def train(self):
LOGGER.info("Starting to train...")
ticker_names = get_ticker_names(self.db, "TSX")[:10]
random.shuffle(ticker_names)
sdate = datetime.today().date() - timedelta(days=2*365)
edate = sdate + timedelta(days=365)
data, targets = [], []
for t in ticker_names:
c = self.rc.calculate_return(t, sdate, edate)
if c is None:
continue
d = self.ff.fetch(t, sdate, edate)
d = list(itertools.chain.from_iterable(d))
targets.append(c)
data.append(d)
train_tickers = ceil(len(targets) * self.training_perc)
self.training_data = data[:train_tickers]
self.training_target = targets[:train_tickers]
self.goal_data = data[train_tickers:]
self.goal_target = targets[train_tickers:]
LOGGER.info("Starting to train...")
print("TEST", self.training_data, self.training_target)
self.lr.fit(self.training_data, self.training_target)
def ticker_history(self, start, end, ticker, info='quote', start_dic={}):
"""
Gets and returns the historic prices for a given ticker for between
the time period provided. Inclusive.
"""
start_str = start.strftime('%Y%m%d')
end_str = end.strftime('%Y%m%d')
# info = 'quote', 'dividend', 'split'
try:
data = yqd.load_yahoo_quote(ticker, start_str, end_str, info=info)
except (HTTPError, URLError, gaierror) as e:
LOGGER.info("Yahoo request failed. Blocked?")
return []
titles = tuple(t.replace(' ', '_').lower() for t in data[0].split(','))
history = []
for row in data[1:-1]:
history_row = {k: v for k, v in start_dic.items()}
iter_list = row.split(',')
for element, title in zip(iter_list, titles):
converted = self.convert_yahoo_element(element)
history_row[title] = converted
history.append(history_row)
return history
def train(self):
LOGGER.info("Starting to train...")
train_data, train_targets, test_data, test_targets = self.fh.fetch_feature_data(
)
tickers = self.fh.fetch_feature_tickers()
print("Shapes", train_data.shape, train_targets.shape, test_data.shape,
test_targets.shape)
features_names = self.fh.fetch_feature_names()
self.kbest.fit(train_data, train_targets)
feature_scores = self.kbest.scores_
combined = [
tuple((s, f)) for s, f in zip(feature_scores, features_names)
if not np.isnan(s)
]
best_features = sorted(combined, key=lambda x: -x[0])
scores, names = zip(*best_features)
self.graph(scores, names)
print('Best Features', scores[:5])
print('Worst Features', best_features[-5:])
def binary_train(self):
LOGGER.info("Starting to train...")
train_data, train_targets, test_data, test_targets = self.fh.fetch_binary_feature_data(
)
train_tickers, test_tickers = self.fh.fetch_feature_tickers()
scaler = StandardScaler()
train_data = scaler.fit_transform(train_data)
test_data = scaler.fit_transform(test_data)
print("Shapes", train_data.shape, train_targets.shape, test_data.shape,
test_targets.shape)
self.mlpc.fit(train_data, train_targets)
predictions = self.mlpc.predict(test_data)
acc_score = accuracy_score(test_targets, predictions)
roc_score = roc_auc_score(test_targets, predictions)
print('???', accuracy_score(test_targets, [True] * len(test_targets)))
print("Accuracy Score", acc_score, 'ROC Score', roc_score)
print("Average True Return", sum(train_targets) / len(train_targets))
def get_historic_events(self):
"""
Gets all the historical events from yahoo, updating only the new entries
based on the date of the last fetch.
