# QTPyLib: Quantitative Trading Python Library # https://github.com/ranaroussi/qtpylib # # Copyright 2016-2018 Ran Aroussi # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # import warnings from datetime import datetime, timedelta import numpy as np import pandas as pd from pandas.core.base import PandasObject # ============================================= warnings.simplefilter(action="ignore", category=RuntimeWarning) # ============================================= def numpy_rolling_window(data, window): shape = data.shape[:-1] + (data.shape[-1] - window + 1, window) strides = data.strides + (data.strides[-1],) return np.lib.stride_tricks.as_strided(data, shape=shape, strides=strides) def numpy_rolling_series(func): def func_wrapper(data, window, as_source=False): series = data.values if isinstance(data, pd.Series) else data new_series = np.empty(len(series)) * np.nan calculated = func(series, window) new_series[-len(calculated) :] = calculated if as_source and isinstance(data, pd.Series): return pd.Series(index=data.index, data=new_series) return new_series return func_wrapper @numpy_rolling_series def numpy_rolling_mean(data, window, as_source=False): return np.mean(numpy_rolling_window(data, window), axis=-1) @numpy_rolling_series def numpy_rolling_std(data, window, as_source=False): return np.std(numpy_rolling_window(data, window), axis=-1, ddof=1) # --------------------------------------------- def session(df, start="17:00", end="16:00"): """remove previous globex day from df""" if df.empty: return df # get start/end/now as decimals int_start = list(map(int, start.split(":"))) int_start = (int_start[0] + int_start[1] - 1 / 100) - 0.0001 int_end = list(map(int, end.split(":"))) int_end = int_end[0] + int_end[1] / 100 int_now = df[-1:].index.hour[0] + (df[:1].index.minute[0]) / 100 # same-dat session? is_same_day = int_end > int_start # set pointers curr = prev = df[-1:].index[0].strftime("%Y-%m-%d") # globex/forex session if not is_same_day: prev = (datetime.strptime(curr, "%Y-%m-%d") - timedelta(1)).strftime("%Y-%m-%d") # slice if int_now >= int_start: df = df[df.index >= curr + " " + start] else: df = df[df.index >= prev + " " + start] return df.copy() # --------------------------------------------- def heikinashi(bars): bars = bars.copy() bars["ha_close"] = (bars["open"] + bars["high"] + bars["low"] + bars["close"]) / 4 # ha open bars.at[0, "ha_open"] = (bars.at[0, "open"] + bars.at[0, "close"]) / 2 for i in range(1, len(bars)): bars.at[i, "ha_open"] = (bars.at[i - 1, "ha_open"] + bars.at[i - 1, "ha_close"]) / 2 bars["ha_high"] = bars.loc[:, ["high", "ha_open", "ha_close"]].max(axis=1) bars["ha_low"] = bars.loc[:, ["low", "ha_open", "ha_close"]].min(axis=1) return pd.DataFrame( index=bars.index, data={ "open": bars["ha_open"], "high": bars["ha_high"], "low": bars["ha_low"], "close": bars["ha_close"], }, ) # --------------------------------------------- def tdi(series, rsi_lookback=13, rsi_smooth_len=2, rsi_signal_len=7, bb_lookback=34, bb_std=1.6185): rsi_data = rsi(series, rsi_lookback) rsi_smooth = sma(rsi_data, rsi_smooth_len) rsi_signal = sma(rsi_data, rsi_signal_len) bb_series = bollinger_bands(rsi_data, bb_lookback, bb_std) return pd.DataFrame( index=series.index, data={ "rsi": rsi_data, "rsi_signal": rsi_signal, "rsi_smooth": rsi_smooth, "rsi_bb_upper": bb_series["upper"], "rsi_bb_lower": bb_series["lower"], "rsi_bb_mid": bb_series["mid"], }, ) # --------------------------------------------- def awesome_oscillator(df, weighted=False, fast=5, slow=34): midprice = (df["high"] + df["low"]) / 2 if weighted: ao = (midprice.ewm(fast).mean() - midprice.ewm(slow).mean()).values else: ao = numpy_rolling_mean(midprice, fast) - numpy_rolling_mean(midprice, slow) return pd.Series(index=df.index, data=ao) # --------------------------------------------- def nans(length=1): mtx = np.empty(length) mtx[:] = np.nan return mtx # --------------------------------------------- def typical_price(bars): res = (bars["high"] + bars["low"] + bars["close"]) / 3.0 return pd.Series(index=bars.index, data=res) # --------------------------------------------- def mid_price(bars): res = (bars["high"] + bars["low"]) / 2.0 return pd.Series(index=bars.index, data=res) # --------------------------------------------- def ibs(bars): """Internal bar