Plot embeddings from Next Pilot
Molly Lewis
2017-03-26
Read in raw trial data
p1_young= read.csv("/Documents/GRADUATE_SCHOOL/Projects/next_kids/experiments/nextKids_pilot1/data/next/3-4yo/participant_data_34.csv") %>%
mutate(exp = 1,
age_group = "young")
p1_mid= read.csv("/Documents/GRADUATE_SCHOOL/Projects/next_kids/experiments/nextKids_pilot1/data/next/5-6yo/participant_data_56.csv") %>%
mutate(exp = 1,
age_group = "mid")
p2_young= read.csv("/Documents/GRADUATE_SCHOOL/Projects/next_kids/experiments/nextKids_pilot2/data/3-4yo/participant_data_34.csv") %>%
mutate(exp = 2,
age_group = "young")
p2_mid= read.csv("/Documents/GRADUATE_SCHOOL/Projects/next_kids/experiments/nextKids_pilot2/data/5-6yo/participant_data_56.csv") %>%
mutate(exp = 2,
age_group = "mid")
p2_old= read.csv("/Documents/GRADUATE_SCHOOL/Projects/next_kids/experiments/nextKids_pilot2/data/7-8yo/participant_data_78.csv") %>%
mutate(exp = 2,
age_group = "old")Merge and munge
d = rbind(p1_young, p1_mid, p2_young, p2_mid, p2_old) %>%
filter(alg_id == "UncertaintySampling") %>%
mutate(participant_uid, subid = lapply(str_split(participant_uid, "_"),
function(x) {x[2]})) %>%
select(exp, subid, age_group, target_center, target_left, target_right, target_winner) %>%
mutate_each(funs(lapply(str_split(., ".com/|.jpg"),
function(x) {x[2]})), -exp, -age_group, -subid) %>%
mutate_each(funs(as.factor(unlist(.))),-exp, -age_group) %>%
mutate(age_group = fct_relevel(age_group, "young", "mid"))
d = filter(d, subid != "7smVlvhKMXzrayO8MtwFKuo23EWP9N")Demographics
Number of participants
d %>%
group_by(subid, age_group,exp) %>%
slice(1) %>%
group_by(exp, age_group) %>%
summarize(n = n()) %>%
kable()| exp | age_group | n |
|---|---|---|
| 1 | young | 5 |
| 1 | mid | 3 |
| 2 | young | 12 |
| 2 | mid | 8 |
| 2 | old | 3 |
Number of trials
d %>%
group_by(exp, age_group) %>%
summarize(n = n()) %>%
kable()| exp | age_group | n |
|---|---|---|
| 1 | young | 25 |
| 1 | mid | 15 |
| 2 | young | 223 |
| 2 | mid | 163 |
| 2 | old | 56 |
Munge data for python
Convert names to numeric (for python)
#images = unique(d$i)
#image_dict = data.frame(images = images, image_num = 0:(length(images)-1))
#write.csv(image_dict, "image_dict.csv", row.names = F)
image_dict = read.csv("image_dict.csv")d = d %>%
mutate(i = target_center,
j = ifelse(target_left == target_winner, as.character(target_right), as.character(target_left)),
k = target_winner)
d_numeric = d %>%
select(exp, age_group, i, j, k) %>%
left_join(select(image_dict, images, image_num), by=c("i"="images")) %>%
select(-i) %>%
rename(i = image_num) %>%
left_join(select(image_dict, images, image_num), by=c("j"="images")) %>%
select(-j) %>%
rename(j = image_num) %>%
left_join(select(image_dict, images, image_num), by=c("k"="images")) %>%
select(-k) %>%
rename(k = image_num)Save for python
write.csv(d_numeric, "triplet_data.csv", row.names = F)Run get_embeddings.ipynb to get embeddings (computeEmbeddings python) jupyter notebook get_embeddings.ipynb.
