https://travis-ci.org/kaelzhang/stock-pandas.svg?branch=master https://codecov.io/gh/kaelzhang/stock-pandas/branch/master/graph/badge.svg https://img.shields.io/pypi/v/stock-pandas.svg https://img.shields.io/pypi/l/stock-pandas.svg

stock-pandas

stock-pandas inherits and extends pandas.DataFrame to support:

  • Stock Statistics
  • Stock Indicators, including:
    • Trend-following momentum indicators, such as MA, EMA, MACD, BBI
    • Dynamic support and resistance indicators, such as BOLL
    • Over-bought / over-sold indicators, such as KDJ, RSI
    • Other indicators, such as LLV, HHV
    • For more indicators, welcome to request a proposal, or fork and send me a pull request, or extend stock-pandas yourself. You might read the Advanced Sections below.
  • To cumulate kline data based on a given time frame, so that it could easily handle real-time data updates.

stock-pandas makes automatical trading much easier. stock-pandas requires Python >= 3.6 and Pandas >= 1.0.0(for now)

With the help of stock-pandas and mplfinance, we could easily draw something like:

_images/boll.png

The code example is available at here.

Install

For now, before installing stock-pandas in your environment

Have g++ compiler installed

# With yum, for CentOS, Amazon Linux, etc
yum install gcc-c++

# With apt-get, for Ubuntu
apt-get install g++

# For macOS, install XCode commandline tools
xcode-select --install

If you use docker with Dockerfile and use python image,

FROM python:3.8

...

The default python:3.8 image already contains g++, so we do not install g++ additionally.

Install stock-pandas

# Installing `stock-pandas` requires `numpy` to be installed first
pip install numpy

pip install stock-pandas

Be careful, you still need to install numpy explicitly even if numpy and stock-pandas both are contained in requirement.txt

numpy
stock-pandas
other-dependencies
...
pip install numpy

pip install -r requirement.txt

Usage

from stock_pandas import StockDataFrame

# or
import stock_pandas as spd

We also have some examples with annotations in the example directory, you could use JupyterLab or Jupyter notebook to play with them.

StockDataFrame

StockDataFrame inherits from pandas.DataFrame, so if you are familiar with pandas.DataFrame, you are already ready to use stock-pandas

import pandas as pd
stock = StockDataFrame(pd.read_csv('stock.csv'))

As we know, we could use [], which called pandas indexing (a.k.a. __getitem__ in python) to select out lower-dimensional slices. In addition to indexing with colname (column name of the DataFrame), we could also do indexing by directives.

stock[directive] # Gets a pandas.Series

stock[[directive0, directive1]] # Gets a StockDataFrame

We have an example to show the most basic indexing using [directive]

stock = StockDataFrame({
    'open' : ...,
    'high' : ...,
    'low'  : ...,
    'close': [5, 6, 7, 8, 9]
})

stock['ma:2']

# 0    NaN
# 1    5.5
# 2    6.5
# 3    7.5
# 4    8.5
# Name: ma:2,close, dtype: float64

Which prints the 2-period simple moving average on column "close".

Parameters

  • date_col Optional[str] = None If set, then the column named date_col will convert and set as DateTimeIndex of the data frame
  • to_datetime_kwargs dict = {} the keyworded arguments to be passed to pandas.to_datetime(). It only takes effect if date_col is specified.
  • time_frame str | TimeFrame | None = None time frame of the stock. For now, only the following time frames are supported:
    • '1m' or TimeFrame.M1
    • '3m' or TimeFrame.M3
    • '5m' or TimeFrame.M5
    • '15m' or TimeFrame.M15
    • '30m' or TimeFrame.M30
    • '1h' or TimeFrame.H1
    • '2h' or TimeFrame.H2
    • '4h' or TimeFrame.H4
    • '6h' or TimeFrame.H6
    • '8h' or TimeFrame.H8
    • '12h' or TimeFrame.H12

stock.exec(directive: str, create_column: bool=False) -> np.ndarray

Executes the given directive and returns a numpy ndarray according to the directive.

stock['ma:5'] # returns a Series

stock.exec('ma:5', create_column=True) # returns a numpy ndarray
# This will only calculate without creating a new column in the dataframe
stock.exec('ma:20')

The difference between stock[directive] and stock.exec(directive) is that

  • the former will create a new column for the result of directive as a cache for later use, while stock.exec(directive) does not unless we pass the parameter create_column as True
  • the former one accepts other pandas indexing targets, while stock.exec(directive) only accepts a valid stock-pandas directive string
  • the former one returns a pandas.Series or StockDataFrame object while the latter one returns an np.ndarray

stock.alias(alias: str, name: str) -> None

Defines column alias or directive alias

  • alias str the alias name
  • name str the name of an existing column or the directive string
# Some plot library such as `mplfinance` requires a column named capitalized `Open`,
# but it is ok, we could create an alias.
stock.alias('Open', 'open')

stock.alias('buy_point', 'kdj.j < 0')

stock.get_column(key: str) -> pd.Series

Directly gets the column value by key, returns a pandas Series.

