Scripts for tables transformation
Test different features combinations by fitting rotating features to linear regression from the pool and cross-validate models on N folds, thus total quantity is: combinations * N
args:
- file_name - {str} name of the file with transformed features
- date_time_col - {str} [date_time] feature column name. filled with N values (1, 2, 3, ... n_observations)
- max_date_num_col - {int} last number [date_time] value / n_observation
- const_features - {list} of constant features that don't have to be rotated !!!Y_val HAVE TO BE FIRST IN LIST!!!
- to_rotate - {list} list of regular expressions for each media feature to rotate ['SUB_OLV_Imp_d0.?[2-8]', 'SUB_OOH_d0.?[2-8]']
- regex_decode - {list} list of media to rotate in normal form ['OLV', 'OOH']
- nfolds - {int} number of CV splits
- save - {str} name of file to save with
return:
- res - {pd.DataFrame} pivoted look like feature and stats table
from cv_model_selection import model_selection
model_selection(
file_name = 'bike_sharing_demand',
date_time_col = False,
max_date_num_col = False,
const_features = ['season', 'holiday', 'weather', 'usd'],
to_rotate = ['TV_d?[0-9]', 'OOH_d?[0-9]', 'OLV_d?[0-9]', 'Comp_d?[0-9]'],
regex_decode = ['TV', 'OOH', 'OLV', 'Comp'],
nfolds = 5,
save = 'filename'
)
Calculate distance beween each competitior/OOH object and client store
- file_name - name of file with client and competitor/OOH objects, latitude, longitude and open-close dates
- client - column of client's objects names
- competitor - column of competitor's objects names
- competitor_latitude - column of competitor's latitude
- client_latitude - column of client's latitude
- competitor_longitude - column of competitor's longitude
- client_longitude - column of client's longitude
- threshold_tuple - interval (min_dist, max_dist) between objects you need to calculate feature within in metres
- open_date - column with competitor open dates
- close_date - column with competitor close dates
- model_start_date - first model's monday
- model_end_date - the last monday of model
Input and output dataframes:
e.g.:
comp comp_latitude comp_longitude client client_latitude client_longitude comp_open_date comp_close_date
1 30.3746659 60.0453568 store_1 59.9268 30.3161 2016-04-04 2019-11-08
2 30.3144672 59.9571587 store_2 59.9258 30.3144 2016-04-04
n 30.338264 59.862598 store_n 60.0264 30.2229 2016-04-11 2019-06-04
to
date client_store value
2016-03-28 store_1 0
2016-04-04 store_1 1
2016-04-11 store_1 2
2016-04-18 store_1 2
Main func of the script
from distance_calculation import dist_feature
dist_feature(file_name = 'dist_filename',
client = 'client_store_col_name',
competitor = 'competitor_store_col_name',
competitor_latitude = 'competitor_latitude_col_name',
client_latitude = 'client_latitude_col_name',
competitor_longitude = 'competitor_longitude_col_name',
client_longitude = 'client_latitude_col_name',
threshold_tuple = (0, 200),
open_date = 'competitor_store_open_date_col_name',
close_date = 'competitor_store_close_date_col_name',
model_start_date = '2017-01-02',
model_end_date = '2020-02-17'
)
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent