This part of the code is provided for the project.
library(ggplot2)
library(lubridate)
library(caTools)
library(zoo)
library(forecast)
require(jsonlite)
require(httr)
require(data.table)
library(rvest)
get_token <- function(username, password, url_site){
post_body = list(username=username,password=password)
post_url_string = paste0(url_site,'/token/')
result = POST(post_url_string, body = post_body)
token=NULL
# error handling (wrong credentials)
if(result$status_code==400){
print('Check your credentials')
return(0)
}
else if (result$status_code==201){
output = content(result)
token = output$key
}
return(token)
}
#get_data <- function(start_date='2020-03-20', token, url_site){ #hocanK n
get_data <- function(start_date='2020-03-20', token, url_site,username, password){
post_body = list(start_date=start_date,username=username,password=password)
post_url_string = paste0(url_site,'/dataset/')
header = add_headers(c(Authorization=paste('Token',token,sep=' ')))
result = GET(post_url_string, header, body = post_body)
output = content(result)
data = data.table::rbindlist(output)
data[,event_date:=as.Date(event_date)]
data = data[order(product_content_id,event_date)]
return(data)
}
send_submission <- function(predictions, token, url_site, submit_now=F){
format_check=check_format(predictions)
if(!format_check){
return(FALSE)
}
post_string="list("
for(i in 1:nrow(predictions)){
post_string=sprintf("%s'%s'=%s",post_string,predictions$product_content_id[i],predictions$forecast[i])
if(i<nrow(predictions)){
post_string=sprintf("%s,",post_string)
} else {
post_string=sprintf("%s)",post_string)
}
}
submission = eval(parse(text=post_string))
json_body = jsonlite::toJSON(submission, auto_unbox = TRUE)
submission=list(submission=json_body)
print(submission)
# {"31515569":2.4,"32939029":2.4,"4066298":2.4,"6676673":2.4,"7061886":2.4,"85004":2.4}
if(!submit_now){
print("You did not submit.")
return(FALSE)
}
header = add_headers(c(Authorization=paste('Token',token,sep=' ')))
post_url_string = paste0(url_site,'/submission/')
result = POST(post_url_string, header, body=submission)
if (result$status_code==201){
print("Successfully submitted. Below you can see the details of your submission")
} else {
print("Could not submit. Please check the error message below, contact the assistant if needed.")
}
print(content(result))
}
check_format <- function(predictions){
if(is.data.frame(predictions) | is.data.frame(predictions)){
if(all(c('product_content_id','forecast') %in% names(predictions))){
if(is.numeric(predictions$forecast)){
print("Format OK")
return(TRUE)
} else {
print("forecast information is not numeric")
return(FALSE)
}
} else {
print("Wrong column names. Please provide 'product_content_id' and 'forecast' columns")
return(FALSE)
}
} else {
print("Wrong format. Please provide data.frame or data.table object")
return(FALSE)
}
}
# this part is main code
subm_url = 'http://157.230.99.81'
# change your credentials accordingly
u_name = 'Group1'
p_word = 'JcIcMxP7CWUusF84'
submit_now = FALSE
token = get_token(username=u_name, password=p_word, url=subm_url)
data = get_data(token=token,url=subm_url,username=u_name, password=p_word)
predictions=unique(data[,list(product_content_id)])
predictions[,forecast:=2.3]
send_submission(predictions, token, url=subm_url, submit_now=F)After reading in the data; we transformed -1 values to NA, because -1 values represented the sales=0 data for related products. Then we performed imputation with zoo package, to fill in these values with the closes neighboring value.