strength""" res = np.round((bars["close"] - bars["low"]) / (bars["high"] - bars["low"]), 2) return pd.Series(index=bars.index, data=res) # --------------------------------------------- def true_range(bars): return pd.DataFrame( { "hl": bars["high"] - bars["low"], "hc": abs(bars["high"] - bars["close"].shift(1)), "lc": abs(bars["low"] - bars["close"].shift(1)), } ).max(axis=1) # --------------------------------------------- def atr(bars, window=14, exp=False): tr = true_range(bars) if exp: res = rolling_weighted_mean(tr, window) else: res = rolling_mean(tr, window) return pd.Series(res) # --------------------------------------------- def crossed(series1, series2, direction=None): if isinstance(series1, np.ndarray): series1 = pd.Series(series1) if isinstance(series2, (float, int, np.ndarray, np.integer, np.floating)): series2 = pd.Series(index=series1.index, data=series2) if direction is None or direction == "above": above = pd.Series((series1 > series2) & (series1.shift(1) <= series2.shift(1))) if direction is None or direction == "below": below = pd.Series((series1 < series2) & (series1.shift(1) >= series2.shift(1))) if direction is None: return above | below return above if direction == "above" else below def crossed_above(series1, series2): return crossed(series1, series2, "above") def crossed_below(series1, series2): return crossed(series1, series2, "below") # --------------------------------------------- def rolling_std(series, window=200, min_periods=None): min_periods = window if min_periods is None else min_periods if min_periods == window and len(series) > window: return numpy_rolling_std(series, window, True) else: try: return series.rolling(window=window, min_periods=min_periods).std() except Exception as e: # noqa: F841 return pd.Series(series).rolling(window=window, min_periods=min_periods).std() # --------------------------------------------- def rolling_mean(series, window=200, min_periods=None): min_periods = window if min_periods is None else min_periods if min_periods == window and len(series) > window: return numpy_rolling_mean(series, window, True) else: try: return series.rolling(window=window, min_periods=min_periods).mean() except Exception as e: # noqa: F841 return pd.Series(series).rolling(window=window, min_periods=min_periods).mean() # --------------------------------------------- def rolling_min(series, window=14, min_periods=None): min_periods = window if min_periods is None else min_periods try: return series.rolling(window=window, min_periods=min_periods).min() except Exception as e: # noqa: F841 return pd.Series(series).rolling(window=window, min_periods=min_periods).min() # --------------------------------------------- def rolling_max(series, window=14, min_periods=None): min_periods = window if min_periods is None else min_periods try: return series.rolling(window=window, min_periods=min_periods).max() except Exception as e: # noqa: F841 return pd.Series(series).rolling(window=window, min_periods=min_periods).max() # --------------------------------------------- def rolling_weighted_mean(series, window=200, min_periods=None): min_periods = window if min_periods is None else min_periods try: return series.ewm(span=window, min_periods=min_periods).mean() except Exception as e: # noqa: F841 return pd.ewma(series, span=window, min_periods=min_periods) # --------------------------------------------- def hull_moving_average(series, window=200, min_periods=None): min_periods = window if min_periods is None else min_periods ma = (2 * rolling_weighted_mean(series, window / 2, min_periods)) - rolling_weighted_mean( series, window, min_periods ) return rolling_weighted_mean(ma, np.sqrt(window), min_periods) # --------------------------------------------- def sma(series, window=200, min_periods=None): return rolling_mean(series, window=window, min_periods=min_periods) # --------------------------------------------- def wma(series, window=200, min_periods=None): return rolling_weighted_mean(series, window=window, min_periods=min_periods) # --------------------------------------------- def hma(series, window=200, min_periods=None): return hull_moving_average(series, window=window, min_periods=min_periods) # --------------------------------------------- def vwap(bars): """ calculate vwap of entire time series (input can be