mirror of
https://github.com/LCTT/TranslateProject.git
synced 2024-12-29 21:41:00 +08:00
Xingyu.Wang
GitHub
commit
560be880f4
@ -1,424 +0,0 @@
|
||||
[#]: collector: (lujun9972)
|
||||
[#]: translator: (wyxplus)
|
||||
[#]: reviewer: ( )
|
||||
[#]: publisher: ( )
|
||||
[#]: url: ( )
|
||||
[#]: subject: (How to automate your cryptocurrency trades with Python)
|
||||
[#]: via: (https://opensource.com/article/20/4/python-crypto-trading-bot)
|
||||
[#]: author: (Stephan Avenwedde https://opensource.com/users/hansic99)
|
||||
|
||||
How to automate your cryptocurrency trades with Python
|
||||
======
|
||||
In this tutorial, learn how to set up and use Pythonic, a graphical
|
||||
programming tool that makes it easy for users to create Python
|
||||
applications using ready-made function modules.
|
||||
![scientific calculator][1]
|
||||
|
||||
Unlike traditional stock exchanges like the New York Stock Exchange that have fixed trading hours, cryptocurrencies are traded 24/7, which makes it impossible for anyone to monitor the market on their own.
|
||||
|
||||
Often in the past, I had to deal with the following questions related to my crypto trading:
|
||||
|
||||
* What happened overnight?
|
||||
* Why are there no log entries?
|
||||
* Why was this order placed?
|
||||
* Why was no order placed?
|
||||
|
||||
|
||||
|
||||
The usual solution is to use a crypto trading bot that places orders for you when you are doing other things, like sleeping, being with your family, or enjoying your spare time. There are a lot of commercial solutions available, but I wanted an open source option, so I created the crypto-trading bot [Pythonic][2]. As [I wrote][3] in an introductory article last year, "Pythonic is a graphical programming tool that makes it easy for users to create Python applications using ready-made function modules." It originated as a cryptocurrency bot and has an extensive logging engine and well-tested, reusable parts such as schedulers and timers.
|
||||
|
||||
### Getting started
|
||||
|
||||
This hands-on tutorial teaches you how to get started with Pythonic for automated trading. It uses the example of trading [Tron][4] against [Bitcoin][5] on the [Binance][6] exchange platform. I choose these coins because of their volatility against each other, rather than any personal preference.
|
||||
|
||||
The bot will make decisions based on [exponential moving averages][7] (EMAs).
|
||||
|
||||
![TRX/BTC 1-hour candle chart][8]
|
||||
|
||||
TRX/BTC 1-hour candle chart
|
||||
|
||||
The EMA indicator is, in general, a weighted moving average that gives more weight to recent price data. Although a moving average may be a simple indicator, I've had good experiences using it.
|
||||
|
||||
The purple line in the chart above shows an EMA-25 indicator (meaning the last 25 values were taken into account).
|
||||
|
||||
The bot monitors the pitch between the current EMA-25 value (t0) and the previous EMA-25 value (t-1). If the pitch exceeds a certain value, it signals rising prices, and the bot will place a buy order. If the pitch falls below a certain value, the bot will place a sell order.
|
||||
|
||||
The pitch will be the main indicator for making decisions about trading. For this tutorial, it will be called the _trade factor_.
|
||||
|
||||
### Toolchain
|
||||
|
||||
The following tools are used in this tutorial:
|
||||
|
||||
* Binance expert trading view (visualizing data has been done by many others, so there's no need to reinvent the wheel by doing it yourself)
|
||||
* Jupyter Notebook for data-science tasks
|
||||
* Pythonic, which is the overall framework
|
||||
* PythonicDaemon as the pure runtime (console- and Linux-only)
|
||||
|
||||
|
||||
|
||||
### Data mining
|
||||
|
||||
For a crypto trading bot to make good decisions, it's essential to get open-high-low-close ([OHLC][9]) data for your asset in a reliable way. You can use Pythonic's built-in elements and extend them with your own logic.