#Converting -1 values to NA for 5 variables that contain NaN values in original set
data$price[data$price==-1] <- NA
data$category_visits[data$category_visits==-1] <- NA
data$ty_visits[data$ty_visits==-1] <- NA
data$category_sold[data$category_sold==-1] <- NA
data$category_brand_sold[data$category_brand_sold==-1] <- NA
#Changing variable types to numeric and factor
initialdata=data
initialdata$product_content_id <- as.factor(initialdata$product_content_id)
initialdata$sold_count <- as.numeric(initialdata$sold_count)
initialdata$visit_count <- as.numeric(initialdata$visit_count)
initialdata$favored_count <- as.numeric(initialdata$favored_count)
initialdata$basket_count <- as.numeric(initialdata$basket_count)
initialdata$category_sold <- as.numeric(initialdata$category_sold)
initialdata$category_brand_sold <- as.numeric(initialdata$category_brand_sold)
initialdata$category_visits <- as.numeric(initialdata$category_visits)
initialdata$ty_visits <- as.numeric(initialdata$ty_visits)
#Filling in NA prices - na.locf function in zoo package
initialdata <- na.locf(initialdata, na.rm=FALSE, option = "locf")
initialdata <- na.locf(initialdata, na.rm=FALSE, option = "locf", fromLast = TRUE)
#Plots of price change over time for each product
plot(initialdata[product_content_id==85004]$event_date,initialdata[product_content_id==85004]$price)
plot(initialdata[product_content_id==4066298]$event_date,initialdata[product_content_id==4066298]$price)
plot(initialdata[product_content_id==6676673]$event_date,initialdata[product_content_id==6676673]$price)
plot(initialdata[product_content_id==7061886]$event_date,initialdata[product_content_id==7061886]$price)
plot(initialdata[product_content_id==31515569]$event_date,initialdata[product_content_id==31515569]$price)
plot(initialdata[product_content_id==32939029]$event_date,initialdata[product_content_id==32939029]$price)
plot(initialdata[product_content_id==5926527]$event_date,initialdata[product_content_id==5926527]$price)
plot(initialdata[product_content_id==3904356]$event_date,initialdata[product_content_id==3904356]$price)
#Product List is here for easily matching product_content_id with products
product_list=predictions[,1]
top_hierarchy=data.table(c("Giyim","SB8permarket","Giyim","SB8permarket","Giyim","Elektronik","Elektronik","Kozmetik"))
bottom_hierarchy=data.table(c("Tayt","o,arj Edebilir DiK FK rCasK ","Mont","Islak Mendil & Havlu","Bikini b 9stB8","Telefon Bluetooth KulaklK k","SB8pB8rge","YB8z Temizleyici"))
brand=data.table(c("TRENDYOLMb :LLA","Oral-B","Koton","Sleepy","TRENDYOLMb :LLA","Xiaomi","Fakir","La Roche Posay"))
product_list=cbind(product_list,top_hierarchy,bottom_hierarchy,brand)
colnames(product_list)=c("product_content_id","top_hierarchy","bottom_hierarchy","brand")
product_listSince price is a major factor in purchasing decisions and there are daily price changes/discounts in Trendyol; we scraped current up-to-date price from their websites to adjust our predictions accordingly.
url31515569 <- 'https://www.trendyol.com/trendyolmilla/siyah-yuksek-bel-toparlayici-orme-tayt-twoaw20ta0087-p-31515569'
url32939029 <- 'https://www.trendyol.com/oral-b/vitality-100-cross-action-white-elektrikli-dis-fircasi-p-32939029'
url3904356 <- 'https://www.trendyol.com/koton/erkek-lacivert-fermuar-detayli-uzun-kollu-cep-detayli-kapitone-mont-9kam21129nw-p-3904356'
url4066298 <- 'https://www.trendyol.com/sleepy/natural-yenidogan-islak-pamuklu-havlu-40-li-x-12-480-yaprak-p-4066298'
url5926527 <- 'https://www.trendyol.com/trendyolmilla/siyah-kapli-cicek-desenli-bikini-ustu-tbess19wv0005-p-5926527'
url6676673 <- 'https://www.trendyol.com/xiaomi/redmi-airdots-tws-bluetooth-basic-5-0-kulaklik-p-6676673'
url7061886 <- 'https://www.trendyol.com/fakir/lucky-dikey-elektrikli-supurge-kum-beji-p-7061886'
url85004 <- 'https://www.trendyol.com/la-roche-posay/effaclar-yuz-temizleme-jeli-yagli-akneye-egilim-gosteren-ciltler-siyah-nokta-karsiti-200ml-p-85004'
page85004 <- read_html(url85004)
page4066298 <- read_html(url4066298)
page32939029 <- read_html(url32939029)
page31515569 <- read_html(url31515569)
page3904356 <- read_html(url3904356)
page5926527 <- read_html(url5926527)
page6676673 <- read_html(url6676673)
page7061886 <- read_html(url7061886)
today=predictions
colnames(today)=c("product_content_id","price")
price_31515569 <- page31515569 %>%
html_nodes(".prc-slg") %>%
html_text()
new <- unlist(strsplit(price_31515569, "\\ "))
new <- gsub(",", ".", new[1])
new=as.numeric(new)
today[1,2]=new
price_32939029 <- page32939029 %>%
html_nodes(".prc-slg") %>%
html_text()
new <- unlist(strsplit(price_32939029, "\\ "))
new <- gsub(",", ".", new[1])
new=as.numeric(new)
today[2,2]=new
price_3904356 <- page3904356 %>%
html_nodes(".prc-slg") %>%
html_text()
new <- unlist(strsplit(price_3904356, "\\ "))
new <- gsub(",", ".", new[1])
new=as.numeric(new)
today[3,2]=0 #when this product is soldout,we get an error.