If the given key is an alias name, it will return the value of corresponding original column.

If the column is not found, a KeyError will be raised.

stock = StockDataFrame({
    'open' : ...,
    'high' : ...,
    'low'  : ...,
    'close': [5, 6, 7, 8, 9]
})

stock.get_column('close')
# 0    5
# 1    6
# 2    7
# 3    8
# 4    9
# Name: close, dtype: float64
try:
    stock.get_column('Close')
except KeyError as e:
    print(e)

    # KeyError: column "Close" not found

stock.alias('Close', 'close')

stock.get_column('Close')
# The same as `stock.get_column('close')`

stock.append(other, *args, **kwargs) -> StockDataFrame

Appends rows of other to the end of caller, returning a new object.

This method has nearly the same hehavior of pandas.DataFrame.append(), but instead it returns an instance of StockDataFrame, and it applies date_col to the newly-appended row(s) if possible.

stock.directive_stringify(directive: str) -> str

Since 0.26.0

Gets the full name of the directive which is also the actual column name of the data frame

stock.directive_stringify('kdj.j')
# "kdj.j:9,3,3,50.0"

And also

from stock_pandas import

directive_stringify('kdj.j')
# "kdj.j:9,3,3,50.0"

Actually, directive_stringify does not rely on StockDataFrame instances.

stock.rolling_calc(size, on, apply, forward, fill) -> np.ndarray

Since 0.27.0

Applies a 1-D function along the given column or directive on

  • size int the size of the rolling window
  • on str | Directive along which the function should be applied
  • apply Callable[[np.ndarray], Any] the 1-D function to apply
  • forward? bool = False whether we should look backward (default value) to get each rolling window or not
  • fill? Any = np.nan the value used to fill where there are not enough items to form a rolling window
stock.rolling_calc(5, 'open', max)

# Whose return value equals to
stock['hhv:5,open'].to_numpy()

stock.cumulate() -> StockDataFrame

Cumulate the current data frame stock based on its time frame setting

StockDataFrame(one_minute_kline_data_frame, time_frame='5m').cumulate()

# And you will get a 5-minute kline data

see Cumulation and DatetimeIndex for details

stock.cum_append(other: DataFrame) -> StockDataFrame

Append other to the end of the current data frame stock and apply cumulation on them. And the following slice of code is equivalent to the above one:

StockDataFrame(time_frame='5m').cum_append(one_minute_kline_data_frame)

see Cumulation and DatetimeIndex for details

stock.fulfill() -> self

Since 1.2.0

Fulfill all stock indicator columns. By default, adding new rows to a StockDataFrame will not update stock indicators of the new row.

Stock indicators will only be updated when accessing the stock indicator column or calling stock.fulfill()

Check the test cases for details

directive_stringify(directive_str) -> str

since 0.30.0

Similar to stock.directive_stringify() but could be called without class initialization

from stock_pandas import directive_stringify

directive_stringify('boll')
# boll:21,close

Cumulation and DatetimeIndex

Suppose we have a csv file containing kline data of a stock in 1-minute time frame

csv = pd.read_csv(csv_path)

print(csv)
                   date   open   high    low  close    volume
0   2020-01-01 00:00:00  329.4  331.6  327.6  328.8  14202519
1   2020-01-01 00:01:00  330.0  332.0  328.0  331.0  13953191
2   2020-01-01 00:02:00  332.8  332.8  328.4  331.0  10339120
3   2020-01-01 00:03:00  332.0  334.2  330.2  331.0   9904468
4   2020-01-01 00:04:00  329.6  330.2  324.9  324.9  13947162
5   2020-01-01 00:04:00  329.6  330.2  324.8  324.8  13947163    <- There is an update of
                                                                    2020-01-01 00:04:00
...
16  2020-01-01 00:16:00  333.2  334.8  331.2  334.0  12428539
17  2020-01-01 00:17:00  333.0  333.6  326.8  333.6  15533405
18  2020-01-01 00:18:00  335.0  335.2  326.2  327.2  16655874
19  2020-01-01 00:19:00  327.0  327.2  322.0  323.0  15086985
Noted that duplicated records of a same timestamp will not be cumulated. The records except the latest one will be disgarded.
stock = StockDataFrame(
    csv,
    date_col='date',
    # Which is equivalent to `time_frame=TimeFrame.M5`
    time_frame='5m'
)

print(stock)
                      open   high    low  close    volume
2020-01-01 00:00:00  329.4  331.6  327.6  328.8  14202519
2020-01-01 00:01:00  330.0  332.0  328.0  331.0  13953191
2020-01-01 00:02:00  332.8  332.8  328.4  331.0  10339120
2020-01-01 00:03:00  332.0  334.2  330.2  331.0   9904468
2020-01-01 00:04:00  329.6  330.2  324.9  324.9  13947162
2020-01-01 00:04:00  329.6  330.2  324.8  324.8  13947162
...
2020-01-01 00:16:00  333.2  334.8  331.2  334.0  12428539
2020-01-01 00:17:00  333.0  333.6  326.8  333.6  15533405
2020-01-01 00:18:00  335.0  335.2  326.2  327.2  16655874
2020-01-01 00:19:00  327.0  327.2  322.0  323.0  15086985

You must have figured it out that the data frame now has DatetimeIndexes.

But it will not become a 15-minute kline data unless we cumulate it, and only cumulates new frames if you use stock.cum_append(them) to cumulate them.

stock_15m = stock.cumulate()

print(stock_15m)

Now we get a 15-minute kline

                      open   high    low  close      volume
2020-01-01 00:00:00  329.4  334.2  324.8  324.8  62346461.0
2020-01-01 00:05:00  325.0  327.8  316.2  322.0  82176419.0
2020-01-01 00:10:00  323.0  327.8  314.6  327.6  74409815.0
2020-01-01 00:15:00  330.0  335.2  322.0  323.0  82452902.0

For more details and about how to get full control of everything, check the online Google Colab notebook here.

Syntax of directive

directive := command | command operator expression
operator := '/' | '\' | '><' | '<' | '<=' | '==' | '>=' | '>'
expression := float | command

command := command_name | command_name : arguments
command_name := main_command_name | main_command_name.sub_command_name
main_command_name := alphabets
sub_command_name := alphabets

arguments := argument | argument , arguments
argument := empty_string | string | ( directive )

directive Example

Here lists several use cases of column names

# The middle band of bollinger bands
#   which is actually a 20-period (default) moving average
stock['boll']

# kdj j less than 0
# This returns a series of bool type
stock['kdj.j < 0']

# kdj %K cross up kdj %D
stock['kdj.k / kdj.d']

# 5-period simple moving average
stock['ma:5']

# 10-period simple moving average on open prices
stock['ma:10,open']

# Dataframe of 5-period, 10-period, 30-period ma
stock[[
    'ma:5',
    'ma:10',
    'ma:30'
]]

# Which means we use the default values of the first and the second parameters,
# and specify the third parameter
stock['macd:,,10']

# We must wrap a parameter which is a nested command or directive
stock['increase:(ma:20,close),3']

# stock-pandas has a powerful directive parser,
# so we could even write directives like this:
stock['''
repeat
    :
        (
            column:close > boll.upper
        ),
        5
''']

Built-in Commands of Indicators

Document syntax explanation:

  • param0 int which means param0 is a required parameter of type int.
  • param1? str='close' which means parameter param1 is optional with default value 'close'.

Actually, all parameters of a command are of string type, so the int here means an interger-like string.

ma, simple Moving Averages

ma:<period>,<column>

Gets the period-period simple moving average on column named column.

SMA is often confused between simple moving average and smoothed moving average.

So stock-pandas will use ma for simple moving average and smma for smoothed moving average.

  • period int (required)
  • column? enum<'open'|'high'|'low'|'close'>='close' Which column should the calculation based on. Defaults to 'close'
# which is equivalent to `stock['ma:5,close']`
stock['ma:5']

stock['ma:10,open']

ema, Exponential Moving Average

ema:<period>,<column>

Gets the Exponential Moving Average, also known as the Exponential Weighted Moving Average.

