def make_exchange_body(name, ex, C, SU, TU, nu):
rows = []
U = ["KRW", "BTC", "ETH", "USDT"]
for u in U:
if u in SU:
currencyOpt = u == "KRW" or u == "USDT"
if (issubclass(type(ex), get_type_u(u))):
rows.append(["%s (%s)" % (name, u)] + util.make_row(
C, get_u_last(ex, u), currencyOpt=currencyOpt))
if (u != nu and issubclass(type(ex), get_type_u(u))):
currencyOpt = nu == "KRW" or nu == "USDT"
if u in TU:
NP = make_NP(ex=ex, tu=u, nu=nu)
rows.append(
["%s (%s->%s)" % (name, u, nu)] + util.make_row(
C, NP, roundOpt=8, currencyOpt=currencyOpt))
return rows
def make_NRPR_body(
EX, RP, C, TU, nu, SA
): # EX:Exchange, RP : RNPR pairs, C : Currencies, TU : Target Unit, nu : norm unit, SA : slack alarm
rows = []
for tu in TU:
for rp in RP:
if (rp[0] == rp[1] and tu == nu):
continue
NP_0 = make_NP(ex=rp[0], tu=tu, nu=nu)
NP_1 = make_NP(ex=rp[1], tu=nu, nu=nu)
if NP_0 is None or NP_1 is None:
continue
elif issubclass(type(rp[0]), get_type_u(tu)) == False:
continue
diff = util.get_diff_last(NP_0, NP_1)
rows.append(
["%s(%s)/%s(%s) (+)" % (rp[0].symbol, tu, rp[1].symbol, nu)] +
util.make_row(C, {k: v
for (k, v) in diff.items() if v > 100}, 4))
rows.append(
["%s(%s)/%s(%s) (-)" % (rp[0].symbol, tu, rp[1].symbol, nu)] +
util.make_row(C, {k: v
for (k, v) in diff.items() if v <= 100}, 4))
return rows
def make_ts(cursor):
i = 0
for row in cursor:
out = make_row(row, location, ds, si)
yield out
def run_model(coll, data, location, si, ds, count):
name = 'measured_flow'
MODEL_PARAMS = model_params3
# MODEL_PARAMS['modelParams']['sensorParams']['encoders'][name] = input_encoder
shifter = InferenceShifter()
# classifier_encoder = {k: v for k, v in input_encoder.items()}
# classifier_encoder['classifierOnly'] = True
# classifier_encoder['name'] = '_classifierInput'
# MODEL_PARAMS['modelParams']['sensorParams']['encoders']['_classifierInput'] = classifier_encoder
model = ModelFactory.create(MODEL_PARAMS)
model.enableInference({'predictedField': 'measured_flow'})
readings_per_week = 288 * 7
if ds == 'sm':
# get approx readings per week
sdate = datetime(2012, 3, 1)
readings_per_week = coll.database['scats_sm_small'].find({
"site_no":
location['site_no'],
"strategic_input":
int(si),
"datetime": {
"$gte": sdate,
"$lte": sdate + timedelta(days=7)
}
}).count()
print("Readings per week for {}: {}".format(ds, readings_per_week))
anomaly_likelihood_helper = anomaly_likelihood.AnomalyLikelihood(
historicWindowSize=readings_per_week)
# output = nupic_anomaly_output.NuPICPlotOutput(location['site_no'])
prog = tqdm(total=count, desc="HTM")
for row in data:
to_process = make_row(row, location, ds, si)
result = model.run(to_process)
result = shifter.shift(result)
raw_anomaly_score = result.inferences['anomalyScore']
likelihood = anomaly_likelihood_helper.anomalyProbability(
to_process['measured_flow'], raw_anomaly_score,
to_process['datetime'])
pred = result.inferences["multiStepBestPredictions"][1]
# output.write(to_process['datetime'], to_process['measured_flow'], pred, raw_anomaly_score)
# print("observed:", last, "predicted:", pred)
# last = to_process['measured_flow']
# print("raw anomaly:", raw_anomaly_score, "likelihood:", likelihood)
if likelihood >= threshold:
try:
doc = {
'site_no': location['site_no'],
'strategic_input': si,
'algorithm': 'HTM',
'datetime': row['datetime'],
'ds': ds,
'other': {
'likelihood': float(likelihood),
'score': float(raw_anomaly_score)
}
}
# print(doc)
coll.insert_one(doc)
except Exception as e:
print(e)
prog.update()