today[3,2]=new
price_4066298 <- page4066298 %>%
html_nodes(".prc-slg") %>%
html_text()
new <- unlist(strsplit(price_4066298, "\\ "))
new <- gsub(",", ".", new[1])
new=as.numeric(new)
today[4,2]=new
price_5926527 <- page5926527 %>%
html_nodes(".prc-slg") %>%
html_text()
new <- unlist(strsplit(price_5926527, "\\ "))
new <- gsub(",", ".", new[1])
new=as.numeric(new)
today[5,2]=new
price_6676673 <- page6676673 %>%
html_nodes(".prc-slg") %>%
html_text()
new <- unlist(strsplit(price_6676673, "\\ "))
new <- gsub(",", ".", new[1])
new=as.numeric(new)
today[6,2]=new
price_7061886 <- page7061886 %>%
html_nodes(".prc-slg") %>%
html_text()
new <- unlist(strsplit(price_7061886, "\\ "))
new <- gsub(",", ".", new[1])
new=as.numeric(new)
today[7,2]=new
price_85004 <- page85004 %>%
html_nodes(".prc-slg") %>%
html_text()
new <- unlist(strsplit(price_85004, "\\ "))
new <- gsub(",", ".", new[1])
new=as.numeric(new)
today[8,2]=new
todayIn the beginning, we developed several naive approaches for predictions while we develop our final model. These helped us to get even more familiar with the data and observe trends closer.
#Looking at the last X days averages
for(i in 1:8){
code=as.numeric(predictions[i,1])
predictions[i,2]=predict365(data,code)
}
our_predictions=predictions
our_predictions[,last365days:=predictions[,2]]
predictions=predictions[,1:2]
for(i in 1:8){
code=as.numeric(predictions[i,1])
predictions[i,2]=predict_lastXdays(data,code,7)
}
our_predictions[,last7days:=predictions[,2]]
for(i in 1:8){
code=as.numeric(predictions[i,1])
predictions[i,2]=predict_avglastXdays(data,code,7)
}
our_predictions[,last7days_avg:=predictions[,2]]
for(i in 1:8){
code=as.numeric(predictions[i,1])
predictions[i,2]=predict_avglastXdays(data,code,30)
}
our_predictions[,last30days_avg:=predictions[,2]]
# Prediction from last year
predict365=function(data,code){
product=data[data$product_content_id==code]
prediction_date=ymd(Sys.Date()) - years(1)
prediction=product[product$event_date==prediction_date,]
#print(prediction$sold_count)
return(prediction$sold_count)
}
# Prediction from last X days
predict_lastXdays=function(data,code,n_days){
product=data[data$product_content_id==code]
prediction_date=ymd(Sys.Date()) - days(n_days)
prediction=product[product$event_date==prediction_date,]
return(prediction$sold_count)
}
# Prediction with average sales of last X days
predict_avglastXdays=function(data,code,n_days){
sum=0
product=data[data$product_content_id==code]
for(i in 1:n_days){
prediction_date=ymd(Sys.Date()) - days(i)
prediction=product[product$event_date==prediction_date,]
sum=sum+prediction$sold_count
}
average_sold_count=sum/n_days
return(average_sold_count)
}
#prediction from last 2 weeks mean with sold_count
for(p_id in 1:8){
code=as.numeric(predictions[p_id,1])
p1=data[data$product_content_id==code & data$event_date>=ymd(Sys.Date()) - days(14)]
mean1=mean(p1$sold_count[1:7])
mean2=mean(p1$sold_count[7:14])
change=(mean1-mean2)/mean1
predicted_sold_count=p1$sold_count[p1$event_date==ymd(Sys.Date()) - days(6)]*(1+change)
if(is.nan(predicted_sold_count)){
predicted_sold_count=0
}
predictions[p_id,2]=predicted_sold_count
}
our_predictions[,last14days_meanWithSoldCount:=predictions[,2]]
#prediction from last 2 weeks mean with basket_count
for(p_id in 1:8){
code=as.numeric(predictions[p_id,1])
p1=data[data$product_content_id==code & data$event_date>=ymd(Sys.Date()) - days(14)]
