public-projects/InferringIntention
1
0
Fork 0
Code Issues Pull requests Releases Wiki Activity

first commit

Browse source
This commit is contained in:
Lei Shi 2024-03-24 23:42:27 +01:00
commit 83b04e2133
109 changed files with 12081 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

132
watch_and_help/stan/plot_user_length.py Normal file
View file

@ -0,0 +1,132 @@
import numpy as np
from numpy import genfromtxt
import matplotlib.pyplot as plt
import argparse
import pathlib
def main(args):
if args.task_type == 'new_test_task':
user = 9
N = 1
if args.task_type == 'test_task':
user = 92
N = 1
rate = 100
widths = [-0.1, 0, 0.1]
user_table = [6, 13, 15, 19, 20, 23, 27, 30, 33, 44, 46, 49, 50, 51, 52, 53, 54, 56, 65, 71, 84]
# read data
model_data_list = []
user_list = []
if not args.plot_user_list:
for i in range(user):
path = "result/"+args.task_type+"/user"+str(user)+"/"+args.loss_type+"/N"+ str(N) + "/" + args.model_type + "_N" + str(N) + "_result_" + str(rate) + "_user" + str(i) +".csv"
data = genfromtxt(path, delimiter=',', skip_header =1)
data = data[[1,2,3,5,6,7,9,10,11],:][:,[2,4,6,7]]
model_data_list.append(data)
if args.task_type == 'test_task':
user_list.append(np.transpose(data[:,[0]]))
else:
for i in range(user):
for t in user_table:
if t == i+1:
path = "result/"+args.task_type+"/user"+str(user)+"/"+args.loss_type+"/N"+ str(N) + "/" + args.model_type + "_N" + str(N) + "_result_" + str(rate) + "_user" + str(i) +".csv"
data = genfromtxt(path, delimiter=',', skip_header =1)
data = data[[1,2,3,5,6,7,9,10,11],:][:,[2,4,6,7]]
model_data_list.append(data)
user_list.append(np.transpose(data[:,[0]]))
color = ['royalblue', 'lightgreen', 'tomato']
legend = ['put fridge', 'put\n dishwasher', 'read book']
fig, axs = plt.subplots(3, sharex=True, sharey=True)
fig.set_figheight(10) # all sample rate: 10; 3 row: 8
fig.set_figwidth(20)
for ax in range(3):
y_total = []
y_low_total = []
y_high_total = []
for j in range(3):
y= []
y_low = []
y_high = []
for i in range(len(model_data_list)):
y.append(model_data_list[i][j+ax*3][0])
y_low.append(model_data_list[i][j+ax*3][2])
y_high.append(model_data_list[i][j+ax*3][3])
y_total.append(y)
y_low_total.append(y_low)
y_high_total.append(y_high)
print()
print("user mean of mean prob: ", np.mean(y))
print("user mean of sd prob: ", np.std(y))
for i in range(3):
if args.plot_type == 'line':
axs[ax].plot(range(user), y_total[i], color=color[i], label=legend[i])
axs[ax].fill_between(range(user), y_low_total[i], y_high_total[i], color=color[i],alpha=0.3 )
if args.plot_type == 'bar':
if args.task_type == 'new_test_task':
widths = [-0.25, 0, 0.25]
yerror = [np.array(y_total[i])-np.array(y_low_total[i]), np.array(y_high_total[i])-np.array(y_total[i])]
axs[0].text(-0.19, 0.9, 'True Intention:', horizontalalignment='center', verticalalignment='center', transform=axs[0].transAxes, fontsize= 36)
axs[ax].bar(np.arange(user)+widths[i],y_total[i], width=0.2, yerr=yerror, color=color[i], label=legend[i])
axs[ax].tick_params(axis='x', which='both', pad=15, length=0)
plt.xticks(range(user), range(1,user+1))
axs[ax].set_ylabel('prob', fontsize= 36) # was 22
axs[ax].text(-0.19, 0.5, legend[ax], horizontalalignment='center', verticalalignment='center', transform=axs[ax].transAxes, fontsize= 36, color=color[ax])
