y[, 1] == 1), which(MLData$train_y[,
3] == 1), sample(which(MLData$train_y[, 2] == 1), mean(counts[-2])))
length(subsample)
MLData$train_y <- MLData$train_y[subsample, ]
MLData$train_x <- MLData$train_x[subsample, ]
# sapply(MLData, function(x) if(is.list(x)) sapply(x, head) else
# head(x))
if (anyNA(MLData$train_y) | anyNA(MLData$train_x)) stop("NAs in data")
# initialize model
model <- keras_model_sequential()
model %>%
layer_dense(units = 256, activation = "relu", input_shape = c(MLData$parameters$N_input),
kernel_regularizer = regularizer_l2(l = 0)) %>%
layer_dropout(rate = 0.4) %>%
layer_dense(units = 128, activation = "relu", kernel_regularizer = regularizer_l2(l = 0)) %>%
layer_dropout(rate = 0.3) %>%
layer_dense(units = MLData$parameters$N_features, activation = "softmax")
# define optimizer
model %>%
compile(loss = "categorical_crossentropy", optimizer = optimizer_rmsprop(),
metrics = c("accuracy"))
# train model
history <- model %>%
fit(MLData$train_x, MLData$train_y, epochs = 50, batch_size = 32,
validation_split = 0)
# #we now use this model to classify the behavior in the cross
# validation file
for (i in names(TrackingAll)) TrackingAll[[i]] <- ClassifyBehaviors(t = TrackingAll[[i]],
model, model_parameters = MLData$parameters)
# # } PlotLabels(t =
# TrackingAll$`RAFI-10_pre_APODLC_resnet50_rafi_ratas_cinJan8shuffle1_90000.csv`)
# PlotLabels(TrackingAll$`RAFII-01
# AMPHDLC_resnet50_rafi_ratas_cinJan8shuffle1_90000.csv`)
saveRDS(TrackingAll, "./data/processed/tracking_data_dlc_analyzer_raf_evaluated.RDS")4.2.2 With cross-validation
Code
evaluation <- list()
for (i in names(TrackingAll)) {
# combine training data from all files except the one we
# want to evaluate
MLData <- CombineTrainingsData(TrackingAll[names(TrackingAll) !=
i], shuffle = TRUE)
# initialize model
model <- keras_model_sequential()
model %>%
layer_dense(units = 256, activation = "relu", input_shape = c(MLData$parameters$N_input),
kernel_regularizer = regularizer_l2(l = 0)) %>%
layer_dropout(rate = 0.4) %>%
layer_dense(units = 128, activation = "relu", kernel_regularizer = regularizer_l2(l = 0)) %>%
layer_dropout(rate = 0.3) %>%
layer_dense(units = MLData$parameters$N_features, activation = "softmax")
# define optimizer
model %>%
compile(loss = "categorical_crossentropy", optimizer = optimizer_rmsprop(),
metrics = c("accuracy"))
# train model
history <- model %>%
fit(MLData$train_x, MLData$train_y, epochs = 20, batch_size = 32,
validation_split = 0)
# we now use this model to classify the behavior in the
# cross validation file
TrackingAll[[i]] <- ClassifyBehaviors(t = TrackingAll[[i]], model,
model_parameters = MLData$parameters)
}
PlotLabels(t = TrackingAll$`RAFI-10_pre_APODLC_resnet50_rafi_ratas_cinJan8shuffle1_90000.csv`)
PlotLabels(TrackingAll$`RAFII-01 AMPHDLC_resnet50_rafi_ratas_cinJan8shuffle1_90000.csv`)