The arguments of this command is the same as ma.

macd, Moving Average Convergence Divergence

macd:<fast_period>,<slow_period>
macd.signal:<fast_period>,<slow_period>,<signal_period>
macd.histogram:<fast_period>,<slow_period>,<signal_period>
  • fast_period? int=12 fast period (short period). Defaults to 12.
  • slow_period? int=26 slow period (long period). Defaults to 26
  • signal_period? int=9 signal period. Defaults to 9
# macd
stock['macd']
stock['macd.dif']

# macd signal band, which is a shortcut for stock['macd.signal']
stock['macd.s']
stock['macd.signal']
stock['macd.dea']

# macd histogram band, which is equivalent to stock['macd.h']
stock['macd.histogram']
stock['macd.h']
stock['macd.macd']

boll, BOLLinger bands

boll:<period>,<column>
boll.upper:<period>,<times>,<column>
boll.lower:<period>,<times>,<column>
  • period? int=20
  • times? float=2.
  • column? str='close'
# boll
stock['boll']

# bollinger upper band, a shortcut for stock['boll.upper']
stock['boll.u']
stock['boll.upper']

# bollinger lower band, which is equivalent to stock['boll.l']
stock['boll.lower']
stock['boll.l']

rsv, Raw Stochastic Value

rsv:<period>

Calculates the raw stochastic value which is often used to calculate KDJ

kdj, a variety of stochastic oscillator

The variety of Stochastic Oscillator indicator created by Dr. George Lane, which follows the formula:

RSV = rsv(period_rsv)
%K = ema(RSV, period_k)
%D = ema(%K, period_d)
%J = 3 * %K - 2 * %D

And the ema here is the exponential weighted moving average with initial value as init_value.

PAY ATTENTION that the calculation forumla is different from wikipedia, but it is much popular and more widely used by the industry.

Directive Arguments:

kdj.k:<period_rsv>,<period_k>,<init_value>
kdj.d:<period_rsv>,<period_k>,<period_d>,<init_value>
kdj.j:<period_rsv>,<period_k>,<period_d>,<init_value>
  • period_rsv? int=9 The period for calculating RSV, which is used for K%
  • period_k? int=3 The period for calculating the EMA of RSV, which is used for K%
  • period_d? int=3 The period for calculating the EMA of K%, which is used for D%
  • init_value? float=50.0 The initial value for calculating ema. Trading softwares of different companies usually use different initial values each of which is usually 0.0, 50.0 or 100.0.
# The %D series of KDJ
stock['kdj.d']
# which is equivalent to
stock['kdj.d:9,3,3,50.0']

# The KDJ serieses of with parameters 9, 9, and 9
stock[['kdj.k:9,9', 'kdj.d:9,9,9', 'kdj.j:9,9,9']]

kdjc, another variety of stochastic oscillator

Unlike kdj, kdjc uses close value instead of high and low value to calculate rsv, which makes the indicator more sensitive than kdj

The arguments of kdjc are the same as kdj

rsi, Relative Strength Index

rsi:<period>

Calculates the N-period RSI (Relative Strength Index)

  • period int The period to calculate RSI. period should be an int which is larger than 1

bbi, Bull and Bear Index

bbi:<a>,<b>,<c>,<d>

Calculates indicator BBI (Bull and Bear Index) which is the average of ma:3, ma:6, ma:12, ma:24 by default

  • a? int=3
  • b? int=6
  • c? int=12
  • d? int=24

llv, Lowest of Low Values

llv:<period>,<column>

Gets the lowest of low prices in N periods

  • period int
  • column? str='low' Defaults to 'low'. But you could also get the lowest value of close prices
# The 10-period lowest prices
stock['llv:10']

# The 10-period lowest close prices
stock['llv:10,close']

hhv, Highest of High Values

hhv:<period>,<column>

Gets the highest of high prices in N periods. The arguments of hhv is the same as llv

Built-in Commands for Statistics

column

column:<name>

Just gets the series of a column. This command is designed to be used together with an operator to compare with another command or as a parameter of some statistics command.

  • name str the name of the column
# A bool-type series indicates whether the current price is higher than the upper bollinger band
stock['column:close > boll.upper']

increase

increase:<on>,<repeat>,<step>

Gets a bool-type series each item of which is True if the value of indicator on increases in the last period-period.