mean1=mean(p1$basket_count[1:7])
mean2=mean(p1$basket_count[7:14])
change=(mean1-mean2)/mean1
predicted_sold_count=p1$sold_count[p1$event_date==ymd(Sys.Date()) - days(6)]*(1+change)
if(is.nan(predicted_sold_count)){
predicted_sold_count=0
}
predictions[p_id,2]=predicted_sold_count
}
our_predictions[,last14days_meanWithBasketCount:=predictions[,2]]
our_predictions
predictions
#Change rates of last 2 weeks data
p_id=6
code=as.numeric(predictions[p_id,1])
p1=data[data$product_content_id==code & data$event_date>=ymd(Sys.Date()) - days(14)]
mean1=mean(p1$sold_count[1:7])
mean2=mean(p1$sold_count[7:14])
change=(mean1-mean2)/mean1
predicted_sold_count=p1$sold_count[p1$event_date==ymd(Sys.Date()) - days(6)]*(1+change)
if(is.nan(predicted_sold_count)){
predicted_sold_count=0
}
predictions[p_id,2]=predicted_sold_count
product_list$bottom_hierarchy[p_id]
mean2
mean1
change
p1$sold_count[p1$event_date==ymd(Sys.Date()) - days(6)]
predicted_sold_count
p1
plot(p1$sold_count,p1$sold_count)
#Using max and min predicted values for predictions
for(i in 1:8){
code=as.numeric(predictions[i,1])
predictions[i,2]=predict_lastXdays(data,code,7)
}
lag7=predictions
max_prediction=predictions[,2]
min_prediction=predictions[,2]
for(i in 1:8){
max_prediction[i]=max(pr_sold[i,2],pr_basket[i,2],lag7[i,2])
min_prediction[i]=abs(min(pr_sold[i,2],pr_basket[i,2],lag7[i,2]))
}
cbind(pr_sold,pr_basket,lag7,max_prediction,min_prediction)
predictions$forecast=max_prediction
predictionsDuring our analysis, we also formed a linear regression model. After careful consideration; we decided that making predictions with a linear model in such setting would require predicting the predictors as well, leading to inflated error rates. So this linear regression model was not extended, and none of our predictions were made using this model.
# prediction with Linear regression
predict_lin=function(data,code){
product=data[data$product_content_id==code]
#product=product[,-season]
product$product_content_id= as.numeric(product$product_content_id)
ss <- sample(1:2,size=nrow(product),replace=TRUE,prob=c(0.7,0.3))
train <- product[ss==1,]
test <- product[ss==2,]
linmod=lm(sold_count~., train)
prediction=predict(linmod,test)
return(prediction)
}
product=data[data$product_content_id==6676673]
product$product_content_id= as.numeric(product$product_content_id)
ss <- sample(1:2,size=nrow(product),replace=TRUE,prob=c(0.7,0.3))
train <- product[ss==1,]
test <- product[ss==2,]
linmod=lm(sold_count~.-1, train)
prediction=predict(linmod,test)
summary(linmod)Our final models are using forecast package. Using 30-day and 90-day data, 2 different forecasts are provided.
# Forecast with 90 days
for(i in 1:8){
code=as.numeric(predictions[i,1])
tdata=data[data$product_content_id==code ]
tdata=tdata[((.N)-90):.N,]
time_data=as.ts(tdata)
d=forecast(time_data,h=2)
x=data.frame(d)
x=predicted_soldcount=x[x$Series=="sold_count",3]
predictions[i,2]=x[2]
}
f90=predictions
# Forecast with 30 days
for(i in 1:8){
code=as.numeric(predictions[i,1])
tdata=data[data$product_content_id==code ]
tdata=tdata[((.N)-30):.N,]
time_data=as.ts(tdata)
d=forecast(time_data,h=2)
x=data.frame(d)
x=predicted_soldcount=x[x$Series=="sold_count",3]
predictions[i,2]=x[2]
}
f30=predictions[,2]
combined=cbind(f90,f30)
colnames(combined)=c("product_content_id","forecast 90","forecast 30")
combined