EvaluateClassification(ts = TrackingAll)
PlotLabels.Multi.PDF(TrackingAll)5 Random Forest on CIN data
5.1 Training model
Code
TrackingAll <- readRDS("./data/processed/tracking_data_dlc_analyzer_raf.RDS")
MLData <- readRDS("./data/processed/training_data_for_ML.RDS")
# subsample data to balance it out
counts <- apply(MLData$train_y, 2, sum)
mean(counts[-2])
subsample <- c(which(MLData$train_y[, 1] == 1), which(MLData$train_y[,
3] == 1), sample(which(MLData$train_y[, 2] == 1), mean(counts[-2])))
length(subsample)
MLData$train_y <- MLData$train_y[subsample, ]
MLData$train_x <- MLData$train_x[subsample, ]
MLData$labels <- MLData$labels[subsample]
dat <- data.frame(MLData$train_x, labels = MLData$labels)
dat$labels <- as.factor(dat$labels)
# untunned random forest
rf <- randomForest::randomForest(labels ~ ., data = dat, ntree = 1000)
saveRDS(rf, "./data/processed/untunned_random_forest.RDS")
# tunning random forest
library(doParallel)
registerDoParallel(cores = 16)
# Random Search
control <- trainControl(method = "oob", number = 10, search = "random",
allowParallel = TRUE)
set.seed(123)
mtry <- sqrt(ncol(x))
rf_random <- train(labels ~ ., data = dat, method = "rf", metric = "Accuracy",
tuneLength = 15, trControl = control)
new_dat <- as.data.frame(MLData$train_x)
names(new_dat) <- names(rf_random$trainingData)[-1]
predicted_class <- predict(rf_random, newdata = new_dat)
# save confusion matrix
conf_mat <- confusionMatrix(predicted_class, as.factor(MLData$labels))
conf_df <- as.data.frame(conf_mat$table)
conf_df$total <- sapply(conf_df$Reference, function(x) sum(MLData$labels ==
x))
conf_df$proportion <- conf_df$Freq/conf_df$total
saveRDS(list(predicted_class = predicted_class, conf_mat = conf_mat,
conf_df = conf_df, new_dat = new_dat, files = MLData$file.names,
class = MLData$labels), "./data/processed/tunned_random_forest_model_results.RDS")
saveRDS(rf_random, "./data/processed/tunned_random_forest_model.RDS")5.2 Results
5.2.1 Accuracy by class
Code
rf_results <- readRDS("./data/processed/tunned_random_forest_model_results.RDS")
ggplot(rf_results$conf_df, aes(x = Reference, y = Prediction, fill = proportion)) +
geom_tile() + theme_bw() + coord_equal() + scale_fill_distiller(palette = "Greens",
direction = 1) + geom_text(aes(label = round(proportion, 2)),
color = "black") + theme_classic() + theme(axis.text.x = element_text(color = "black",
size = 11, angle = 30, vjust = 0.8, hjust = 0.8))
Confusion Matrix and Statistics
Reference
Prediction left None right
left 33708 10144 0
None 0 106441 57
right 0 6177 23924
Overall Statistics
Accuracy : 0.9092
95% CI : (0.9079, 0.9106)
No Information Rate : 0.6803
P-Value [Acc > NIR] : < 2.2e-16
Kappa : 0.8291
Mcnemar's Test P-Value : NA
Statistics by Class:
Class: left Class: None Class: right
Sensitivity 1.0000 0.8671 0.9976
