def getprediction(series, rdate, elements):
rdateg = mjd2g(rdate)
qdate = str(rdateg[0]) + '-' + str(rdateg[1]) + '-' + str(rdateg[2])
elements = 1 #!??!?!?!? I must have copied the select statement in without considering approach, rethink it!
# do we get this from econ or econsmooth??
# todo: test with element =1 adjust query to get n elements and sort to take the
# top one on the list
cursor.execute(
"""SELECT value FROM econ
WHERE series=%s AND date > %s order by date desc LIMIT %s""",
(series, qdate, elements))
data = cursor.fetchall()
assert cursor.rowcount > 0, 'no data returned - rdate=%d (%s)' % (rdate,
qdate)
return data
def validmktday(series, rdate):
rdateg = mjd2g(rdate)
qdate = str(rdateg[0]) + '-' + str(rdateg[1]) + '-' + str(rdateg[2])
#print "validmktday: ", series, rdate, qdate
cursor.execute(
"""SELECT value FROM econ
WHERE series=%s AND date=%s""", (series, qdate))
data = cursor.fetchall()
assert cursor.rowcount == 1 or cursor.rowcount == 0, 'validmktday: invalid row count %s %s' % (
series, qdate)
# print "validmktday: ", cursor.rowcount
if cursor.rowcount == 1:
validday = 1
else:
validday = 0
print validday
return validday
def getseriesdata(x, rdate, datasrc):
# recheck how this was done before
series = x[0]
elements = int(x[1])
lag = 0 # lag will be determined by doing a dictionary lookup lag=lags{series}?
if nolags:
lag = 0 # used for production forecasting when we want the best data avail.
rdateg = mjd2g(rdate - lag)
qdate = str(rdateg[0]) + '-' + str(rdateg[1]) + '-' + str(rdateg[2])
if datasrc == 'econsmooth':
cursor.execute(
"""SELECT value FROM econsmooth WHERE series=%s AND date < %s order by date desc LIMIT %s""",
(series, qdate, elements))
else:
cursor.execute(
"""SELECT value FROM econ WHERE series=%s AND date < %s order by date desc LIMIT %s""",
(series, qdate, elements))
data = cursor.fetchall()
# seems to come out of here as a 2xn array, but the [0] column is used and the [1] is empty
assert cursor.rowcount > 0, 'no data returned - rdate=%d series=%s' % (
rdate, series)
return data
def makerecs(forecastmode):
global tinputs
global toutput
global firstdatemjd
global lastdatemjd
#
# value - refers to a data element that is fully prepared to go into trng file
if (forecastmode == "yes"):
lastdatemjd = firstdatemjd + 1
print "forecast mode is YES so first and lastmjd is the same"
print firstdatemjd
print lastdatemjd
for rdate in range(firstdatemjd, lastdatemjd):
tinputs = []
toutput = []
# todo: check for valid market day
# todo op: check for valid market day in forecast
rdateg = mjd2g(rdate)
print 'rdate is: %d (%4d-%02d-%02d)' % (rdate, rdateg[0], rdateg[1],
rdateg[2])
print 'mktseries: ', mktseries
if validmktday(mktseries, rdate):
# for series in
for x in parm_seriesprm: # x is (series, elements, lag)
valuesrc = x[0]
datasrc = 'econsmooth'
data = getseriesdata(x, rdate, datasrc)
for value in data:
trainingdata(
value[0],
valuesrc) # use [0] to make sure it is a simple value
# for seriesma in
for x in parm_seriesmaprm:
valuesrc = x[0]
datasrc = 'econ'
value = getseriesmadata(x, rdate, datasrc)
trainingdata(
value, valuesrc
) # the value is simple now after going through average
# do remaining items
if (timelinefmjd > 0):
valuesrc = '*'
value = (rdate - timelinefmjd) * timeline_sfactor
#print value
trainingdata(value, valuesrc)
if (parm_anlsaw):
valuesrc = '*'
value = doy(rdateg) / 366
trainingdata(value, valuesrc)
if (parm_anlsin):
valuesrc = '*'
value = sin(doy(rdateg) / 366 * pi) / 2 + 0.5
trainingdata(value, valuesrc)
if (parm_anlcos):
valuesrc = '*'
value = cos(doy(rdateg) / 366 * pi) / 2 + 0.5
trainingdata(value, valuesrc)
# forecast mode: would skip this section, write a zero or leave the record one element short?
# get prediction
fcseries = 'INDEX_SPX'
fcelements = 1
value = getprediction(fcseries, rdate, fcelements)
valuesrc = fcseries
trainingdata(
value[0][0], valuesrc,
"forecast") # use [0] to make sure it is a scalor value
# print 'forecast value: ', value[0][0], rdate
close_training_rec()
else:
pass
print "non-market day: ", mjd2g(rdate)
titems_p = titems_p + (parm_anlsaw + parm_anlsin + parm_anlcos)
if (timelinefmjd): titems_p = titems_p + 1
# print ident
print 'ident: ', parm_ident
print 'annual cycles(s,s,c): ', parm_anlsaw, parm_anlsin, parm_anlcos
print 'series slots...'
for row in parm_seriesprm:
print row
print 'seriesma - moving averages...'
for row in parm_seriesmaprm:
print row
print 'seriesslp - series slope...'
for row in parm_seriesslpprm:
print row
fdateg = mjd2g(firstdatemjd)
fdate = str(fdateg[0]) + '-' + str(fdateg[1]) + '-' + str(fdateg[2])
ldateg = mjd2g(lastdatemjd)
ldate = str(ldateg[0]) + '-' + str(ldateg[1]) + '-' + str(ldateg[2])
print 'rdate range: %s (%d) - %s (%d)' % (fdate, firstdatemjd, ldate,
lastdatemjd)
print 'elements: ', titems_p + 1
print '--- assembling records now ', time.strftime(STR_TIME_FORMAT,
time.localtime())
sys.stdout.flush()
# now assemble the records
makerecs(forecastmode)
# assert (trec_p= trec) print "ERROR: param specification of elements
# different than actual number of records present:" trecs, trecs_p