  • on str the command name of an indicator on what the calculation should be based
  • repeat? int=1
  • direction? 1 | -1 the direction of “increase”. -1 means decreasing

For example:

# Which means whether the `ma:20,close` line
# (a.k.a. 20-period simple moving average on column `'close'`)
# has been increasing repeatedly for 3 times (maybe 3 days)
stock['increase:(ma:20,close),3']

# If the close price has been decreasing repeatedly for 5 times (maybe 5 days)
stock['increase:close,5,-1']

style

style:<style>

Gets a bool-type series whether the candlestick of a period is of style style

  • style 'bullish' | 'bearish'
stock['style:bullish']

repeat

repeat:(<bool_directive>),<repeat>

The repeat command first gets the result of directive bool_directive, and detect whether True is repeated for repeat times

  • bool_directive str the directive which should returns a series of bools. PAY ATTENTION, that the directive should be wrapped with parantheses as a parameter.
  • repeat? int=1 which should be larger than 0
# Whether the bullish candlestick repeats for 3 periods (maybe 3 days)
stock['repeat:(style:bullish),3']

change

change:<on>,<period>

Percentage change between the current and a prior element on a certain series

Computes the percentage change from the immediately previous element by default. This is useful in comparing the percentage of change in a time series of prices.

  • on str the directive which returns a series of numbers, and the calculation will based on the series.
  • period? int=2 2 means we computes with the start value and the end value of a 2-period window.
# Percentage change of 20-period simple moving average
stock['change:(ma:20)']

Operators

left operator right

Operator: /

whether left crosses through right from the down side of right to the upper side which we call it as “cross up”.

Operator: \

whether left crosses down right.

# Which we call them "dead crosses"
stock['macd \\ macd.signal']

PAY ATTENTION, in the example above, we should escape the backslash, so we’ve got double backslashes '\\'

Operator: ><

whether left crosses right, either up or down.

Operator: < | <= | == | >= | >

For a certain record of the same time, whether the value of left is less than / less than or equal to / equal to / larger than or equal to / larger than the value of right.

Errors

from stock_pandas import (
    DirectiveSyntaxError,
    DirectiveValueError
)

DirectiveSyntaxError

Raises if there is a syntax error in the given directive.

stock['''
repeat
    :
        (
            column:close >> boll.upper
        ),
        5
''']

DirectiveSyntaxError might print some messages like this:

File "<string>", line 5, column 26

   repeat
       :
           (
>              column:close >> boll.upper
           ),
           5

                            ^
DirectiveSyntaxError: ">>" is an invalid operator

DirectiveValueError

Raises if

  • there is an unknown command name
  • something is wrong about the command arguments
  • etc.

Advanced Sections

How to extend stock-pandas and support more indicators,
This section is only recommended for contributors, but not for normal users, for that the definition of COMMANDS might change in the future.
from stock_pandas import COMMANDS, CommandPreset

To add a new indicator to stock-pandas, you could update the COMMANDS dict.

# The value of 'new-indicator' is a tuple
COMMANDS['new-indicator'] = (
    # The first item of the tuple is a CommandPreset instance
    CommandPreset(
        formula,
        args_setting
    ),
    sub_commands_dict,
    aliases_of_sub_commands
)

You could check here to figure out the typings for COMMANDS.

For a simplest indicator, such as simple moving average, you could check the implementation here.

formula(df, s, *args) -> Tuple[np.ndarray, int]

formula is a Callable[[StockDataFrame, slice, ...], [ndarray, int]].

  • df StockDataFrame the first argument of formula is the stock dataframe itself
  • s slice sometimes, we don’t need to calculate the whole dataframe but only part of it. This argument is passed into the formula by stock_pandas and should not be changed manually.
  • args Tuple[Any] the args of the indicator which is defined by args_setting

The Callable returns a tuple:

  • The first item of the tuple is the calculated result which is a numpy ndarray.
  • The second item of the tuple is the mininum periods to calculate the indicator.

args_setting: [(default, validate_and_coerce), …]

args_setting is a list of tuples.

  • The first item of each tuple is the default value of the parameter, and it could be None which implies it has no default value and is required.
  • The second item is a raisable callable which receives user input, validates it, coerces the type of the value and returns it. If the parameter has a default value and user don’t specified a value, the function will be skipped.

sub_commands_dict: Dict[str, CommandPreset]

A dict to declare sub commands, such as boll.upper.

sub_commands_dict could be None which indicates the indicator has no sub commands

aliases_of_sub_commands: Dict[str, Optional[str]]

Which declares the shortcut or alias of the commands, such as boll.u

dict(
    u='upper'
)

If the value of an alias is None, which means it is an alias of the main command, such as macd.dif

dict(
    dif=None
)

Development

First, install conda (recommended), and generate a conda environment for this project

conda create -n stock-pandas python=3.11

conda activate stock-pandas

# Install requirements
make install

# Build python ext (C++)
make build-ext

# Run unit tests
make test