Specificity 0.9309 0.9990 0.9605
Pos Pred Value 0.7687 0.9995 0.7948
Neg Pred Value 1.0000 0.7793 0.9996
Prevalence 0.1868 0.6803 0.1329
Detection Rate 0.1868 0.5899 0.1326
Detection Prevalence 0.2430 0.5902 0.1668
Balanced Accuracy 0.9654 0.9330 0.97915.2.2 Prediction by file
These graphs show the observed behavior (left colum) and the predicted by the model (right column) for each video (rows)
Code
dat <- data.frame(file = gsub("_resnet50_rafi_ratas_cinJan8shuffle1_90000.csv",
"", rf_results$files), class = c(as.character(rf_results$predicted_class),
rf_results$class), type = rep(c("predicted", "observed"), each = length(rf_results$files)))
dat$frame <- 1:(unique(table(dat$file))/2)
dat$time <- dat$frame/20
ggplot(data = dat, aes(x = time, y = class, color = class)) + geom_point(size = 4,
shape = 124) + facet_grid(file ~ type, scales = "free_x") + scale_color_viridis_d() +
theme_bw()
Scatterplots of the correlation between number of predicted frames and actual turns observed
Code
turn_annot <- read.csv("./data/raw/turning_annotations_fixed.csv")
turn_annot$file <- gsub("_resnet50_rafi_ratas_cinJan8shuffle1_90000.csv",
"", turn_annot$CSVname)
agg_turns <- aggregate(ID ~ file + type, turn_annot, length)
names(agg_turns)[ncol(agg_turns)] <- "quarter.turns"
agg_turns <- agg_turns[order(agg_turns$file), ]
agg_turns$frames <- sapply(1:nrow(agg_turns), function(x) {
sum(dat$file == agg_turns$file[x] & dat$class == agg_turns$type[x])
})
ggplot(agg_turns, aes(y = frames, x = quarter.turns)) + geom_smooth(method = "lm",
col = viridis(10, alpha = 0.4)[3]) + geom_point(size = 4, pch = 21,
fill = viridis(10, alpha = 0.4)[8]) + facet_wrap(~type, nrow = 1,
scales = "free") + labs(x = "Quarter of turns count (observed)",
y = "Number of frames (predicted)") + theme_classic()
[1] "Overall correlation:"[1] 0.8486071[1] "Left turn correlation:"Code
[1] 0.8627682[1] "Right turn correlation:"Code
[1] 0.75974136 Create label videos (srt files)
Code
library(srt)
x <- read_srt(srt_example(), collapse = " ")
dif <- x$end - x$start
x$start <- x$start - abs(min(x$start))
x$end <- x$start + dif
x <- x[x$end < 960, ]
head(x)
# write_srt(x, tempfile(fileext = '.srt'), wrap = FALSE)
write_srt(x, "~/Downloads/delete.srt", wrap = FALSE)
labeling.data <- read.csv("./data/raw/turning_annotations_fixed.csv")
for (i in unique(labeling.data$CSVname)) {
labs <- labeling.data[labeling.data$CSVname == i, ]
labs$start <- labs$from
labs$end <- labs$to
labs$n <- 1:nrow(labs)
labs$subtitle <- labs$type
file <- gsub("DLC_resnet50_rafi_ratas_cinJan8shuffle1_90000.csv",
"", labs$CSVname[1])
labs <- labs[, c("n", "start", "end", "subtitle")]
labs <- rbind(labs, data.frame(n = nrow(labs) + 1, start = 300,
end = 360, subtitle = "******* END OF MANUAL TRACKING *******"))
write_srt(labs, paste0("/media/m/Seagate Portable Drive/rafII_videos/subtitle_files/",
