Meli 2021 Dataset

library(arrow)
library(skimr)
library(jsonlite)
library(dplyr)
library(skimr)

Reading Parquet files

train_data<-arrow::read_parquet("datasets/train_data.parquet")
train_data %>% group_by(sku) %>% summarise(n=n())

train_data %>% filter(sku==12101)  

NA

skimr::skim(
train_data %>% select_if(is.numeric)) %>%
  yank("numeric")

── Variable type: numeric ───────────────────────────────────────────────────────────────────────────────────────────────────────────────────
  skim_variable  n_missing complete_rate       mean         sd    p0      p25    p50     p75      p100 hist 
1 sku                    0             1 330374.     190792.    0    165125   330380 495596     660915 ▇▇▇▇▇
2 sold_quantity          0             1      0.990       9.99  0         0        0      1       6951 ▇▁▁▁▁
3 current_price          0             1   2358.    1263914.    0.87     48.7    119    299. 999999999 ▇▁▁▁▁
4 minutes_active         0             1   1043.        630.    0         0     1440   1440       1440 ▃▁▁▁▇

Read JSON lines files

library(tidyjson)
library(tidyr)
metadata<-jsonlite::stream_in(file("datasets/items_static_metadata_full.jl"),verbose = 0)
metadata %>% filter(item_domain_id=="MLB-SNEAKERS") 

Countries and item_id are separated into two different columns

metadata<-metadata %>% separate(item_domain_id,sep = "-", into=c("country","item_domain_id"))
metadata  %>% group_by(country,item_domain_id) %>% summarise(n=n())

`summarise()` has grouped output by 'country'. You can override using the `.groups` argument.

JOIN both data.frames

train_data_meta<-train_data %>% inner_join(metadata %>% select(country,sku,item_domain_id), by="sku")
train_data_meta

Now we can explore…

library(ggplot2)
library(lubridate)
library(ggdark)
serie_ex<-train_data_meta %>% filter(country=="MLM" & item_domain_id=="ADHESIVE_TAPES") %>% group_by(date) %>% summarise(sold=sum(sold_quantity)) 


serie_ex %>% ggplot()+
  geom_point(aes(x=ymd(date),y=sold),color='orange')+
  geom_line(aes(x=ymd(date),y=sold),color='red')+
  ggdark::dark_theme_bw()+
  xlab("Date")+
  ylab("Sold Quantity")+
  labs(title="SNEAKERS Sold in Mexico for 30 days")+
  theme(axis.text.x = element_text(angle = 45, vjust = 0.5, hjust=1))

#train_data_meta %>% filter(country=="MLM" & item_domain_id=="SNEAKERS") %>% filter(date=="2021-02-01") %>% filter(sold_quantity!=0)

Boxplots

train_data_meta_sum<-train_data_meta %>% group_by(sku,dow) %>% summarise(total_sold_qty=sum(sold_quantity)) 

`summarise()` has grouped output by 'sku'. You can override using the `.groups` argument.
Warning message:
package ‘arrow’ was built under R version 4.0.5 

train_data_meta_sum<-train_data_meta %>% group_by(sku,dow) %>% summarise(total_sold_qty=sum(sold_quantity)) 

`summarise()` has grouped output by 'sku'. You can override using the `.groups` argument.

train_data_meta_sum %>%
  ggplot(aes(x=dow,y=total_sold_qty))+
 # geom_point()+
  #geom_violin()+
  geom_boxplot(outlier.shape=NA)+
  ylim(0,50)+
  xlab("day_of_the_week")+
  ggdark::dark_theme_bw()

Inverted geom defaults of fill and color/colour.
To change them back, use invert_geom_defaults().

Multitple time series.

train_data_meta<-train_data_meta %>% mutate(dow=ymd(date) %>% wday(label=TRUE), week=week(date))
train_data_meta_sum_m <- train_data_meta %>% group_by(item_domain_id,week,dow) %>% summarise(total_sold_qty=sum(sold_quantity)) 

`summarise()` has grouped output by 'item_domain_id', 'week'. You can override using the `.groups` argument.

sku_ex<-train_data_meta_sum_m  %>% filter(week!=5) %>% filter(week!=13) %>% filter(item_domain_id=="ADHESIVE_TAPES")
sku_ex %>%
  ggplot(aes(x=dow,y=total_sold_qty))+
 # facet_wrap(~week)+
  geom_point(color='orange')+
  geom_smooth(aes(x=as.numeric(dow)),method = "loess",se = TRUE)+
  #geom_area(aes(x=dow,y=total_sold_qty,group=week,fill = week), position = "stack") +
  #geom_line(aes(group=week))+
  #geom_smooth(aes(x=dow,y=total_sold_qty, gr))+
  xlab("DOW")+
  
  ylab("Sold Quantity")+
  ggdark::dark_theme_bw()+
  labs(title="ADHESIVE_TAPES: Total Qty sold per day of the week")+
  theme(axis.text.x = element_text(angle = 45, vjust = 0.5, hjust=1))

## read https://stackoverflow.com/questions/10357768/plotting-lines-and-the-group-aesthetic-in-ggplot2
  

sku_ex %>%
  ggplot(aes(x=dow,y=total_sold_qty))+
  geom_point(color='orange')+
  geom_line(aes(group=week,color=as.factor(week)))+
  xlab("DOW")+
  ylab("Sold Quantity")+
  ggdark::dark_theme_bw()+
  labs(title="ADHESIVE_TAPES: Total Qty sold per day of the week")+
  theme(axis.text.x = element_text(angle = 45, vjust = 0.5, hjust=1))

using FACET

sku_ex %>%
  ggplot(aes(x=dow,y=total_sold_qty))+
  facet_wrap(~week)+
  geom_point(color='orange')+
  geom_line(aes(group=1))+
  xlab("DOW")+
  ylab("Sold Quantity")+
  ggdark::dark_theme_bw()+
  labs(title="ADHESIVE_TAPES: Total Qty sold per day of the week")+
  theme(axis.text.x = element_text(angle = 45, vjust = 0.5, hjust=1))