plt.xlabel('user', fontsize= 40) # was 22
for k, x in enumerate(np.arange(user)+widths[i]):
y = y_total[i][k] + yerror[1][k]
axs[ax].annotate(f'{y_total[i][k]:.2f}', (x, y), textcoords='offset points', xytext=(-15, 3), fontsize=14)
if args.task_type == 'test_task':
if not args.plot_user_list:
axs[0].text(-0.19, 0.9, 'True Intention:', horizontalalignment='center', verticalalignment='center', transform=axs[0].transAxes, fontsize= 36)
axs[ax].errorbar(np.arange(user)+widths[i],y_total[i], yerr=[np.array(y_total[i])-np.array(y_low_total[i]), np.array(y_high_total[i])-np.array(y_total[i])],markerfacecolor=color[i], ecolor=color[i], markeredgecolor=color[i], label=legend[i],fmt='.k')
axs[ax].tick_params(axis='x', which='both', pad=15, length=0)
plt.xticks(range(user)[::5], range(1,user+1)[::5])
axs[ax].set_ylabel('prob', fontsize= 36)
axs[ax].text(-0.19, 0.5, legend[ax], horizontalalignment='center', verticalalignment='center', transform=axs[ax].transAxes, fontsize= 36, color=color[ax])
plt.xlabel('user', fontsize= 40)
else:
axs[0].text(-0.19, 0.9, 'True Intention:', horizontalalignment='center', verticalalignment='center', transform=axs[0].transAxes, fontsize= 36)
axs[ax].errorbar(np.arange(len(model_data_list))+widths[i],y_total[i], yerr=[np.array(y_total[i])-np.array(y_low_total[i]), np.array(y_high_total[i])-np.array(y_total[i])],markerfacecolor=color[i], ecolor=color[i], markeredgecolor=color[i], label=legend[i],fmt='.k')
axs[ax].tick_params(axis='x', which='both', pad=15, length=0)
plt.xticks(range(len(model_data_list)), user_table)
axs[ax].set_ylabel('prob', fontsize= 36)
#axs[ax].set_yticks(range(0.0,1.0, 0.25))
axs[ax].text(-0.19, 0.5, legend[ax], horizontalalignment='center', verticalalignment='center', transform=axs[ax].transAxes, fontsize= 36, color=color[ax])
plt.xlabel('user', fontsize= 40)
axs[ax].tick_params(axis='both', which='major', labelsize=30)
handles, labels = axs[0].get_legend_handles_labels()
plt.ylim([0, 1.08])
plt.tight_layout()
pathlib.Path("result/"+args.task_type+"/user"+str(user)+"/"+args.loss_type+"/figure/").mkdir(parents=True, exist_ok=True)
if args.task_type == 'test_task':
if not args.plot_user_list:
plt.savefig("result/"+args.task_type+"/user"+str(user)+"/"+args.loss_type+"/figure/"+"N"+ str(N)+"_"+args.model_type+"_rate_"+str(rate)+"_"+args.plot_type+"_test_set_1.png", bbox_inches='tight')
else:
plt.savefig("result/"+args.task_type+"/user"+str(user)+"/"+args.loss_type+"/figure/"+"N"+ str(N)+"_"+args.model_type+"_rate_"+str(rate)+"_"+args.plot_type+"_test_set_1_user_analysis.png", bbox_inches='tight')
if args.task_type == 'new_test_task':
plt.savefig("result/"+args.ask_type+"/user"+str(user)+"/"+args.loss_type+"/figure/"+"N"+ str(N)+"_"+args.model_type+"_rate_"+str(rate)+"_"+args.plot_type+"_test_set_2.png", bbox_inches='tight')
plt.show()
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='')
parser.add_argument('--loss_type', type=str, default='ce')
parser.add_argument('--model_type', type=str, default="lstmlast" )
parser.add_argument('--plot_type', type=str, default='bar') # bar or line
parser.add_argument('--task_type', type=str, default='test_task')
parser.add_argument('--plot_user_list', action='store_true') # plot user_table or not
args = parser.parse_args()
main(args)

