def __init__(self):
'''Init function to initialize logger, variables for trading, and previous EMA for continuing data'''
for handler in logging.root.handlers[:]:
logging.root.removeHandler(handler)
logging.basicConfig(format='%(asctime)s: %(message)s', datefmt='%m/%d/%Y %H:%M:%S %p', level=logging.CRITICAL, handlers=[
logging.FileHandler("LOG.log"),
logging.StreamHandler()
])
self.last_completed_candle = requests.get("https://api-fxpractice.oanda.com/v3/instruments/{}/candles?count=2&price=M&granularity={}".format(INSTRUMENT, GRANULARITY), headers=HEADERS).json()['candles'][0]
self.in_long = False
self.in_short = False
self.begin_trading = False
self.fast = 0
self.slow = 0
# initialize prev EMAs with last 5000 candlesticks of data to use for live price EMA calculations
prev_data = requests.get("https://api-fxpractice.oanda.com/v3/instruments/{}/candles?count=5000&price=M&granularity={}".format(INSTRUMENT, GRANULARITY), headers=HEADERS).json()['candles'][:-2]
prev_data_df = pd.DataFrame([float(candle['mid']['c']) for candle in prev_data], columns=['close'])
self.prev_fast = btalib.ema(prev_data_df, period=FAST, _seed=prev_data_df['close'].iloc[0]).df['ema'].iloc[-1]
self.prev_slow = btalib.ema(prev_data_df, period=SLOW, _seed=prev_data_df['close'].iloc[0]).df['ema'].iloc[-1]
import plotly.express as px
import plotly.graph_objs as go
data = pd.read_csv(
'/Users/cosmos/GitHub/haasomeapitools/market_data/BINANCE|ADA|BNB.csv', parse_dates=True)
data.fillna(0)
print(data)
sma = btalib.sma(data, period=4)
# # sma = btalib.sma(df, period=15)
# print(sma.params.period)
# print(sma.p.period)
# print(sma.params['period'])
# print(list(sma.params))
# print(dict(sma.params))
# stochastic = btalib.stochastic(data,period = 5)
# print(stochastic.df)
first_close_price = data.Close.iloc[0]
ema = btalib.ema(data,_seed =first_close_price)
bbands = btalib.bbands(data,_ma=btalib.wma)
print(bbands.df)
newcols = []
# data2 = pd.merge(data,bbands.df,how='right',on='index')
# print(bbands)
data2 = data.join(bbands.df,how='outer')
data2.rename(columns={'mid': f"{bbands.p.period} mid",
'top': f"{bbands.p.devs} top", 'bot': f"{bbands.p.devs} bot"}.)
print(data2)
def price_history(symbol_pair, time_step):
client = client_test_net()
if time_step == '5m':
klines = client.get_historical_klines(symbol_pair,
Client.KLINE_INTERVAL_5MINUTE,
"2 day ago UTC")
elif time_step == '15m':
klines = client.get_historical_klines(symbol_pair,
Client.KLINE_INTERVAL_15MINUTE,
"6 day ago UTC")
elif time_step == '30m':
klines = client.get_historical_klines(symbol_pair,
Client.KLINE_INTERVAL_30MINUTE,
"12 day ago UTC")
else:
klines = client.get_historical_klines(symbol_pair,
Client.KLINE_INTERVAL_1HOUR,
"24 day ago UTC")
# Remove unnecessary data
for line in klines:
del line[6:]
# Generate csv file of RAW data
df1 = pd.DataFrame(
klines, columns=['date', 'open', 'high', 'low', 'close', 'volume'])
df1.set_index('date', inplace=True)
df1.to_csv('{}_history_{}.csv'.format(symbol_pair, time_step))
# Prepare to calculate indicator
df = pd.read_csv('{}_history_{}.csv'.format(symbol_pair, time_step),
index_col=0)
df.index = pd.to_datetime(df.index, unit='ms') + pd.DateOffset(hours=7)
# Average of high-low *note: is high-low-close better?
df['avg'] = (df['high'] + df['low']) / 2
# Calculate moving average using Pandas
df['50ma'] = df.close.rolling(50).mean()
df['200ma'] = df.close.rolling(200).mean()
# VWAP
df['vwap'] = 0
volume = df['volume']
high = df['high']
low = df['low']
close = df['close']
typical_price = (high + low + close) / 3
for i in range(len(df)):
if df.index.hour[i] == 23 and df.index.minute[i] == 0:
df.loc[i:, 'vwap'] = (
volume[i:] * typical_price[i:]).cumsum() / volume[i:].cumsum()
# EMA
ema200 = btalib.ema(df.close, period=200)
# MACD
macd = btalib.macd(df.close, pfast=12, pslow=26, psignal=9)
df = df.join([ema200.df, macd.df])
# Export RAW calculated file
df.to_csv('{}_indicator_{}.csv'.format(symbol_pair, time_step))
# Remove blank data from calculated file
df_plot = df[199:-1]
df_plot.to_csv('{}_indicator_cutted_{}.csv'.format(symbol_pair, time_step))
return df_plot
in_position = False
print(in_position)
# Read a csv file into a pandas dataframe
try:
df = pd.read_csv('data/ohlc/' + SYMBOL + '1.txt',
parse_dates=True,
index_col='Date')
except Exception as e:
logging.error(e)
time.sleep(.01)
df = pd.read_csv('data/ohlc/' + SYMBOL + '1.txt',
parse_dates=True,
index_col='Date')
ema = btalib.ema(df)
rsi = btalib.rsi(df)
df['ema'] = ema.df
df['rsi'] = rsi.df
macd = btalib.macd(df)
df['macd'] = macd.df['macd']
df['signal'] = macd.df['signal']
df['histogram'] = macd.df['histogram']
print(df)
try:
df.rsi.to_csv('data/technicals/' + SYMBOL + 'rsi1.csv',
header=True,
index=False)
def generate_history(symbol_pair,
time_step,
day_ago=None,
date_start=None,
date_end=None,
until_now=True):
day_ago = str(day_ago)
client = client_test_net()
if until_now:
klines = client.get_historical_klines(symbol_pair, time_step,
'{} day ago UTC'.format(day_ago))
else:
klines = client.get_historical_klines(symbol_pair, time_step,
date_start, date_end)
for line in klines:
del line[6:]
# Generate csv file of RAW data
df1 = pd.DataFrame(
klines, columns=['date', 'open', 'high', 'low', 'close', 'volume'])
df1.set_index('date', inplace=True)
df1.to_csv('{}_long_history_{}.csv'.format(symbol_pair, time_step))
# Prepare to calculate indicator
df = pd.read_csv('{}_long_history_{}.csv'.format(symbol_pair, time_step),
index_col=0)
df.index = pd.to_datetime(df.index, unit='ms') + pd.DateOffset(hours=7)