|
||||
|
||||
The general workflow is:
|
||||
|
||||
1. Synchronize with Binance time
|
||||
2. Download OHLC data
|
||||
3. Load existing OHLC data from the file into memory
|
||||
4. Compare both datasets and extend the existing dataset with the newer rows
|
||||
|
||||
|
||||
|
||||
This workflow may be a bit overkill, but it makes this solution very robust against downtime and disconnections.
|
||||
|
||||
To begin, you need the **Binance OHLC Query** element and a **Basic Operation** element to execute your own code.
|
||||
|
||||
![Data-mining workflow][10]
|
||||
|
||||
Data-mining workflow
|
||||
|
||||
The OHLC query is set up to query the asset pair **TRXBTC** (Tron/Bitcoin) in one-hour intervals.
|
||||
|
||||
![Configuration of the OHLC query element][11]
|
||||
|
||||
Configuring the OHLC query element
|
||||
|
||||
The output of this element is a [Pandas DataFrame][12]. You can access the DataFrame with the **input** variable in the **Basic Operation** element. Here, the **Basic Operation** element is set up to use Vim as the default code editor.
|
||||
|
||||
![Basic Operation element set up to use Vim][13]
|
||||
|
||||
Basic Operation element set up to use Vim
|
||||
|
||||
Here is what the code looks like:
|
||||
|
||||
|
||||
```
|
||||
import pickle, pathlib, os
|
||||
import pandas as pd
|
||||
|
||||
outout = None
|
||||
|
||||
if isinstance(input, pd.DataFrame):
|
||||
file_name = 'TRXBTC_1h.bin'
|
||||
home_path = str(pathlib.Path.home())
|
||||
data_path = os.path.join(home_path, file_name)
|
||||
|
||||
try:
|
||||
df = pickle.load(open(data_path, 'rb'))
|
||||
n_row_cnt = df.shape[0]
|
||||
df = pd.concat([df,input], ignore_index=True).drop_duplicates(['close_time'])
|
||||
df.reset_index(drop=True, inplace=True)
|
||||
n_new_rows = df.shape[0] - n_row_cnt
|
||||
log_txt = '{}: {} new rows written'.format(file_name, n_new_rows)
|
||||
except:
|
||||
log_txt = 'File error - writing new one: {}'.format(e)
|
||||
df = input
|
||||
|
||||
pickle.dump(df, open(data_path, "wb" ))
|
||||
output = df
|
||||
```
|
||||
|
||||
First, check whether the input is the DataFrame type. Then look inside the user's home directory (**~/**) for a file named **TRXBTC_1h.bin**. If it is present, then open it, concatenate new rows (the code in the **try** section), and drop overlapping duplicates. If the file doesn't exist, trigger an _exception_ and execute the code in the **except** section, creating a new file.
|
||||
|
||||
As long as the checkbox **log output** is enabled, you can follow the logging with the command-line tool **tail**:
|
||||
|
||||
|
||||
```
|
||||
`$ tail -f ~/Pythonic_2020/Feb/log_2020_02_19.txt`
|
||||
```
|
||||
|
||||
For development purposes, skip the synchronization with Binance time and regular scheduling for now. This will be implemented below.
|
||||
|
||||
### Data preparation
|
||||
|
||||
The next step is to handle the evaluation logic in a separate grid; therefore, you have to pass over the DataFrame from Grid 1 to the first element of Grid 2 with the help of the **Return element**.
|
||||
|
||||
In Grid 2, extend the DataFrame by a column that contains the EMA values by passing the DataFrame through a **Basic Technical Analysis** element.
|
||||
|
||||
![Technical analysis workflow in Grid 2][14]
|
||||
|
||||
Technical analysis workflow in Grid 2
|
||||
|
||||
Configure the technical analysis element to calculate the EMAs over a period of 25 values.
