from ReadData import ALPACA_REST,runTicker,ConfigTable,ALPHA_TIMESERIES,GetTimeSlot,SQL_CURSOR import sys import sqlite3 ts = ALPHA_TIMESERIES() sqlcursor = SQL_CURSOR() sc = sqlcursor.cursor() doClean=False doReload=False ticker='DUG' daily_prices,j = ConfigTable(ticker, sqlcursor,ts,'full',hoursdelay=18) daily_prices_365d = GetTimeSlot(daily_prices, days=365) split_dates = daily_prices_365d[daily_prices_365d.splitcoef!=1.0] if len(split_dates)>0: print(split_dates) #print(daily_prices.to_string()) #sc.execute('DROP TABLE DUG') #list_of_tables = table_names = sc.execute("SELECT name from sqlite_master WHERE type ='table' AND name NOT LIKE 'sqlite_%';").fetchall() print("tables: %s" %len(table_names)) #splitID = sc.execute('SELECT COUNT(splitcoef) from DUG WHERE splitcoef!=1.0 AND Date>2021-01-23').fetchall()[0][0] splitID = sc.execute("SELECT COUNT(splitcoef) from SPY WHERE splitcoef!=1.0 AND Date>'2021-01-23'").fetchall()[0][0] #splitID = sc.execute("SELECT * from DUG WHERE splitcoef!=1.0 AND Date>'2021-01-23'").fetchall()[0][0] splitIDl = sc.execute("SELECT * from DUG WHERE splitcoef!=1.0 AND Date>'2021-01-23'").fetchall()[0] print(splitID) print(splitIDl) # create a list to reload
'180d_return', 'volatitilty', '5d_vol', '30d_vol', '180d_vol' ] earn_points = [ 'ShortPercentFloat', 'PercentInsiders', 'PercentInstitutions', 'PERatio', 'ForwardPE', 'MarketCapitalization', 'AnalystTargetPrice', 'Industry', 'Sector' ] data_points_kmeans = data_points stock_returns = pd.DataFrame(columns=['ticker'] + data_points) if doEarnings: stock_returns = pd.DataFrame(columns=['ticker'] + data_points + earn_points) data_points_kmeans = data_points + earn_points runListA = b.stock_list runList = [] allList = sqlcursor.cursor().execute( 'SELECT name FROM sqlite_master WHERE type="table"').fetchall() for a in runListA: if [a[0]] not in runList: runList += [[a[0]]] for a in allList: #if len(runList)>1000: # break if [a[0]] not in runList: runList += [[a[0]]] print('Running: ', len(runList)) #for s in b.etfs: for s in runList: if s[0].count('^'): continue print(s[0])
from ReadData import SQL_CURSOR import sqlite3 s = SQL_CURSOR() sc = s.cursor() table_names = sc.execute( "SELECT name from sqlite_master WHERE type ='table' AND name NOT LIKE 'sqlite_%';" ).fetchall() for tname in table_names: #print(tname[0]) if tname[0].count('-'): continue try: #print(sc.execute('SELECT MIN(rowid) from SPY GROUP BY Date').fetchall()) distin = sc.execute('SELECT COUNT(DISTINCT Date) from %s' % tname[0]).fetchall()[0][0] allD = sc.execute('SELECT COUNT(Date) from %s' % tname[0]).fetchall()[0][0] if abs(allD - distin) > 0: print(distin, allD, tname[0]) #print(sc.execute('SELECT Date from %s' %tname[0]).fetchall()) print( 'DELETE FROM %s WHERE rowid NOT IN ( SELECT MIN(rowid) from %s GROUP BY Date)' % (tname[0], tname[0])) #sc.execute('DELETE FROM %s WHERE rowid NOT IN ( SELECT MIN(rowid) from %s GROUP BY Date)' %(tname[0],tname[0])) sc.execute('DROP TABLE %s' % (tname[0])) #print(sc.execute('SELECT COUNT( Date) from %s' %tname[0]).fetchall()[0][0]) except sqlite3.OperationalError: print('Could not load!') sc.close() #DELETE FROM lipo