# Average of high-low *note: is close-open better?
df['avg'] = (df['high'] + df['low']) / 2
# Calculate moving average using Pandas
df['50ma'] = df.close.rolling(50).mean()
df['200ma'] = df.close.rolling(200).mean()
# VWAP
df['vwap'] = 0
volume = df['volume']
high = df['high']
low = df['low']
close = df['close']
typical_price = (high + low + close) / 3
for i in range(len(df)):
if df.index.hour[i] == 23 and df.index.minute[i] == 0:
df.loc[i:, 'vwap'] = (
volume[i:] * typical_price[i:]).cumsum() / volume[i:].cumsum()
# EMA
ema200 = btalib.ema(df.close, period=200)
# MACD
macd = btalib.macd(df.close, pfast=12, pslow=26, psignal=9)
df = df.join([ema200.df, macd.df])
# Export RAW calculated file
df.to_csv('{}_history_indicator_{}.csv'.format(symbol_pair, time_step))
# Remove blank data from calculated file
df_plot = df[199:-1]
df_plot.to_csv('{}_history_indicator_cutted_{}.csv'.format(
symbol_pair, time_step))
return df_plot
def tech_ind_features(data):
"""
Generate technical indicators
* :param data(pd.DataFrame): Raw data for processing
:return transformed_df(pd.DataFrame): Dataframe of features, sample balanced and normalised
"""
data['smoothed_close'] = data.close.rolling(9).mean().rolling(21).mean().shift(-15)
data['dx'] = np.diff(data['smoothed_close'], prepend=data['smoothed_close'][0])
data['dx_signal'] = pd.Series(data['dx']).rolling(9).mean()
data['ddx'] = np.diff(np.diff(data['smoothed_close']), prepend=data['smoothed_close'][0:2])
data['labels'] = np.zeros(len(data))
data['labels'].iloc[[(data.ddx < 0.1) & (data.dx <= 0) & (data.dx_signal > 0)]] = 1
data['labels'].iloc[[(data.ddx > -0.075) & (data.dx >= 0) & (data.dx_signal < 0)]] = 2
#Filter and drop all columns except close price, volume and date (for indexing)
relevant_cols = list(
compress(
data.columns,
[False if i in [1, 3, 4, 5, 6, len(data.columns)-1] else True for i in range(len(data.columns))]
)
)
data = data.drop(columns=relevant_cols).rename(columns={'open_date_time': 'date'})
data.set_index('date', inplace = True)
#Define relevant periods for lookback/feature engineering
periods = [
9, 14, 21,
30, 45, 60,
90, 100, 120
]
#Construct technical features for image synthesis
for period in periods:
data[f'ema_{period}'] = btalib.ema(
data.close,
period = period
).df['ema']
data[f'ema_{period}_dx'] = np.append(np.nan, np.diff(btalib.ema(
data.close,
period = period
).df['ema']))
data[f'rsi_{period}'] = btalib.rsi(
data.close,
period = period
).df['rsi']
data[f'cci_{period}'] = btalib.cci(
data.high,
data.low,
data.close,
period = period
).df['cci']
data[f'macd_{period}'] = btalib.macd(
data.close,
pfast = period,
pslow = period*2,
psignal = int(period/3)
).df['macd']
data[f'signal_{period}'] = btalib.macd(
data.close,
pfast = period,
pslow = period*2,
psignal = int(period/3)
).df['signal']
data[f'hist_{period}'] = btalib.macd(
data.close,
pfast = period,
pslow = period*2,
psignal = int(period/3)
).df['histogram']
data[f'volume_{period}'] = btalib.sma(
data.volume,
period = period
).df['sma']
data[f'change_{period}'] = data.close.pct_change(periods = period)
data = data.drop(data.query('labels == 0').sample(frac=.90).index)
data = data.replace([np.inf, -np.inf], np.nan).dropna()
data_trimmed = data.loc[:, 'ema_9':]
data_trimmed = pd.concat(
[data_trimmed,
data_trimmed.shift(1),
data_trimmed.shift(2)],
axis = 1
)
mm_scaler = MinMaxScaler(feature_range=(0, 1))
transformed_data = mm_scaler.fit_transform(data_trimmed[24:])
transformed_data = np.c_[
transformed_data,
pd.to_numeric(
data.labels[24:],
downcast = 'signed'
).to_list()
]
return transformed_data