pandas series or numpy array) bars are usually mid [ (h+l)/2 ] or typical [ (h+l+c)/3 ] """ raise ValueError( "using `qtpylib.vwap` facilitates lookahead bias. Please use " "`qtpylib.rolling_vwap` instead, which calculates vwap in a rolling manner." ) # typical = ((bars['high'] + bars['low'] + bars['close']) / 3).values # volume = bars['volume'].values # return pd.Series(index=bars.index, # data=np.cumsum(volume * typical) / np.cumsum(volume)) # --------------------------------------------- def rolling_vwap(bars, window=200, min_periods=None): """ calculate vwap using moving window (input can be pandas series or numpy array) bars are usually mid [ (h+l)/2 ] or typical [ (h+l+c)/3 ] """ min_periods = window if min_periods is None else min_periods typical = (bars["high"] + bars["low"] + bars["close"]) / 3 volume = bars["volume"] left = (volume * typical).rolling(window=window, min_periods=min_periods).sum() right = volume.rolling(window=window, min_periods=min_periods).sum() return ( pd.Series(index=bars.index, data=(left / right)) .replace([np.inf, -np.inf], float("NaN")) .ffill() ) # --------------------------------------------- def rsi(series, window=14): """ compute the n period relative strength indicator """ # 100-(100/relative_strength) deltas = np.diff(series) seed = deltas[: window + 1] # default values ups = seed[seed > 0].sum() / window downs = -seed[seed < 0].sum() / window rsival = np.zeros_like(series) rsival[:window] = 100.0 - 100.0 / (1.0 + ups / downs) # period values for i in range(window, len(series)): delta = deltas[i - 1] if delta > 0: upval = delta downval = 0 else: upval = 0 downval = -delta ups = (ups * (window - 1) + upval) / window downs = (downs * (window - 1.0) + downval) / window rsival[i] = 100.0 - 100.0 / (1.0 + ups / downs) # return rsival return pd.Series(index=series.index, data=rsival) # --------------------------------------------- def macd(series, fast=3, slow=10, smooth=16): """ compute the MACD (Moving Average Convergence/Divergence) using a fast and slow exponential moving avg' return value is emaslow, emafast, macd which are len(x) arrays """ macd_line = rolling_weighted_mean(series, window=fast) - rolling_weighted_mean( series, window=slow ) signal = rolling_weighted_mean(macd_line, window=smooth) histogram = macd_line - signal # return macd_line, signal, histogram return pd.DataFrame( index=series.index, data={"macd": macd_line.values, "signal": signal.values, "histogram": histogram.values}, ) # --------------------------------------------- def bollinger_bands(series, window=20, stds=2): ma = rolling_mean(series, window=window, min_periods=1) std = rolling_std(series, window=window, min_periods=1) upper = ma + std * stds lower = ma - std * stds return pd.DataFrame(index=series.index, data={"upper": upper, "mid": ma, "lower": lower}) # --------------------------------------------- def weighted_bollinger_bands(series, window=20, stds=2): ema = rolling_weighted_mean(series, window=window) std = rolling_std(series, window=window) upper = ema + std * stds lower = ema - std * stds return pd.DataFrame( index=series.index, data={"upper": upper.values, "mid": ema.values, "lower": lower.values} ) # --------------------------------------------- def returns(series): try: res = (series / series.shift(1) - 1).replace([np.inf, -np.inf], float("NaN")) except Exception as e: # noqa: F841 res = nans(len(series)) return pd.Series(index=series.index, data=res) # --------------------------------------------- def log_returns(series): try: res = np.log(series / series.shift(1)).replace([np.inf, -np.inf], float("NaN")) except Exception as e: # noqa: F841 res = nans(len(series)) return pd.Series(index=series.index, data=res) # --------------------------------------------- def implied_volatility(series, window=252): try: logret = np.log(series / series.shift(1)).replace([np.inf, -np.inf], float("NaN")) res = numpy_rolling_std(logret, window) * np.sqrt(window) except Exception as e: # noqa: F841 res = nans(len(series)) return pd.Series(index=series.index, data=res) # --------------------------------------------- def keltner_channel(bars, window=14, atrs=2): typical_mean = rolling_mean(typical_price(bars), window) atrval = atr(bars, window) * atrs upper = typical_mean + atrval lower = typical_mean - atrval return