def eval_impl(self):
self.eval_note = START_PENDING_EVAL_NOTE
period_length = 14
if len(self.data[
PriceIndexes.IND_PRICE_HIGH.value]) > period_length + 10:
min_adx = 7.5
max_adx = 45
neutral_adx = 25
adx = tulipy.adx(self.data[PriceIndexes.IND_PRICE_HIGH.value],
self.data[PriceIndexes.IND_PRICE_LOW.value],
self.data[PriceIndexes.IND_PRICE_CLOSE.value],
period_length)
instant_ema = tulipy.ema(
self.data[PriceIndexes.IND_PRICE_CLOSE.value], 2)
slow_ema = tulipy.ema(
self.data[PriceIndexes.IND_PRICE_CLOSE.value], 20)
adx = DataUtil.drop_nan(adx)
if len(adx):
current_adx = adx[-1]
current_slows_ema = slow_ema[-1]
current_instant_ema = instant_ema[-1]
multiplier = -1 if current_instant_ema < current_slows_ema else 1
# strong adx => strong trend
if current_adx > neutral_adx:
# if max adx already reached => when ADX forms a top and begins to turn down, you should look for a
# retracement that causes the price to move toward its 20-day exponential moving average (EMA).
adx_last_values = adx[-15:]
adx_last_value = adx_last_values[-1]
local_max_adx = adx_last_values.max()
# max already reached => trend will slow down
if adx_last_value < local_max_adx:
self.eval_note = multiplier * (
current_adx - neutral_adx) / (local_max_adx -
neutral_adx)
# max not reached => trend will continue, return chances to be max now
else:
crossing_indexes = TrendAnalysis.get_threshold_change_indexes(
adx, neutral_adx)
chances_to_be_max = \
TrendAnalysis.get_estimation_of_move_state_relatively_to_previous_moves_length(
crossing_indexes, adx) if len(crossing_indexes) > 2 \
else 0.75
proximity_to_max = min(1, current_adx / max_adx)
self.eval_note = multiplier * proximity_to_max * chances_to_be_max
# weak adx => change to come
else:
self.eval_note = multiplier * min(1, (
(neutral_adx - current_adx) / (neutral_adx - min_adx)))
async def evaluate(self, cryptocurrency, symbol, time_frame, close_candles,
high_candles, low_candles, candle):
self.eval_note = commons_constants.START_PENDING_EVAL_NOTE
if len(close_candles) >= self.minimal_data:
min_adx = 7.5
max_adx = 45
neutral_adx = 25
adx = tulipy.adx(high_candles, low_candles, close_candles,
self.period_length)
instant_ema = data_util.drop_nan(tulipy.ema(close_candles, 2))
slow_ema = data_util.drop_nan(tulipy.ema(close_candles, 20))
adx = data_util.drop_nan(adx)
if len(adx):
current_adx = adx[-1]
current_slows_ema = slow_ema[-1]
current_instant_ema = instant_ema[-1]
multiplier = -1 if current_instant_ema < current_slows_ema else 1
# strong adx => strong trend
if current_adx > neutral_adx:
# if max adx already reached => when ADX forms a top and begins to turn down, you should look for a
# retracement that causes the price to move toward its 20-day exponential moving average (EMA).
adx_last_values = adx[-15:]
adx_last_value = adx_last_values[-1]
local_max_adx = adx_last_values.max()
# max already reached => trend will slow down
if adx_last_value < local_max_adx:
self.eval_note = multiplier * (
current_adx - neutral_adx) / (local_max_adx -
neutral_adx)
# max not reached => trend will continue, return chances to be max now
else:
crossing_indexes = EvaluatorUtil.TrendAnalysis.get_threshold_change_indexes(
adx, neutral_adx)
chances_to_be_max = \
EvaluatorUtil.TrendAnalysis.get_estimation_of_move_state_relatively_to_previous_moves_length(
crossing_indexes, adx) if len(crossing_indexes) > 2 else 0.75
proximity_to_max = min(1, current_adx / max_adx)
self.eval_note = multiplier * proximity_to_max * chances_to_be_max
# weak adx => change to come
else:
self.eval_note = multiplier * min(1, (
(neutral_adx - current_adx) / (neutral_adx - min_adx)))
await self.evaluation_completed(
cryptocurrency,
symbol,
time_frame,
eval_time=evaluators_util.get_eval_time(full_candle=candle,
time_frame=time_frame))
def test_adx(data):
"""Spot check against ti seems accurate, slight differences (1e-02) with
Oanda display"""
arr_high = data["high"].to_numpy().astype(float)
arr_low = data["low"].to_numpy().astype(float)
arr_close = data["close"].to_numpy().astype(float)
ti_adx = np.append([np.nan for i in range(26)],
ti.adx(arr_high, arr_low, arr_close, period=14))
adx = indicator.Indicate(data, exp=6)\
.average_directional_movement()['adx']\
.astype(float)
assert np.allclose(ti_adx[-10:], adx[-10:], atol=1e-03, equal_nan=True)
def analyse(self):
close = self.bars[:, 0].copy(order='C').astype('float64')
high = self.bars[:, 3].copy(order='C').astype('float64')
low = self.bars[:, 4].copy(order='C').astype('float64')
if not (close.size + 2 - 2 * self.parameter > 0):
raise RuntimeError
my_result = ti.adx(high, low, close, self.parameter)
median = np.median(my_result)
if not np.isnan(median):
self.calc = float(median)
else:
self.logger.warning("Data causes nan. {}".format(close))
return super(AdxFilter, self).analyse()
def calculate(self, candles):
candles_len = len(candles)
if candles_len < self.period + 20: #make sure 20 more candles are available to work with
return 0
close_array = np.array(
[float(x['ohlc'].close) for x in candles[-(self.period + 20):]])
high_array = np.array(
[float(x['ohlc'].high) for x in candles[-(self.period + 20):]])
low_array = np.array(
[float(x['ohlc'].low) for x in candles[-(self.period + 20):]])
#calculate
adx = ti.adx(high_array, low_array, close_array, self.period)
return adx[-1]
def get_ADX(high, low, close, signal):
adx = np.zeros(len(high))
di = np.zeros([2,len(high)])
adx[signal*2-2:] = ti.adx(high, low, close, signal)
di[:,signal-1:] = np.array(ti.di(high, low, close, signal))
return [adx, di[0], di[1]]
def add_adx(df, period=10):
l = len(df.close.values)
df[f'adx_{period}'] = pad_left(ti.adx(
np.array(df.high),
np.array(df.low),
np.array(df.close), period), l)