5
watch_and_help/stan/plot_user_length.sh Normal file
View file

@ -0,0 +1,5 @@
python3 plot_user_length.py \
--loss_type ce \
--model_type lstmlast \
--plot_type bar \
--task_type test_task

88
watch_and_help/stan/plot_user_length_10_steps.py Normal file
View file

@ -0,0 +1,88 @@
import numpy as np
from numpy import genfromtxt
import matplotlib.pyplot as plt
import argparse
import pathlib
def main():
parser = argparse.ArgumentParser(description='')
parser.add_argument('--loss_type', type=str, default='ce')
parser.add_argument('--model_type', type=str, default="lstmlast" )
parser.add_argument('--task_type', type=str, default='test_task')
args = parser.parse_args()
if args.task_type == 'new_test_task':
user = 9
N = 1
if args.task_type == 'test_task':
user = 92
N = 1
#rate = range(0,101,10)
rate_user_data_list = []
for r in range(0,101,10): # rate = range(0,101,10)
# read data
print(r)
model_data_list = []
for i in range(user):
path = "result/"+args.task_type+"/user"+str(user)+"/"+args.loss_type+"/N"+ str(N) + "/" + args.model_type + "_N" + str(N) + "_result_" + str(r) + "_user" + str(i) +".csv"
data = genfromtxt(path, delimiter=',', skip_header =1)
data = data[[1,2,3,5,6,7,9,10,11],:][:,[2,4,6,7]]
model_data_list.append(data)
#print(type(data))
model_data_list_total = np.stack(model_data_list)
mean_user_data = np.mean(model_data_list_total,axis=0)
rate_user_data_list.append(mean_user_data)
color = ['royalblue', 'lightgreen', 'tomato']
legend = ['put fridge', 'put\n dishwasher', 'read book']
fig, axs = plt.subplots(3, sharex=True, sharey=True)
fig.set_figheight(10) # all sample rate: 10; 3 row: 8
fig.set_figwidth(20)
axs[0].text(-0.145, 0.9, 'True Intention:', horizontalalignment='center', verticalalignment='center', transform=axs[0].transAxes, fontsize= 25) # all: -0.3,0.5 3rows: -0.5,0.5
for ax in range(3):
y_total = []
y_low_total = []
y_high_total = []
for j in range(3):
y= []
y_low = []
y_high = []
for i in range(len(rate_user_data_list)):
y.append(rate_user_data_list[i][j+ax*3][0])
y_low.append(rate_user_data_list[i][j+ax*3][2])
y_high.append(rate_user_data_list[i][j+ax*3][3])
y_total.append(y)
y_low_total.append(y_low)
y_high_total.append(y_high)
print()
print("user mean of mean prob: ", np.mean(y))
print("user mean of sd prob: ", np.std(y))
for i in range(3):
axs[ax].plot(range(0,101,10), y_total[i], color=color[i], label=legend[i])
axs[ax].fill_between(range(0,101,10), y_low_total[i], y_high_total[i], color=color[i],alpha=0.3 )
axs[ax].set_xticks(range(0,101,10))
axs[ax].set_ylabel('probability', fontsize=22)
axs[ax].text(-0.145, 0.5, legend[ax], horizontalalignment='center', verticalalignment='center', transform=axs[ax].transAxes, fontsize= 25, color=color[ax])
axs[ax].tick_params(axis='both', which='major', labelsize=18)
plt.xlabel('Percentage of observed actions in one action sequence', fontsize= 22)
handles, labels = axs[0].get_legend_handles_labels()
plt.xlim([0, 101])
plt.ylim([0, 1])
pathlib.Path("result/"+args.task_type+"/user"+str(user)+"/"+args.loss_type+"/figure/").mkdir(parents=True, exist_ok=True)
if args.task_type == 'test_task':
plt.savefig("result/"+args.task_type+"/user"+str(user)+ "/"+args.loss_type+"/figure/N"+ str(N) + "_"+args.model_type+"_rate_full_test_set_1.png", bbox_inches='tight')
if args.task_type == 'new_test_task':
plt.savefig("result/"+args.task_type+"/user"+str(user)+ "/"+args.loss_type+"/figure/N"+ str(N) + "_"+args.model_type+"_rate_full_test_set_2.png", bbox_inches='tight')
plt.show()
if __name__ == '__main__':
main()

