The pandemic and quarantine have changed the life and behavior of almost every inhabitant of the planet. At the same time, some changes are short-term and disappear with the lifting of quarantine measures, while others may remain with us for a long time, perhaps even forever.
, Dentsu Aegis Network, , . , 2020 , .
, () , . (, , ) , .
, , , .
, , , .
.
:
— , ( , - , , , .., , - ).
— , . , .
TVR — , .
Affinity — , , , (TVR / TVR ).
.
· . Mediascope, . – . , (, ) . , , . : , , , ..
· . () . , : \ , , (, ..). .
:
· COVID-19. : , , . - 2020 2019 , - 2020 , - , .
· 2020 . 100 (100+), 2020 , 100 (100-). , ( 100 ) , ( 100 ). , . , , , TVR.
· . “” TVR 0.10. - TVR 0.05 0.15 ( ), 0.05 / 0.10 = 50% 0.15 / 0.10 = 150%. .
· , , . .
· .
·
, , : MS SQL+Python+LightGBM(CPU) – , .
: TVR, , , . , . ( ).
, Mediascope . 6 . , 4 , , , (12*4*[ =130] *[ -=24] = 150. ) .
. 40 , (60*24=1440), 36 . , 6 ( ).
2 – . , , , , .
, , . 130 .
. - , , , . , .
. - , - . , :
· : PCA .
· 100- . , 100-. .
· Anomaly detection. \ TVR . isolation forest - .
· -.
· : , , , -, , , 30 , (, “ ”), . , .
· . , :
1.
a. dummy + “100-”, , dummy .
b. – , .
c. . .
2. .1 + dummy + “100-” ( ), dummy , .
3. dummy : . , “100-”, .
4. .3 + + «100-», , . .
5. (!) : + «100-», .
· COVID-19.
1. .
2. :
a. 2020-03-05 - “”
b. 2020-03-25 -
c. 2020-03-28 –
d. 2020-04-06 -
e. 2020-06-01 -
f. 2020-06-14 -
3. .1 - . .
:
· ,
· ,
· ( ) , .
, (, ). , . boxplot () 2 4 .
· .
1. 95 ( ) . , .
2. : TVR ( .1), TVR .
3. TVR , .
· TVR log(TVR+1). TVR, . .
. , - .
https://download.dentsuaegis.ru/index.php/s/emtvlKKdPkUCdvn
Machine Learning Engineer developer. ML Engineer Clean Code Paradigm, . Data Engineer’. , : MS SQL server .
-, MS SQL. . SQLAlchemy . , , (, MS SQL ?). : deadlock’, . , , pyodbc. :
1. ( ) . constraints, race condition deadlock.
2. , . , :
a. Insert into table(col1,col2) values (1,1);
Insert into table(col1,col2) values (1,2);
b. Insert into table(col1,col2) values
(1,1),
(1,2);
, ) 2 , ) , . 1000 .
3. \ merge tables target table ( target table).
. 2 10 - . , , - . , , , MS SQL.
-, . , : . , , , . . :
, , \ . (, ) .
.1.
, . . . . . tableau. , , , , . avg[(fact – predict) / (fact +predict +epsilon)]. boxplot . . :
· TVR . R_squared – .
· \ \
· - TVR.
· – , .
, 22 . . 2 4 : , .
:
· (, ..).
· -: - , - ( ).
· : TVR 2 , .
· - . .