file, ".srt"), wrap = FALSE)
}
7 Takeaways
- Left and right turns can be predicted with good accuracy form deeplabcut pose estimations
Session information
R version 4.1.0 (2021-05-18)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Ubuntu 20.04.2 LTS
Matrix products: default
BLAS: /usr/lib/x86_64-linux-gnu/atlas/libblas.so.3.10.3
LAPACK: /usr/lib/x86_64-linux-gnu/atlas/liblapack.so.3.10.3
locale:
[1] LC_CTYPE=pt_BR.UTF-8 LC_NUMERIC=C
[3] LC_TIME=es_CR.UTF-8 LC_COLLATE=pt_BR.UTF-8
[5] LC_MONETARY=es_CR.UTF-8 LC_MESSAGES=pt_BR.UTF-8
[7] LC_PAPER=es_CR.UTF-8 LC_NAME=C
[9] LC_ADDRESS=C LC_TELEPHONE=C
[11] LC_MEASUREMENT=es_CR.UTF-8 LC_IDENTIFICATION=C
attached base packages:
[1] stats graphics grDevices utils datasets methods base
other attached packages:
[1] viridis_0.6.2 viridisLite_0.4.1 caret_6.0-88
[4] lattice_0.20-44 srt_1.0.3 adehabitatLT_0.3.25
[7] CircStats_0.2-6 boot_1.3-28 MASS_7.3-54
[10] adehabitatMA_0.3.14 ade4_1.7-17 tensorflow_2.9.0
[13] keras_2.9.0 corrplot_0.90 cowplot_1.1.1
[16] data.table_1.14.0 ggmap_3.0.0 ggplot2_3.4.0
[19] imputeTS_3.3 sp_1.5-1 xaringanExtra_0.7.0
[22] rprojroot_2.0.3 formatR_1.11 knitr_1.42
[25] kableExtra_1.3.4
loaded via a namespace (and not attached):
[1] colorspace_2.0-3 rjson_0.2.21 class_7.3-19
[4] base64enc_0.1-3 gridtext_0.1.5 ggtext_0.1.2
[7] rstudioapi_0.14 farver_2.1.1 remotes_2.4.2
[10] lubridate_1.7.10 prodlim_2019.11.13 fansi_1.0.3
[13] xml2_1.3.3 codetools_0.2-18 splines_4.1.0
[16] zeallot_0.1.0 jsonlite_1.8.4 pROC_1.17.0.1
[19] packrat_0.9.0 png_0.1-7 tfruns_1.5.1
[22] compiler_4.1.0 httr_1.4.4 assertthat_0.2.1
[25] Matrix_1.5-1 fastmap_1.1.0 cli_3.6.0
[28] htmltools_0.5.4 tools_4.1.0 gtable_0.3.1
[31] glue_1.6.2 reshape2_1.4.4 dplyr_1.0.10
[34] Rcpp_1.0.9 fracdiff_1.5-1 vctrs_0.5.2
[37] urca_1.3-3 svglite_2.1.0 nlme_3.1-152
[40] iterators_1.0.13 lmtest_0.9-40 timeDate_3043.102
[43] xfun_0.36 gower_0.2.2 stringr_1.5.0
[46] rvest_1.0.3 lifecycle_1.0.3 zoo_1.8-11
[49] scales_1.2.1 ipred_0.9-11 parallel_4.1.0
[52] RColorBrewer_1.1-3 yaml_2.3.7 quantmod_0.4.20
[55] curl_4.3.3 gridExtra_2.3 reticulate_1.26
[58] rpart_4.1-15 stringi_1.7.12 tseries_0.10-52
[61] foreach_1.5.1 TTR_0.24.3 lava_1.6.9
[64] RgoogleMaps_1.4.5.3 rlang_1.0.6 pkgconfig_2.0.3
[67] systemfonts_1.0.4 bitops_1.0-7 evaluate_0.20
[70] purrr_1.0.0 labeling_0.4.2 stinepack_1.4
[73] recipes_0.1.16 htmlwidgets_1.5.4 tidyselect_1.2.0
[76] plyr_1.8.7 magrittr_2.0.3 R6_2.5.1
[79] generics_0.1.3 DBI_1.1.1 mgcv_1.8-36
[82] pillar_1.8.1 whisker_0.4.1 withr_2.5.0
[85] xts_0.12.2 survival_3.2-11 nnet_7.3-16
[88] tibble_3.1.8 crayon_1.5.2 utf8_1.2.2
[91] rmarkdown_2.20 jpeg_0.1-8.1 grid_4.1.0
[94] sketchy_1.0.2 ModelMetrics_1.2.2.2 forecast_8.18
[97] digest_0.6.31 webshot_0.5.4 tidyr_1.1.3
[100] stats4_4.1.0 munsell_0.5.0 quadprog_1.5-8