4
watch_and_help/stan/plot_user_length_10_steps.sh Normal file
View file

@ -0,0 +1,4 @@
python3 plot_user_length_10_steps.py \
--loss_type ce \
--model_type lstmlast \
--task_type test_task

64
watch_and_help/stan/sampler_user.py Normal file
View file

@ -0,0 +1,64 @@
import numpy as np
from numpy import genfromtxt
import csv
import pandas
import argparse
def sample_predciton(path, rate):
data = pandas.read_csv(path).values
task_list = [0, 1, 2]
start = 0
stop = 0
num_unique = np.unique(data[:,1])
#print('unique number', num_unique)
samples = []
for j in task_list:
for i in num_unique:
inx = np.where((data[:,1] == i) & (data[:,-2] == j))
samples.append(data[inx])
for i in range(len(samples)):
n = int(len(samples[i])*(100-rate)/100)
samples[i] = samples[i][:-n]
return np.vstack(samples)
def main():
parser = argparse.ArgumentParser(description='')
parser.add_argument('--LOSS', type=str, default='ce')
parser.add_argument('--MODEL_TYPE', type=str, default="lstmlast_cross_entropy_bs_32_iter_2000_train_task_prob" )
parser.add_argument('--EPOCHS', type=int, default=50)
parser.add_argument('--TASK', type=str, default='test_task')
args = parser.parse_args()
task = ['put_fridge', 'put_dishwasher', 'read_book']
sets = [args.TASK]
rate = [10, 20, 30, 40, 50, 60, 70, 80, 90]
for i in task:
for j in rate:
for k in sets:
if k == 'test_task':
user_num = 92
if k == 'new_test_task':
user_num = 9
for l in range(user_num):
pred_path = "prediction/" + k + "/" + "user" + str(user_num) + "/ce/" + i + "/" + "loss_weight_" + args.MODEL_TYPE + "_prediction_" + i + "_user" + str(l) + ".csv"
save_path = "prediction/" + k + "/" + "user" + str(user_num) + "/ce/" + i + "/" + "loss_weight_" + args.MODEL_TYPE + "_prediction_" + i + "_user" + str(l) + "_rate_" + str(j) + ".csv"
data = sample_predciton(pred_path, j)
head = []
for r in range(79):
head.append('act'+str(r+1))
head.append('task_name')
head.append('gt')
head.insert(0,'action_id')
pandas.DataFrame(data[:,1:]).to_csv(save_path, header=head)
if __name__ == '__main__':
main()

5
watch_and_help/stan/sampler_user.sh Normal file
View file

@ -0,0 +1,5 @@
python3 sampler_user.py \
--TASK test_task \
--LOSS ce \
--MODEL_TYPE lstmlast \
--EPOCHS 50