pd.DataFrame( index=bars.index, data={"upper": upper.values, "mid": typical_mean.values, "lower": lower.values}, ) # --------------------------------------------- def roc(series, window=14): """ compute rate of change """ res = (series - series.shift(window)) / series.shift(window) return pd.Series(index=series.index, data=res) # --------------------------------------------- def cci(series, window=14): """ compute commodity channel index """ price = typical_price(series) typical_mean = rolling_mean(price, window) res = (price - typical_mean) / (0.015 * np.std(typical_mean)) return pd.Series(index=series.index, data=res) # --------------------------------------------- def stoch(df, window=14, d=3, k=3, fast=False): """ compute the n period relative strength indicator http://excelta.blogspot.co.il/2013/09/stochastic-oscillator-technical.html """ my_df = pd.DataFrame(index=df.index) my_df["rolling_max"] = df["high"].rolling(window).max() my_df["rolling_min"] = df["low"].rolling(window).min() my_df["fast_k"] = ( 100 * (df["close"] - my_df["rolling_min"]) / (my_df["rolling_max"] - my_df["rolling_min"]) ) my_df["fast_d"] = my_df["fast_k"].rolling(d).mean() if fast: return my_df.loc[:, ["fast_k", "fast_d"]] my_df["slow_k"] = my_df["fast_k"].rolling(k).mean() my_df["slow_d"] = my_df["slow_k"].rolling(d).mean() return my_df.loc[:, ["slow_k", "slow_d"]] # --------------------------------------------- def zlma(series, window=20, min_periods=None, kind="ema"): """ John Ehlers' Zero lag (exponential) moving average https://en.wikipedia.org/wiki/Zero_lag_exponential_moving_average """ min_periods = window if min_periods is None else min_periods lag = (window - 1) // 2 series = 2 * series - series.shift(lag) if kind in ["ewm", "ema"]: return wma(series, lag, min_periods) elif kind == "hma": return hma(series, lag, min_periods) return sma(series, lag, min_periods) def zlema(series, window, min_periods=None): return zlma(series, window, min_periods, kind="ema") def zlsma(series, window, min_periods=None): return zlma(series, window, min_periods, kind="sma") def zlhma(series, window, min_periods=None): return zlma(series, window, min_periods, kind="hma") # --------------------------------------------- def zscore(bars, window=20, stds=1, col="close"): """get zscore of price""" std = numpy_rolling_std(bars[col], window) mean = numpy_rolling_mean(bars[col], window) return (bars[col] - mean) / (std * stds) # --------------------------------------------- def pvt(bars): """Price Volume Trend""" trend = ((bars["close"] - bars["close"].shift(1)) / bars["close"].shift(1)) * bars["volume"] return trend.cumsum() def chopiness(bars, window=14): atrsum = true_range(bars).rolling(window).sum() highs = bars["high"].rolling(window).max() lows = bars["low"].rolling(window).min() return 100 * np.log10(atrsum / (highs - lows)) / np.log10(window) # ============================================= PandasObject.session = session PandasObject.atr = atr PandasObject.bollinger_bands = bollinger_bands PandasObject.cci = cci PandasObject.crossed = crossed PandasObject.crossed_above = crossed_above PandasObject.crossed_below = crossed_below PandasObject.heikinashi = heikinashi PandasObject.hull_moving_average = hull_moving_average PandasObject.ibs = ibs PandasObject.implied_volatility = implied_volatility PandasObject.keltner_channel = keltner_channel PandasObject.log_returns = log_returns PandasObject.macd = macd PandasObject.returns = returns PandasObject.roc = roc PandasObject.rolling_max = rolling_max PandasObject.rolling_min = rolling_min PandasObject.rolling_mean = rolling_mean PandasObject.rolling_std = rolling_std PandasObject.rsi = rsi PandasObject.stoch = stoch PandasObject.zscore = zscore PandasObject.pvt = pvt PandasObject.chopiness = chopiness PandasObject.tdi = tdi PandasObject.true_range = true_range PandasObject.mid_price = mid_price PandasObject.typical_price = typical_price PandasObject.vwap = vwap PandasObject.rolling_vwap = rolling_vwap PandasObject.weighted_bollinger_bands = weighted_bollinger_bands PandasObject.rolling_weighted_mean = rolling_weighted_mean PandasObject.sma = sma PandasObject.wma = wma PandasObject.ema = wma PandasObject.hma = hma PandasObject.zlsma = zlsma PandasObject.zlwma = zlema PandasObject.zlema = zlema PandasObject.zlhma = zlhma PandasObject.zlma = zlma