"""
exchange = "TSX"
listings = self.session.query(
Listings.ticker,
Listings.dateoflisting).filter(Listings.exchange == exchange)
dict_fields = ["index", "action", "value"]
fields = ["exchange", "ticker", "date", "action", "value"]
total_listings = listings.count()
for counter, (ticker, listdate) in enumerate(listings):
lastdate, = self.session.query(func.max(
EventHistory.updatedate)).filter(
EventHistory.exchange == exchange,
EventHistory.ticker == ticker).first()
startdate = listdate if lastdate is None else lastdate + timedelta(
days=1)
rows = []
if startdate < self.today:
yahoo_ticker = ticker + ".TO"
dividend_dict = self.ticker_history(startdate,
self.today,
yahoo_ticker,
info='dividend')
split_dict = self.ticker_history(startdate,
self.today,
yahoo_ticker,
info='split')
rows = []
for row in dividend_dict:
rows.append([
exchange, ticker, row["date"], "DIVIDEND",
row["dividends"], self.today
])
for row in split_dict:
rows.append([
exchange, ticker, row["date"], "SPLIT",
row["stock_splits"], self.today
])
if rows:
LOGGER.info("{}/{} Inserting {} from {} to {}".format(
counter + 1, total_listings, ticker, startdate,
self.today))
stmt = insert(EventHistory).values(
rows).on_conflict_do_nothing(
constraint='event_history_pkey')
self.session.execute(stmt)
self.session.commit()
else:
LOGGER.info("{}/{} Skipping ticker {}".format(
counter + 1, total_listings, ticker))
def fetch_all(self, exchange):
extension = '.TO'
tickers = tuple(x.ticker + extension for x in self.session.query(Listings.ticker).filter(Listings.exchange == exchange, or_(Listings.onyahoo == True, Listings.onyahoo is None)))
ticker_groups = self.chunks(tickers, 200)
LOGGER.info("Fetching/Updating {} urls.".format(len(ticker_groups)))
for ticker_group in ticker_groups:
url = self.create_url(ticker_group)
self.handle_url(ticker_group, url, exchange)
sleep(1) # limit requests to 1/s
def binary_train(self):
LOGGER.info("Starting to train...")
train_data, train_targets, test_data, test_targets = self.fh.fetch_binary_feature_data()
train_tickers, test_tickers = self.fh.fetch_feature_tickers()
print("Shapes", train_data.shape, train_targets.shape, test_data.shape, test_targets.shape)
self.lc.fit(train_data, train_targets)
predictions = self.lc.predict(test_data)
acc_score = accuracy_score(test_targets, predictions)
roc_score = roc_auc_score(test_targets, predictions)
print("Accuracy Score", acc_score, 'ROC Score', roc_score)
print("Average True Return", sum(train_targets) / len(train_targets), sum(test_targets) / len(test_targets))
def get_historic_prices(self):
"""
Gets all the historical prices from yahoo, updating only the new entries
based on the date of the last fetch.
"""
exchange = "TSX"
listings = list(
self.session.query(
Listings.ticker,
Listings.dateoflisting).filter(Listings.exchange == exchange))
total_listings = len(listings)
for counter, (ticker, listdate) in enumerate(listings):
lastdate, = self.session.query(func.max(PriceHistory.date)).filter(
PriceHistory.exchange == exchange,
PriceHistory.ticker == ticker).first()
startdate = listdate if lastdate is None else lastdate + timedelta(
days=1)
his_dict = []
if startdate < self.today:
yahoo_ticker = ticker + ".TO"
start_dic = {"exchange": exchange, "ticker": ticker}
his_dict = self.ticker_history(startdate,
self.today,
yahoo_ticker,
info="quote",
start_dic=start_dic)
if his_dict:
LOGGER.info("{}/{} Inserting {} from {} to {}".format(
counter, total_listings, ticker, startdate, self.today))
for d in his_dict:
stmt = insert(PriceHistory).values(
d).on_conflict_do_update(
constraint='price_history_pkey', set_=d)
self.session.execute(stmt)
self.session.commit()
else:
LOGGER.info("{}/{} Skipping ticker {}".format(
counter, total_listings, ticker))
def train(self):
LOGGER.info("Starting to train...")
train_data, train_targets, test_data, test_targets = self.fh.fetch_feature_data(
)
tickers = self.fh.fetch_feature_tickers()
print("Shapes", train_data.shape, train_targets.shape, test_data.shape,
test_targets.shape)
self.svm.fit(train_data, train_targets)
predictions = self.svm.predict(test_data)
mean_err = mean_absolute_error(test_targets, predictions)
mean_s_err = mean_squared_error(test_targets, predictions)
print("Got Mean error", mean_err, "Squared Err", mean_s_err)
print("Average Expected Return", sum(predictions) / len(predictions))
print("Average True Return", sum(train_targets) / len(train_targets))
def go(self, urls):
LOGGER.info("Preparing threads...")