76
watch_and_help/stan/save_act_series.R Normal file
View file

@ -0,0 +1,76 @@
library(tidyverse)
library(cmdstanr)
library(dplyr)
strategies <- c("put_fridge", "put_dishwasher", "read_book")
model_type <- "lstmlast_cross_entropy_bs_32_iter_2000_train_task_prob"
rate <- "_0"
task_type <- "new_test_task" # new_test_task test_task
loss_type <- "ce"
set.seed(9746234)
if (task_type=="test_task"){
user_num <- 92
user <-c(0:(user_num-1))
N <- 1
}
if (task_type=="new_test_task"){
user_num <- 9
user <-c(0:(user_num-1))
N <- 1
}
total_user_act1 <- vector("list", length(user_num))
total_user_act2 <- vector("list", length(user_num))
sel <- vector("list", length(strategies))
act_series <- vector("list", user_num)
for (u in seq_along(user)){
print('user')
print(u)
dat <- vector("list", length(strategies))
for (i in seq_along(strategies)) {
if (rate=="_0"){
dat[[i]] <- read.csv(paste0("prediction/", task_type, "/user", user_num, "/", loss_type, "/", strategies[[i]], "/loss_weight_", model_type, "_prediction_", strategies[[i]], "_user", user[[u]], "_rate_", "90", ".csv"))
} else if (rate=="_100"){
dat[[i]] <- read.csv(paste0("prediction/", task_type, "/user", user_num, "/", loss_type, "/", strategies[[i]], "/loss_weight_", model_type, "_prediction_", strategies[[i]], "_user", user[[u]], ".csv"))
} else{
dat[[i]] <- read.csv(paste0("prediction/", task_type, "/user", user_num, "/", loss_type, "/", strategies[[i]], "/loss_weight_", model_type, "_prediction_", strategies[[i]], "_user", user[[u]], "_rate", rate, ".csv"))
}
dat[[i]]$assumed_strategy <- strategies[[i]]
dat[[i]]$index <- dat[[i]]$action_id # sample based on intention
dat[[i]]$id <- dat[[i]][,1] # sample based on intention
}
N <- 1
# select all action series and infer every one
sel[[1]]<-dat[[1]] %>%
group_by(task_name) %>%
filter(task_name==1)
sel[[1]] <- data.frame(sel[[1]])
unique_act_id_t1 <- unique(sel[[1]]$action_id)
write.csv(unique_act_id_t1, paste0("result/", task_type, "/user", user_num, "/", loss_type, "/act", "/", "action_series_", "user_",u, "_put_dishwasher", ".csv"))
total_user_act1[[u]] <- unique_act_id_t1
sel[[1]]<-dat[[1]] %>%
group_by(task_name) %>%
filter(task_name==2)
sel[[1]] <- data.frame(sel[[1]])
unique_act_id_t1 <- unique(sel[[1]]$action_id)
write.csv(unique_act_id_t1, paste0("result/", task_type, "/user", user_num, "/", loss_type, "/act", "/", "action_series_", "user_",u, "_read_book", ".csv"))
total_user_act2[[u]] <- unique_act_id_t1
}
write.csv(total_user_act1, paste0("result/", task_type, "/user", user_num, "/", loss_type, "/act", "/", "action_series_", "_put_dishwasher_total", ".csv"))
write.csv(total_user_act2, paste0("result/", task_type, "/user", user_num, "/", loss_type, "/act", "/", "action_series_", "read_book_total", ".csv"))

