seifolah-ghaderi / hub_events_challenge

hubs interview project

The Hub events challenge

Achievements: Calculate two below metrics for hub supplier

• supplier average review
• supplier acceptance ratio

Author: Seifolah Ghaderi Repo: https://github.com/seifolah-ghaderi/hub_events_challenge

Assumptions:

• A postgres database is up and running on localhost and port 5432
• A database with name postgres is ready with following credentials:
  • user: postgres
  • pwd: pass
• Two folder exists in code:
  • sql : contains hubs_events.sql that contain table definition and raw initial data
  • csv : the data loaded first time are saved here to load from any jupyter notebook or other execution when database is not run.

Customize database

if you have your own database you should change the following lines in main.py with your credential :

• db_conn global variable:

db_conn = 'postgresql+psycopg2://postgres:pass@127.0.0.1'

• in function delete_today_records :

conn = psycopg2.connect(database="postgres", user='postgres', password='pass', host='127.0.0.1', port='5432')

Prepare and run code

Database

you can use docker-compose to run a postgres database and automatically load sql script. Be sure that you are in code folder and then run:

  docker-compose up -d

be sure that postgres is run by checking:

docker ps

running code:

This is a pure python code so just make sure you have created a virtual environment and then run code by :

cd Folder(source code):

python3 -m venv venv  
source venv/bin/activate 
python main.py

output

The output for each metric will be print in console window(in vscode or pycharm). and the table supplier_score_metrics will be created and populated in database.

You can query table by running :

SELECT supplier_id, value, metric, calculated_at
FROM public.supplier_score_metrics;

Milestones:

• load dataframe from csv( or from postgres table)
• calculate acceptance ratio metric
• calculate average review metric
• prepare to be ready for insert in database
• write them to metric table

Supplier Metrics:

• Average review
• Acceptance ratio
• Response time (suggested by me)
• Deliver time (suggested by me)

Acceptance ratio percentage:

 ( accepted aorders / total  assigned orders) * 100

• assigned orders could be find in event: order/execute/customer/status/processing
• accepted orders could be find in event : order/execute/customer/status/payment

functions & processes

parser & extract --> calculate aggregate --> calculate metrics

• 1:parser & extractor functions:

- get_assigned_orders 
- get_accepted_orders

Those above function get df as master dataframe and by searching relative event extract json data and convert it to dataframe .

df_mask = df['name'] == 'order/execute/customer/status/processing'
df_mask = df['name'] == 'order/execute/customer/status/payment'

extraxt and convert to datafram done using the pd.json_normalize function.

• 2:calculate aggregate functions:

- cal_all_order_count 
- cal_accepted_orders_count

Those two above function calculate total & accepted orders for each supplier using dataframe returned from parser functions.

df_all.groupby(['supplier_id']).count()

• 3:main function

cal_accept_ratio

this is the final function that calculate acceptance ratio for each supplier.

    df_all_o = get_assigned_orders(df)
    df_accepteds_o = get_accepted_orders(df)

    df_all = cal_all_order_count(df_all_o)
    df_accepteds = cal_accepted_orders_count(df_accepteds_o)

    ratio_df = pd.merge(df_all, df_accepteds)

    ratio_df['value'] = round((ratio_df['accepted'] / ratio_df['total']) * 100).astype(int)

Average review metric

Contains two metrics:

- review_value_speed
- review_value_print_quality

These two metric are available in 'node/review/created' events plus some modification and deleted row in 'node/review/updated' & 'node/review/deleted' evenets. ###challenges : updated and removed reviews should be considered. So we developed three functions:

- get_all_reviews
- get_upd_reviews
- get_del_reviews

These functions in order : get all review , updated and deleted reviews. After getting them we update reviews by apply modified rows and also delete removed reviews. The function that does the main calculation is :

cal_review

I calculated both review_value_speed and review_value_print_quality as two separated metrics for each supplier. And then make a new column to calculate their average .These all done in cal_review function.

  avg ( review_value_speed ) + avg(review_value_print_quality)/ 2

    grouped_df = df_reviews.groupby(['supplier_id']).mean().astype(int)
    grouped_df['value'] = round((grouped_df['review_speed'] + grouped_df['review_quality']) / 2).astype(int)

Checking data

• Update event with missed order_id :

What i found during checking data is that there are some updated review that not exists in created reviews !.maybe they belong to another time windows. For example below record not exists in created event but exists in updated event:

event	            supplier_id	order_id	    timestamp	            customer_id	    review_speed	review_quality
node/review/updated	34872	    3194344	    2017-01-03T22:10:58.000Z	  154527	        100	            80          

• multiple review for single order!: during removing deleted reviewes I found that there are multiple reviews from one customer_id for a single order and supplier !

event	                supplier_id	order_id	    timestamp	            customer_id	    review_speed	review_quality
node/review/deleted	    30352	    3994815	    2017-01-11T08:37:47.000Z	51468	        100	                100
node/review/deleted	    30352	    3994815	    2017-01-11T08:28:40.000Z	51468	        100	                100

This will be meaning full when you decide to delete reviews by only mathing order_id.I had 5 deleted review and when i deleted their order_id from reviews it removed 6 records indicate that we have multiple order_id for resolve it i included customer_id in filtering and removing reviews and then it removed exactly as i expected, 5 rows~

cond = df_reviews['order_id'].isin(df_del_reviews['order_id']) & df_reviews['customer_id'].isin(
        df_del_reviews['customer_id'])
    df_reviews.drop(df_reviews[cond].index, inplace=True)

Extra Metrics for Suppliers

• Average Response time (suggested by me) : avg (accepted_time - assigned time)
• Average Deliver time (suggested by me) : avg (delivered time - start time)

Average Response time:

• assigned time : When customer assign an order to a supplier during 'order/execute/customer/status/processing' event
• accepted time : When supplier review and accept an order during 'order/execute/customer/status/payment' event The Supplier Average Response time indicate how much a supplier is timely and punctual or how much he respects customer time ! This metric calculated in hour and when you lok at the result it's interesting that range of response to an order is varies form less than one hour to 16h !

	supplier_id	avg_respon_h
0	100130	        0
1	114087	        0
2	170225	        0
3	62909	        1
4	18815	        2
5	192417	        2
6	34872	        3
7	84241	        8
8	30352	        11
9	4450	        13
10	69607	        16

The function cal_sup_resp_time used for calculate this metric.

Average Deliver time:

• start time : When supplier start manufacturing during 'order/execute/status/customer/printing' event
• deliver time : When supplier finished order and order delivered to customer during 'order/execute/status/customer/successful' event

Unfortunately i couldn't find any event order/execute/status/customer/successful in the provided data and can not calculate this metrics. But the process is same as Average Response time metric.