manager = Manager()
urlQ = manager.Queue()
for url in urls:
urlQ.put(url)
procs = [
Process(target=get_html,
args=(urlQ, self.callback, self.xpath_hooks),
daemon=True) for i in range(self.nproc)
]
LOGGER.info("Threads started. Fetching n' parsing!")
for proc in procs:
proc.start()
for proc in procs:
proc.join()
def convert_yahoo_element(self, element):
converted = None
try:
converted = float(element)
except ValueError:
try:
converted = datetime.strptime(element, "%Y-%m-%d")
except ValueError:
if element == 'null':
converted = None
elif '/' in element:
try:
a, b = element.split('/')
converted = float(a) / float(b)
except ValueError:
LOGGER.info("Unable to convert {}".format(element))
else:
LOGGER.info("Unable to convert {}".format(element))
return converted
def fetch_all(self, exchange):
q = self.session.query(Listings).filter(
Listings.exchange == exchange,
or_(Listings.onyahoo == True, Listings.onyahoo is None))
extension = '.TO'
urls = [
"https://ca.finance.yahoo.com/quote/{}{}/key-statistics".format(
l.ticker, extension) for l in q
]
xpath_hooks = [
"//section[@data-test='qsp-statistics']",
"//section[@data-test='lookup-page']"
]
LOGGER.info("Fetching/Updating {} urls.".format(len(urls)))
jsps = JSPageScraper(self.dic_parse, xpath_hooks, "key_statistics")
jsps.go(urls)
def get_key_stats(self, ticker, db_exchange="TSX"):
"""
This function get key statistics from
Morning Star.
"""
url = ("http://financials.morningstar.com/ajax/exportKR2CSV.html?t={}&"
"culture=en-CA®ion=CAN&order=asc&r={}").format(
ticker, randint(1, 500000))
req = urllib.request.Request(url, headers=self.headers)
resp = urllib.request.urlopen(req)
csv_r = csv.reader(codecs.iterdecode(resp, 'utf-8'))
on_morningstar = csv_r and resp.headers['content-length'] != '0'
if on_morningstar:
LOGGER.info("Getting key statistics for {}... ".format(ticker))
else:
LOGGER.info("Skipping", ticker)
return 1
col_names, vals = self.parse_csv(
csv_r,
10,
self.special_key_titles,
self.column_key_map,
extra_cols=["ticker", "exchange", "update_date"],
extra_vals=[ticker, db_exchange, self.today])
self.db.insert_into("ms_key_statistics",
col_names,
vals,
unique_conflict=True)
LOGGER.info("Done")
return 0
def get_historic_prices(self):
"""
Gets all the historical prices from yahoo, updating only the new entries
based on the date of the last fetch.
"""
exchange = "TSX"
listings = self.db.select("ticker, dateoflisting",
"listings",
where="exchange = %s",
vals=[exchange])
dict_fields = ["Adj Close", "High", "Close", "Open", "Low", "Date"]
fields = ["exchange", "ticker"] + [x.lower() for x in dict_fields]
total_listings = len(listings)
for counter, (ticker, listdate) in enumerate(listings):
lastdate = self.db.select("MAX(date)",
"price_history",
fetch="one",
where="exchange = %s AND ticker = %s",
vals=[exchange, ticker],
unroll=True)
startdate = listdate if lastdate is None else lastdate + timedelta(
days=1)
his_dict = []
if startdate < self.today:
yahoo_ticker = ticker + ".TO"
his_dict = self.ticker_price_history(startdate, self.today,
yahoo_ticker)
if his_dict:
LOGGER.info("{}/{} Inserting {} from {} to {}".format(
counter, total_listings, ticker, startdate, self.today))
rows = [[exchange, ticker] + [row[k] for k in dict_fields]
for row in his_dict]
self.db.insert_into("price_history", fields, rows)
else:
LOGGER.info("{}/{} Skipping ticker {}".format(
counter, total_listings, ticker))
def get_financial(self, ticker, period_name, exchange="XTSE"):
"""
This function get yearly and quartly information from
Morning Star.