87
watch_and_help/stan/split_user.py Normal file
View file

@ -0,0 +1,87 @@
import numpy as np
import pathlib
import argparse
np.random.seed(seed=100)
def sample_user(data, num_users, split_inx):
np.random.seed(seed=100)
num_unique3 = np.unique(data[:,1])
num_unique2 = num_unique3[0:split_inx[1]]
num_unique = num_unique3[0:split_inx[0]]
user_list1 = [np.random.choice(num_unique, int(len(num_unique)/num_users), replace=False) for i in range(num_users)]
user_list2 = [np.random.choice(num_unique2, int(len(num_unique2)/num_users), replace=False) for i in range(num_users)]
user_list3 = [np.random.choice(num_unique3, int(len(num_unique3)/num_users), replace=False) for i in range(num_users)]
user_data = []
for i in range(num_users): # len(user_list)
user_idx1 = [int(item) for item in user_list1[i]]
user_idx2 = [int(item) for item in user_list2[i]]
user_idx3 = [int(item) for item in user_list3[i]]
data_list = []
for j in range(len(user_idx1)):
inx = np.where((data[:,1] == user_idx1[j]) & (data[:,-2]==0))
data_list.append(data[inx])
for j in range(len(user_idx2)):
inx = np.where((data[:,1] == user_idx2[j]) & (data[:,-2]==1))
data_list.append(data[inx])
for j in range(len(user_idx3)):
inx = np.where((data[:,1] == user_idx3[j]) & (data[:,-2]==2))
data_list.append(data[inx])
user_data.append(np.vstack(data_list))
return user_data
def main():
parser = argparse.ArgumentParser(description='')
parser.add_argument('--LOSS', type=str, default='ce')
parser.add_argument('--MODEL_TYPE', type=str, default="lstmlast_cross_entropy_bs_32_iter_2000_train_task_prob" )
parser.add_argument('--EPOCHS', type=int, default=50)
parser.add_argument('--TASK', type=str, default='test_task')
args = parser.parse_args()
pref = ['put_fridge', 'put_dishwasher', 'read_book']
if args.TASK == 'new_test_task':
NUM_USER = 9 # 9 for 1 user 1 action
SPLIT_INX = [NUM_USER, 45]
if args.TASK == 'test_task':
NUM_USER = 92
SPLIT_INX = [NUM_USER, 229]
head = []
for j in range(79):
head.append('act'+str(j+1))
head.append('task_name')
head.append('gt')
head.insert(0,'action_id')
head.insert(0,'')
for i in pref:
path = "prediction/"+args.TASK+"/" + args.MODEL_TYPE + "/model_" + i + "_strategy_put_fridge" +".csv"
data = np.genfromtxt(path, skip_header=1, delimiter=',')
data_task_name = np.genfromtxt(path, skip_header=1, delimiter=',', usecols=-2, dtype=None)
data_task_name[data_task_name==b'put_fridge'] = 0
data_task_name[data_task_name==b'put_dishwasher'] = 1
data_task_name[data_task_name==b'read_book'] = 2
data[:,-2] = data_task_name.astype(np.float)
print("data length: ", len(data))
users_data = sample_user(data, NUM_USER, SPLIT_INX)
length = 0
pathlib.Path("prediction/"+args.TASK+"/user" + str(NUM_USER) + "/" + args.LOSS + "/" + i).mkdir(parents=True, exist_ok=True)
for j in range(len(users_data)):
save_path = "prediction/"+args.TASK+"/user" + str(NUM_USER) + "/" + args.LOSS + "/" + i +"/loss_weight_"+ args.MODEL_TYPE + "_prediction_"+ i + "_user"+str(j)+".csv"
length = length + len(users_data[j])
np.savetxt(save_path, users_data[j], delimiter=',', header=','.join(head))
print("user data length: ", length)
if __name__ == '__main__':
main()

5
watch_and_help/stan/split_user.sh Normal file
View file

@ -0,0 +1,5 @@
python3 split_user.py \
--TASK test_task \
--LOSS ce \
--MODEL_TYPE lstmlast \
--EPOCHS 50

BIN
watch_and_help/stan/strategy_inference_model Executable file
View file

Binary file not shown.

26
watch_and_help/stan/strategy_inference_model.stan Normal file
View file

@ -0,0 +1,26 @@
data {
int<lower=1> I; // number of question options (22)
int<lower=0> N; // number of questions being asked by the user
int<lower=1> K; // number of strategies
// observed "true" questions of the user
int q[N];
// array of predicted probabilities of questions given strategies
// coming from the forward neural network
matrix[I, K] P_q_S[N];
}
parameters {
// probabiliy vector of the strategies being applied by the user
// to be inferred by the model here
simplex[K] P_S;
}
model {
for (n in 1:N) {
// marginal probability vector of the questions being asked
vector[I] theta = P_q_S[n] * P_S;
// categorical likelihood
target += categorical_lpmf(q[n] | theta);
}
// priors
target += dirichlet_lpdf(P_S | rep_vector(1.0, K));
}

