def predict2csv(self, code, startd, endd, csvpath):
tfl = TfLearning(meigaras=[code], restore_first=False)
(restored, ver, accuracy) = tfl.restore(code)
if restored == False:
return False
(dates, ximage) = tfl.get_meigara_xs(code, startd, endd)
result = tfl.predict(ximage)
data = []
for i in range(len(dates)):
if dates[i] <= ver:
continue
row = [code, dates[i], ver, accuracy]
row.extend(result[i].tolist())
data.append(row)
cf.arr2csv(csvpath, data)
def do_test(startd, endd, meigaras=[]):
interval = 20
if len(meigaras) == 0:
meigaras = kf.get_meigaras()
f.log("Start making reports")
report = []
for code in meigaras:
f.log("Processing meigara:%s" % (code))
kl = KabukaLines(code, startd, endd)
lt = LineTrader(kl)
kabuka = kl.get_kabuka()
if len(kabuka) == 7:
(indexes, dates, open, high, low, close, volume) = kabuka
else:
continue
tmp_endd = dates[-1]
i = 60
old_from_date = ""
while i < len(dates)-interval:
if lt.judge_trade_goodness(i) == False:
i += interval
continue
tmp_startd = dates[i]
tmp_next_startd = dates[i+interval]
(trade_mode_str, interest, from_date, to_date, spent, start_price, end_price, endi) \
= lt.test(tmp_startd, tmp_endd, tmp_next_startd)
if old_from_date != from_date:
if trade_mode_str != "":
report.append([code, trade_mode_str, interest, from_date, to_date, spent, start_price, end_price])
old_from_date = from_date
if endi > i:
i = endi
else:
i += interval
f.arr2csv("%s/report.csv" % (TMP_DIR), report)
f.log("Finished making reports:%s/report.csv" % (TMP_DIR))
def save(self):
path = "%s/%s" % (CSV_DIR, self.cluster_centers_csv)
labels = np.matrix(self.km.predict(self.km.cluster_centers_)).T
labels = np.array(labels)
centers = self.km.cluster_centers_
f.arr2csv(path, np.c_[labels, centers], "w")
def save(self):
path = "%s/%s" % (CSV_DIR, self.cluster_centers_csv)
coef = self.sgdcls.coef_
intercept = self.sgdcls.intercept_
f.arr2csv(self.coef_csv, coef, "w")
f.arr2csv(self.intercept_csv, intercept, "w")
def save(km, cluster_centers_csv):
path = "%s/%s" % (CSV_DIR, cluster_centers_csv)
f.arr2csv(path, km.cluster_centers_, "w")