period_name: "quarter" or "annual"
exchanges: XTSE (TSX),
"""
# this converts the morning star exchange name to our database name
if exchange in self.exchange_map:
db_exchange = self.exchange_map[exchange]
else:
raise ValueError("Exchange unsupported {}".format(exchange))
period = 3 if period_name == "quarter" else 12
url = (
"http://financials.morningstar.com/ajax/ReportProcess4CSV.html?&t="
"{}:{}®ion=can&culture=en-US&cur=&reportType=is&period={}&"
"dataType=A&order=desc&columnYear=5&curYearPart=1st5year&"
"rounding=1&view=raw&r={}&denominatorView=raw&number=1").format(
exchange, ticker, period, randint(1, 500000))
req = urllib.request.Request(url, headers=self.headers)
resp = urllib.request.urlopen(req)
csv_r = csv.reader(codecs.iterdecode(resp, 'utf-8'))
on_morningstar = csv_r and resp.headers['content-length'] != '0'
if on_morningstar:
LOGGER.info("Getting {} financial data for {}... ".format(
period_name, ticker))
else:
LOGGER.info("Skipping {}".format(ticker))
return 1
num_cols = 6 if period_name == "quarter" else 5 # skip last column if not quarter view (removes TTM)
return_dics = self.parse_csv(csv_r,
num_cols,
self.special_financials_titles,
self.column_financials_map,
start_dic={
"ticker": ticker,
"exchange": db_exchange,
"period": period,
"update_date": self.today
})
for d in return_dics:
stmt = insert(MorningStarFinancials).values(
d).on_conflict_do_update(constraint='fiscal_year_unique',
set_=d)
self.session.execute(stmt)
self.session.commit()
LOGGER.info("Done")
return 0
def screen_and_save_feature_data(self):
train_ticker_names, test_ticker_names = self.fetch_feature_tickers()
train_data, train_targets, test_data, test_targets = self.fetch_feature_data(
)
tickers = set(find_small_cap_tickers(
self.sess)) # finds ticker < 10m value
train_rm_indexes = []
for i, (ticker,
target) in enumerate(zip(train_ticker_names, train_targets)):
if target > 10 or ticker in tickers:
train_rm_indexes.append(i)
test_rm_indexes = []
for i, (ticker,
target) in enumerate(zip(test_ticker_names, test_targets)):
if target > 10 or ticker in tickers:
test_rm_indexes.append(i)
train_ticker_names = np.delete(train_ticker_names,
train_rm_indexes,
axis=0)
train_data = np.delete(train_data, train_rm_indexes, axis=0)
train_targets = np.delete(train_targets, train_rm_indexes, axis=0)
test_ticker_names = np.delete(test_ticker_names,
test_rm_indexes,
axis=0)
test_data = np.delete(test_data, test_rm_indexes, axis=0)
test_targets = np.delete(test_targets, test_rm_indexes, axis=0)
LOGGER.info("Saving file at: {}".format(self.file_path))
np.savez(self.file_path,
train_data=train_data,
train_targets=train_targets,
train_ticker_names=train_ticker_names,
test_data=test_data,
test_targets=test_targets,
test_ticker_names=test_ticker_names)
def dic_parse(self, session, url, html):
def innerHtml(ele):
return ele.decode_contents(formatter="html")
soup = BeautifulSoup(html, "lxml")
ticker = self.url_ticker_pat.search(url).group(1)
exchange = "TSX"
on_yahoo = soup.find('section', attrs={'data-test': 'lookup-page'
}) is None
session.query(Listings).filter(Listings.exchange == exchange,
Listings.ticker == ticker).update(
{Listings.onyahoo: on_yahoo})
if not on_yahoo: # if quote not found, exit
LOGGER.error("Failed to find quote for {} skipping".format(url))
return
div_test = soup.find('section', attrs={'data-test': 'qsp-statistics'})
if div_test is None:
LOGGER.error("Unknown error for {} skipping".format(url))
return
db_dic = {}
for table in div_test.find_all('table'):
for row in table.find_all('tr'):
td_list = row.find_all('td')
title = innerHtml(td_list[0].find('span'))