190
watch_and_help/stan/strategy_inference_test.R Normal file
View file

@ -0,0 +1,190 @@
library(tidyverse)
library(cmdstanr)
library(dplyr)
# index order of the strategies assumed throughout
strategies <- c("put_fridge", "put_dishwasher", "read_book")
model_type <- "lstmlast"
rates <- c("_0", "_10", "_20", "_30", "_40", "_50", "_60", "_70", "_80", "_90", "_100")
task_type <- "test_task" # new_test_task test_task
loss_type <- "ce"
set.seed(9746234)
if (task_type=="test_task"){
user_num <- 92
user <-c(38:(user_num-1))
N <- 1
}
if (task_type=="new_test_task"){
user_num <- 9
user <-c(0:(user_num-1))
N <- 1
}
# read data from csv
sel <- vector("list", length(strategies))
act_series <- vector("list", user_num)
for (u in seq_along(user)){
for (rate in rates) {
dat <- vector("list", length(strategies))
for (i in seq_along(strategies)) {
if (rate=="_0"){
dat[[i]] <- read.csv(paste0("prediction/", task_type, "/user", user_num, "/", loss_type, "/", strategies[[i]], "/loss_weight_", model_type, "_prediction_", strategies[[i]], "_user", user[[u]], "_rate_", "10", ".csv")) # _60
} else if (rate=="_100"){
dat[[i]] <- read.csv(paste0("prediction/", task_type, "/user", user_num, "/", loss_type, "/", strategies[[i]], "/loss_weight_", model_type, "_prediction_", strategies[[i]], "_user", user[[u]], ".csv")) # _60
} else{
dat[[i]] <- read.csv(paste0("prediction/", task_type, "/user", user_num, "/", loss_type, "/", strategies[[i]], "/loss_weight_", model_type, "_prediction_", strategies[[i]], "_user", user[[u]], "_rate", rate, ".csv")) # _60
}
# strategy assumed for prediction
dat[[i]]$assumed_strategy <- strategies[[i]]
dat[[i]]$index <- dat[[i]]$action_id # sample based on intention
dat[[i]]$id <- dat[[i]][,1] # sample based on intention
}
# reset N after inference
if (task_type=="test_task"){
N <- 1
}
if (task_type=="new_test_task"){
N <- 1
}
# select one action series from one intention
if (rate == "_0"){
sel[[1]]<-dat[[1]] %>%
group_by(task_name) %>%
sample_n(N)
sel[[1]] <- data.frame(sel[[1]])
act_series[[u]] <- sel[[1]]$action_id
#print(typeof(sel[[1]]))
#print(typeof(dat[[1]]))
#print(sel[[1]]$action_id[2])
}
print(c('unique action id', sel[[1]]$action_id))
# filter data from the selected action series, N series per intention
dat[[1]]<-subset(dat[[1]], dat[[1]]$action_id == sel[[1]]$action_id[1] | dat[[1]]$action_id == sel[[1]]$action_id[2] | dat[[1]]$action_id == sel[[1]]$action_id[3])
dat[[2]]<-subset(dat[[2]], dat[[2]]$action_id == sel[[1]]$action_id[1] | dat[[2]]$action_id == sel[[1]]$action_id[2] | dat[[2]]$action_id == sel[[1]]$action_id[3])
dat[[3]]<-subset(dat[[3]], dat[[3]]$action_id == sel[[1]]$action_id[1] | dat[[3]]$action_id == sel[[1]]$action_id[2] | dat[[3]]$action_id == sel[[1]]$action_id[3])
row.names(dat) <- NULL
print(c('task name 1', dat[[1]]$task_name))
print(c('task name 2', dat[[2]]$task_name))
print(c('task name 3', dat[[3]]$task_name))
print(c('action id 1', dat[[1]]$action_id))
print(c('action id 2', dat[[2]]$action_id))
print(c('action id 3', dat[[3]]$action_id))
# create save path
dir.create(file.path(paste0("result/", task_type, "/user", user_num, "/", loss_type, "/N", N)), showWarnings = FALSE, recursive = TRUE)
dir.create(file.path("temp"), showWarnings = FALSE)