Embeddings
Uncertainty E1 and E2
es = read.csv("all_embeddings_uncertainty.csv") %>%
rename(image_num = X.1) %>%
mutate(age_group = fct_relevel(age_group, "young", "mid")) %>%
left_join(image_dict)Points
ggplot(es, aes(x = X, y = Y, color = type, shape = theme)) +
geom_point(size = 3) +
facet_wrap(~age_group) +
theme_bw() +
theme(panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
strip.text.x = element_text(size = 14),
strip.background = element_rect(colour="grey", fill="grey"),
axis.title=element_blank(),
axis.text=element_blank(),
axis.ticks=element_blank(),
legend.position = "bottom")
Labels
ggplot(es, aes(x = X, y = Y)) +
geom_text(aes(label=images), size = 3) +
facet_wrap(~age_group) +
theme_bw() +
theme(panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
strip.text.x = element_text(size = 14),
strip.background = element_rect(colour="grey", fill="grey"),
axis.title=element_blank(),
axis.text=element_blank(),
axis.ticks=element_blank())
All E1 and E2
es = read.csv("all_embeddings_all.csv") %>%
rename(image_num = X.1) %>%
mutate(age_group = fct_relevel(age_group, "young", "mid")) %>%
left_join(image_dict)Points
ggplot(es, aes(x = X, y = Y, color = type, shape = theme)) +
geom_point(size = 3) +
facet_wrap(~age_group) +
theme_bw() +
theme(panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
strip.text.x = element_text(size = 14),
strip.background = element_rect(colour="grey", fill="grey"),
axis.title=element_blank(),
axis.text=element_blank(),
axis.ticks=element_blank(),
legend.position = "bottom")
Labels
ggplot(es, aes(x = X, y = Y)) +
geom_text(aes(label=images), size = 3) +
facet_wrap(~age_group) +
theme_bw() +
theme(panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
strip.text.x = element_text(size = 14),
strip.background = element_rect(colour="grey", fill="grey"),
axis.title=element_blank(),
axis.text=element_blank(),
axis.ticks=element_blank())
Uncertainty E2
es = read.csv("all_embeddings_2_uncertainty.csv") %>%
rename(image_num = X.1) %>%
mutate(age_group = fct_relevel(age_group, "young", "mid")) %>%
left_join(image_dict)Points
ggplot(es, aes(x = X, y = Y, color = type, shape = theme)) +
geom_point(size = 3) +
facet_wrap(~age_group) +
theme_bw() +
theme(panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
strip.text.x = element_text(size = 14),
strip.background = element_rect(colour="grey", fill="grey"),
axis.title=element_blank(),
axis.text=element_blank(),
axis.ticks=element_blank(),
legend.position = "bottom")
Labels
ggplot(es, aes(x = X, y = Y)) +
geom_text(aes(label=images), size = 3) +
facet_wrap(~age_group) +
theme_bw() +
theme(panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
strip.text.x = element_text(size = 14),
strip.background = element_rect(colour="grey", fill="grey"),
axis.title=element_blank(),
axis.text=element_blank(),
axis.ticks=element_blank())
Next Uncertainty E2
p2e_young= flatten(fromJSON("/Documents/GRADUATE_SCHOOL/Projects/next_kids/experiments/nextKids_pilot2/data/3-4yo/embedding_34_uncertainty.txt")) %>%
select(darray, target.alt_description) %>%
mutate(exp = 2,
age_group = "young")
p2e_mid= flatten(fromJSON("/Documents/GRADUATE_SCHOOL/Projects/next_kids/experiments/nextKids_pilot2/data/5-6yo/embedding_56_uncertainty.txt"))%>%
select(darray, target.alt_description) %>%
mutate(exp = 2,
age_group = "mid")
p2e_old= flatten(fromJSON("/Documents/GRADUATE_SCHOOL/Projects/next_kids/experiments/nextKids_pilot2/data/7-8yo/embedding_78_uncertainty.txt"))%>%
select(darray, target.alt_description) %>%
mutate(exp = 2,
age_group = "old")
next_embed = rbind(p2e_young, p2e_mid, p2e_old) %>%
mutate(images = unlist(lapply(str_split(target.alt_description, ".jpg"),
function(x) {x[1]}))) %>%
select(-target.alt_description) %>%
separate(darray, c("X", "Y"), ",") %>%
mutate(X = as.numeric(gsub( "c\\(", "", X)),
Y = as.numeric(gsub("\\)", "", Y)),
age_group = as.factor(age_group),
images = as.factor(images)) %>%
mutate(age_group = fct_relevel(age_group, "young", "mid")) %>%
left_join(image_dict) %>%
select(-image_num)Points
ggplot(next_embed, aes(x = X, y = Y, color = type, shape = theme)) +
geom_point(size = 3) +
facet_wrap(~age_group) +
theme_bw() +
theme(panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
strip.text.x = element_text(size = 14),
strip.background = element_rect(colour="grey", fill="grey"),
axis.title=element_blank(),
axis.text=element_blank(),
axis.ticks=element_blank(),
legend.position = "bottom")
Labels
ggplot(next_embed, aes(x = X, y = Y)) +
geom_text(aes(label=images), size = 3) +
facet_wrap(~age_group) +
theme_bw() +
theme(panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
strip.text.x = element_text(size = 14),
strip.background = element_rect(colour="grey", fill="grey"),
axis.title=element_blank(),
axis.text=element_blank(),
axis.ticks=element_blank(),)