val = innerHtml(td_list[1]) if td_list[1].find(
'span') is None else innerHtml(td_list[1].find('span'))
if title in self.y_to_db_map:
db_dic[self.y_to_db_map[title]] = self.parse_numeric(val)
if db_dic:
db_dic["ticker"] = ticker
db_dic["exchange"] = exchange
exists = session.query(KeyStatistics).filter_by(
**db_dic).scalar() is not None
if exists:
LOGGER.info("Skipping {} due to prior existence".format(url))
else:
db_dic["update_date"] = self.today
stmt = insert(KeyStatistics).values(
db_dic).on_conflict_do_nothing(
constraint='key_statistics_pkey', )
session.execute(stmt)
session.commit()
LOGGER.info("Done parsing {}".format(url))
else:
LOGGER.info("Skipping {}".format(url))
def dic_parse(self, db, url, html):
def innerHtml(ele):
return ele.decode_contents(formatter="html")
soup = BeautifulSoup(html, "lxml")
ticker = self.url_ticker_pat.search(url).group(1)
exchange = "TSX"
on_yahoo = soup.find('div', attrs={'data-test': 'unknown-quote'
}) is None
db.update("listings", ["onyahoo"], [on_yahoo],
"exchange=%s AND ticker=%s", [exchange, ticker])
if not on_yahoo: # if quote not found, exit
LOGGER.error("Failed to find quote for", url, "skipping")
return
div_test = soup.find('div', attrs={'data-test': 'qsp-statistics'})
if div_test is None:
LOGGER.error("Unknown error for", url, "skipping")
return
db_dic = {}
for table in div_test.find_all('table'):
for row in table.find_all('tr'):
td_list = row.find_all('td')
title = innerHtml(td_list[0].find('span'))
val = innerHtml(td_list[1]) if td_list[1].find(
'span') is None else innerHtml(td_list[1].find('span'))
if title in self.y_to_db_map:
db_dic[self.y_to_db_map[title]] = self.parse_numeric(val)
if db_dic:
db_dic["ticker"] = ticker
db_dic["exchange"] = exchange
col_names, vals = list(db_dic.keys()), list(db_dic.values())
where = db.create_conditional_string(col_names)
if db.exists("key_statistics", where, vals):
LOGGER.info("Skipping {} due to prior existence".format(url))
else:
col_names.append("update_date")
vals.append(self.today)
db.insert_into("key_statistics",
col_names,
vals,
multiple=False)
LOGGER.info("Done parsing {}".format(url))
else:
LOGGER.info("Skipping {}".format(url))
def handle_url(self, tickers, url, exchange):
"""
Fetches the url and inserts the data into the appropriate cols in the DB.
"""
LOGGER.info("Starting to add url: {} ...".format(url))
req = urllib.request.Request(url, headers=self.headers)
resp = urllib.request.urlopen(req)
csv_r = csv.reader(codecs.iterdecode(resp, 'utf-8'))
db_list = []
for row, ticker in zip(csv_r, tickers):
assert(len(row) == len(self.url_flags))
db_dic = {db_col: self.handle_csv_string(cell) for cell, db_col in zip(row, self.db_entries)}
onyahoo = any(v is not None for v in db_dic.values())
self.session.query(Listings).filter(Listings.exchange == exchange,
Listings.ticker == ticker
).update({Listings.onyahoo: onyahoo})
if not onyahoo: # not found, skip
LOGGER.error("Failed to find quote for {} skipping".format(ticker))
continue
db_dic["ticker"] = ticker
db_dic["exchange"] = exchange
exists = self.session.query(YahooKeyStatistics).filter_by(**db_dic).scalar() is not None
if exists:
LOGGER.info("Skipping {} due to prior existence".format(ticker))
continue
db_dic["update_date"] = self.today
# Annoyingly enough, sqlalchemy doesn't allow PostgreSQL bulk inserts
# when checking constraints, RIP performance
stmt = insert(YahooKeyStatistics).values(db_dic).on_conflict_do_nothing(
constraint = 'yahoo_key_statistics_pkey',
)
self.session.execute(stmt)
self.session.commit()
LOGGER.info("Done url.")
def get_key_stats(self, ticker, db_exchange="TSX"):
"""
This function get key statistics from
Morning Star.