save_path <- paste0("result/", task_type, "/user", user_num, "/", loss_type, "/N", N, "/", model_type, "_N", N, "_", "result", rate,"_user", user[[u]], ".csv")
if(task_type=="test_task"){
dat <- do.call(rbind, dat) %>%
mutate(index = as.numeric(as.factor(id))) %>%
rename(true_strategy = task_name) %>%
mutate(
true_strategy = factor(
#true_strategy, levels = 0:3,
true_strategy, levels = 0:2,
labels = strategies
),
q_type = case_when(
gt %in% c(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 19, 20, 22, 23, 29, 30, 31, 32, 33, 34, 35, 37, 38, 39, 40, 42, 43, 44, 58, 59, 64, 65, 68, 69, 70, 71, 72, 73, 74) ~ "put_fridge",
gt %in% c(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 25, 29,30, 31, 32, 33, 34, 37, 38, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57) ~ "put_dishwasher",
gt %in% c(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45) ~ "read_book",
)
)
}
if(task_type=="new_test_task"){
dat <- do.call(rbind, dat) %>%
mutate(index = as.numeric(as.factor(id))) %>%
rename(true_strategy = task_name) %>%
mutate(
true_strategy = factor(
true_strategy, levels = 0:2,
labels = strategies
),
q_type = case_when(
# new_test_set
gt %in% c(1, 5, 6, 7, 8, 9, 10, 12, 13, 14, 15, 16, 19, 20, 22, 23, 25, 29, 30, 31, 32, 33, 34, 35, 40, 42, 43, 44, 46, 47, 52, 53, 55, 56, 58, 59, 60, 64, 65, 68, 69, 70, 71, 72, 73, 74, 75, 77, 78) ~ "put_fridge",
gt %in% c(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 18, 19, 20, 21, 22, 23, 24, 25, 29, 30, 31, 32, 33, 34, 35, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74) ~ "put_dishwasher",
gt %in% c(1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 60, 75, 76, 77, 78) ~ "read_book",
)
)
}
#print(nrow(dat))
#print(dat)
dat_obs <- dat %>% filter(assumed_strategy == strategies[[i]])
N <- nrow(dat_obs)
print(c("N: ", N))
q <- dat_obs$gt
true_strategy <- dat_obs$true_strategy
K <- length(unique(dat$assumed_strategy))
I <- 79
P_q_S <- array(dim = c(N, I, K))
for (n in 1:N) {
P_q_S[n, , ] <- dat %>%
filter(index == n) %>%
select(matches("^act[[:digit:]]+$")) %>%
as.matrix() %>%
t()
for (k in 1:K) {
# normalize probabilities
P_q_S[n, , k] <- P_q_S[n, , k] / sum(P_q_S[n, , k])
}
}
mod <- cmdstan_model(paste0(getwd(),"/strategy_inference_model.stan"))
if (rate=="_0"){
sub <- integer(0)
} else {
sub <- which(true_strategy == "put_fridge")
}
sdata <- list(N = length(sub), K = K, I = I, q = q[sub], P_q_S = P_q_S[sub, , ])
fit_put_fridge <- mod$sample(data = sdata, refresh=0, output_dir=paste0(getwd(),"/temp"))
print(fit_put_fridge$summary(NULL, c("mean","sd")))
if (rate=="_0"){
sub <- integer(0)
} else {
sub <- which(true_strategy == "put_dishwasher")
}
sdata <- list(N = length(sub), K = K, I = I, q = q[sub], P_q_S = P_q_S[sub, , ])
fit_put_dishwasher <- mod$sample(data = sdata, refresh=0, output_dir=paste0(getwd(),"/temp"))
print(fit_put_dishwasher$summary(NULL, c("mean","sd")))
# read_book strategy (should favor index 3)
if (rate=="_0"){
sub <- integer(0)
} else {
sub <- which(true_strategy == "read_book")
}
sdata <- list(N = length(sub), K = K, I = I, q = q[sub], P_q_S = P_q_S[sub, , ])
fit_read_book <- mod$sample(data = sdata, refresh=0, output_dir=paste0(getwd(),"/temp"))
print(fit_read_book$summary(NULL, c("mean","sd")))
# save csv
df <-rbind(fit_put_fridge$summary(), fit_put_dishwasher$summary(), fit_read_book$summary())
write.csv(df,file=save_path,quote=FALSE)
}
}