TAcharts 0.0.29
By: Carl Farterson
This repository provides technical tools to analyze OHLCV data, along with several
TA chart functionalities. These functions are optimized for speed and utilize numpy
vectorization over built-in pandas methods when possible.
Methods
Indicators With Chart Functionality
Bollinger(df=None, filename=None, interval=None, n=20, ndev=2)
: Bollinger BandsIchimoku(df=None, filename=None, interval=None)
: Ichimoku CloudRenko(df=None, filename=None, interval=None)
: Renko Chart
Indicators Without Chart Functionality
atr(high, low, close, n=2)
: average true range from candlestick datacmf(df, n=2)
: Chaikin Money Flow of an OHLCV datasetdouble_smooth(src, n_slow, n_fast)
: The smoothed value of two EMAsema(src, n=2)
: exponential moving average for a list of src
across n
periodsmacd(src, slow=25, fast=13)
: moving average convergence/divergence of src
mmo(src, n=2)
: Murrey Math oscillator of src
roc(src, n=2)
: rate of change of src
across n
periodsrolling(src, n=2, fn=None, axis=1)
: rolling sum
, max
, min
, mean
, or median
of src
across n
periodsrsi(src, n=2)
: relative strength index of src
across n
periodssdev(src, n=2)
: standard deviation across n periodssma(src, n=2)
: simple moving average of src
across n
periodstd_sequential(src, n=2)
: TD sequential of src
across n
periodstsi(src, slow=25, fast=13)
: true strength indicator
utils
area_between(line1, line2)
: find the area between line1 and line2crossover(x1, x2)
: find all instances of intersections between two linesdraw_candlesticks(ax, df)
: add candlestick visuals to a matplotlib chartfill_values(averages, interval, target_len)
: Fill missing values with evenly
spaced samples.
- Example: You're using 15-min candlestick data to find the 1-hour moving average
and want a value at every 15-min mark, and not every 1-hour mark.
group_candles(df, interval=4)
: combine candles so instead of needing a different
dataset for each time interval, you can form time intervals using more precise
data.
- Example: you have 15-min candlestick data but want to test a strategy based
on 1-hour candlestick data (
interval=4
).
intersection(a0, a1, b0, b1)
: find the intersection coordinates between vector
A and vector B
How it works
Create your DataFrame
from TAcharts.utils.ohlcv import OHLCV
df = OHLCV().btc
df.head()
| date | open | high | low | close | volume |
---|
0 | 2019-01-01 00:00:00 | 3699.95 | 3713.93 | 3697.00 | 3703.56 | 660.279771 |
1 | 2019-01-01 01:00:00 | 3703.63 | 3726.64 | 3703.34 | 3713.83 | 823.625491 |
2 | 2019-01-01 02:00:00 | 3714.19 | 3731.19 | 3707.00 | 3716.70 | 887.101362 |
3 | 2019-01-01 03:00:00 | 3716.98 | 3732.00 | 3696.14 | 3699.95 | 955.879034 |
4 | 2019-01-01 04:00:00 | 3699.96 | 3717.11 | 3698.00 | 3713.07 | 534.113945 |
Bollinger Bands
from TAcharts.indicators.bollinger import Bollinger
b = Bollinger(df)
b.build(n=20, ndev=2)
b.plot()
Ichimoku
from TAcharts.indicators.ichimoku import Ichimoku
i = Ichimoku(df)
i.build(20, 60, 120, 30)
i.plot()
Renko
from TAcharts.indicators.renko import Renko
r = Renko(df)
r.set_brick_size(auto=True, atr_interval=2)
r.build()
r.plot()
wrappers
-
@args_to_dtype(dtype)
: Convert all function arguments to a specific data type
from TAcharts.wrappers import args_to_dtype
@args_to_dtype(list)
def rsi(src, n=2):
pass
-
@pd_series_to_np_array
: Convert function arguments from pd.Series
to np.array
using pd.Series.values
. This wrapper is 10x quicker than using @args_to_dtype(np.array)
when working with Pandas series.
from TAcharts.wrappers import pd_series_to_np_array
@pd_series_to_np_array
def atr(high, low, close, n=14):
pass