"""
url = ("http://financials.morningstar.com/ajax/exportKR2CSV.html?t={}&"
"culture=en-CA®ion=CAN&order=asc&r={}").format(
ticker, randint(1, 500000))
req = urllib.request.Request(url, headers=self.headers)
resp = urllib.request.urlopen(req)
csv_r = csv.reader(codecs.iterdecode(resp, 'utf-8'))
on_morningstar = csv_r and resp.headers['content-length'] != '0'
if on_morningstar:
LOGGER.info("Getting key statistics for {}... ".format(ticker))
else:
LOGGER.info("Skipping {}".format(ticker))
return 1
return_dics = self.parse_csv(csv_r,
10,
self.special_key_titles,
self.column_key_map,
start_dic={
"ticker": ticker,
"exchange": db_exchange,
"update_date": self.today
})
for d in return_dics:
stmt = insert(MorningStarKeyStatistics).values(
d).on_conflict_do_update(constraint='ms_key_statistics_pkey',
set_=d)
self.session.execute(stmt)
self.session.commit()
LOGGER.info("Done")
return 0
def get_quotes(self):
"""
This function gets the tickers and various other random information
from the TSX website from a hardcoded file and inserts it into the database
"""
recent_date = self.db.select("MAX(updatedate)",
"listings",
fetch="one",
unroll=True)
if self.url.startswith("http"):
req = create_url_request(self.url)
self.url = urllib.request.urlopen(req)
sheet = pd.read_excel(self.url,
skiprows=5,
header=1,
keep_default_na=False)
sheet.fillna('', inplace=True)
sheet.rename(columns=self.cleanse_str, inplace=True)
file_date = self.find_date_in_list(list(sheet.columns.values))
if recent_date is None or (file_date > recent_date):
xlsx_dict = sheet.to_dict(orient="records")
recent_date = file_date
if self.cache_path:
self.write_cache(recent_date, sheet)
else:
LOGGER.info("Already up to date")
return
row_names = [
"updatedate",
"ticker",
"exchange",
"name",
"sector",
"osshares",
"dateoflisting",
"listingtype",
"volume",
"value",
]
all_excel_names = tuple(xlsx_dict[0].keys())
base_wanted_excel_names = [
"Root Ticker",
"Exchange",
"Name",
"Sector",
"O/S",
"Date of TSX Listing",
"Listing Type",
"Volume YTD",
"Value (C$)",
]
types = [
"str",
"str",
"str",
"str",
"int",
"date",
"str",
"int",
"int",
]
wanted_excel_names = []
for bxn in base_wanted_excel_names:
for xn in all_excel_names:
if xn.startswith(bxn):
wanted_excel_names.append(xn)
break
num_rows = len(wanted_excel_names)
table_name = "listings"
values = []
for row in xlsx_dict:
value_list = [recent_date]
for i in range(num_rows):
excel_name = wanted_excel_names[i]
val = row[excel_name]
if types[i] == "date":
val = datetime.strptime(str(val),
"%Y%m%d") # assume YYYYMMDD
value_list.append(val)
values.append(value_list)
self.db.insert_into(table_name, row_names, values)
def generate_and_save_feature_data(self, independent=True):
rc = ReturnCalculator()
ticker_names = sorted(get_ms_ticker_names(self.sess, "TSX"))
num_tickers = len(ticker_names)
train_data, train_targets = [], []
train_ticker_names = []
test_data, test_targets = [], []
test_ticker_names = []
imp = Imputer(missing_values='NaN', strategy='mean', axis=0)
for i, t in enumerate(ticker_names, 1):
LOGGER.info("[{:d}/{:d}] Working on {}...".format(
i, num_tickers, t))
dates = self.ff.ms_key_stats_date(self.sess, t)
if len(dates) < 1:
continue
date_gap = dates[1] - dates[0] if len(dates) > 2 else timedelta(
days=365)
last_date = dates[-1]
rows = self.ff.ms_key_stats_data(self.sess, t)
if not independent:
# Window sliding for time series
empty_row = tuple((None, )) * len(rows[0])
new_rows = []
for i in range(len(rows)):
first_part = rows[i - 1] if i > 0 else empty_row
second_part = rows[i]
new_rows.append(first_part + second_part)
rows = new_rows
# Add the start date to the list of dates
return_dates = [dates[0] - date_gap] + dates
returns = rc.calculate_return_between_dates(t, return_dates)
for row, date, ret in zip(rows, dates, returns):
if ret is None: # if return date are out of range
continue
if date == last_date:
test_data.append(row)
test_targets.append(ret)
test_ticker_names.append(t)
else:
train_data.append(row)
train_targets.append(ret)
train_ticker_names.append(t)
# Convert the python lists to numpy arrays and fill missing values
train_data = np.array(train_data, dtype=np.float)
imp = imp.fit(train_data)
train_ticker_names = np.array(train_ticker_names, dtype=np.str)
train_data = imp.transform(train_data)
train_targets = np.array(train_targets, dtype=np.float)
test_ticker_names = np.array(test_ticker_names, dtype=np.str)
test_data = imp.transform(np.array(test_data, dtype=np.float))
test_targets = np.array(test_targets, dtype=np.float)
if not os.path.exists(self.dir_path):
os.makedirs(self.dir_path)
LOGGER.info("Saving file at: {}".format(self.file_path))
np.savez(self.file_path,
train_data=train_data,
train_targets=train_targets,
train_ticker_names=train_ticker_names,
test_data=test_data,
test_targets=test_targets,
test_ticker_names=test_ticker_names)
def get_quotes(self):
"""
This function gets the tickers and various other random information
from the TSX website from a hardcoded file and inserts it into the database
"""
recent_date, = self.session.query(func.max(
Listings.updatedate)).first()
if self.url.startswith("http"):
req = create_url_request(self.url)
self.url = urllib.request.urlopen(req)
sheet = pd.read_excel(self.url,
skiprows=5,
header=1,
keep_default_na=False)
sheet.fillna('', inplace=True)
sheet.rename(columns=self.cleanse_str, inplace=True)
file_date = self.find_date_in_list(list(sheet.columns.values))
if recent_date is None or (file_date > recent_date):
xlsx_dict = sheet.to_dict(orient="records")
recent_date = file_date
else:
LOGGER.info("Already up to date")
return
row_names = [
"ticker",
"exchange",
"name",
"sector",
"osshares",
"dateoflisting",
"listingtype",
"volume",
"value",
]
all_excel_names = tuple(xlsx_dict[0].keys())
base_wanted_excel_names = [
"Root Ticker",
"Exchange",
"Name",
"Sector",
"O/S",
"Date of TSX Listing",
"Listing Type",
"Volume YTD",
"Value (C$)",
]
wanted_excel_names = []
for bxn in base_wanted_excel_names:
for xn in all_excel_names:
if xn.startswith(bxn):
wanted_excel_names.append(xn)
break
assert (len(base_wanted_excel_names) == len(wanted_excel_names) ==
len(row_names))
value_dics = []
for row in xlsx_dict:
value_dic = {"updatedate": recent_date}
for excel_name, row_name in zip(wanted_excel_names, row_names):
val = row[excel_name]
if row_name == "dateoflisting":
val = datetime.strptime(str(val),
"%Y%m%d") # assume YYYYMMDD
if val == '':
val = None
value_dic[row_name] = val
value_dics.append(value_dic)
self.session.execute(insert(Listings).values(value_dics))
self.session.commit()
def write_cache(self, date, sheet):
os.makedirs(self.cache_path, exist_ok=True)
json_name = date.strftime('TSX-%Y-%m-%d.json')
full_path = os.path.join(self.cache_path, json_name)
sheet.to_json(full_path, orient="records")
LOGGER.info("Wrote file to {}".format(full_path))