|
||||
|
||||
![Configuration of the technical analysis element][15]
|
||||
|
||||
Configuring the technical analysis element
|
||||
|
||||
When you run the whole setup and activate the debug output of the **Technical Analysis** element, you will realize that the values of the EMA-25 column all seem to be the same.
|
||||
|
||||
![Missing decimal places in output][16]
|
||||
|
||||
Decimal places are missing in the output
|
||||
|
||||
This is because the EMA-25 values in the debug output include just six decimal places, even though the output retains the full precision of an 8-byte float value.
|
||||
|
||||
For further processing, add a **Basic Operation** element:
|
||||
|
||||
![Workflow in Grid 2][17]
|
||||
|
||||
Workflow in Grid 2
|
||||
|
||||
With the **Basic Operation** element, dump the DataFrame with the additional EMA-25 column so that it can be loaded into a Jupyter Notebook;
|
||||
|
||||
![Dump extended DataFrame to file][18]
|
||||
|
||||
Dump extended DataFrame to file
|
||||
|
||||
### Evaluation logic
|
||||
|
||||
Developing the evaluation logic inside Juypter Notebook enables you to access the code in a more direct way. To load the DataFrame, you need the following lines:
|
||||
|
||||
![Representation with all decimal places][19]
|
||||
|
||||
Representation with all decimal places
|
||||
|
||||
You can access the latest EMA-25 values by using [**iloc**][20] and the column name. This keeps all of the decimal places.
|
||||
|
||||
You already know how to get the latest value. The last line of the example above shows only the value. To copy the value to a separate variable, you have to access it with the **.at** method, as shown below.
|
||||
|
||||
You can also directly calculate the trade factor, which you will need in the next step.
|
||||
|
||||
![Buy/sell decision][21]
|
||||
|
||||
Buy/sell decision
|
||||
|
||||
### Determine the trading factor
|
||||
|
||||
As you can see in the code above, I chose 0.009 as the trade factor. But how do I know if 0.009 is a good trading factor for decisions? Actually, this factor is really bad, so instead, you can brute-force the best-performing trade factor.
|
||||
|
||||
Assume that you will buy or sell based on the closing price.
|
||||
|
||||
![Validation function][22]
|
||||
|
||||
Validation function
|
||||
|
||||
In this example, **buy_factor** and **sell_factor** are predefined. So extend the logic to brute-force the best performing values.
|
||||
|
||||
![Nested for loops for determining the buy and sell factor][23]
|
||||
|
||||
Nested _for_ loops for determining the buy and sell factor
|
||||
|
||||
This has 81 loops to process (9x9), which takes a couple of minutes on my machine (a Core i7 267QM).
|
||||
|
||||
![System utilization while brute forcing][24]
|
||||
|
||||
System utilization while brute-forcing
|
||||
|
||||
After each loop, it appends a tuple of **buy_factor**, **sell_factor**, and the resulting **profit** to the **trading_factors** list. Sort the list by profit in descending order.
|
||||
|
||||
![Sort profit with related trading factors in descending order][25]
|
||||
|
||||
Sort profit with related trading factors in descending order
|
||||
|
||||
When you print the list, you can see that 0.002 is the most promising factor.
|
||||
|
||||
![Sorted list of trading factors and profit][26]
|
||||
|
||||
Sorted list of trading factors and profit
|
||||
|
||||
When I wrote this in March 2020, the prices were not volatile enough to present more promising results. I got much better results in February, but even then, the best-performing trading factors were also around 0.002.
|
||||
|
||||
### Split the execution path
|
||||
|
||||
Start a new grid now to maintain clarity. Pass the DataFrame with the EMA-25 column from Grid 2 to element 0A of Grid 3 by using a **Return** element.
|
||||
|
||||
In Grid 3, add a **Basic Operation** element to execute the evaluation logic. Here is the code of that element:
|
||||
|
||||
![Implemented evaluation logic][27]
|
||||
|
||||
Implemented evaluation logic
|
||||
|
||||
The element outputs a **1** if you should buy or a **-1** if you should sell. An output of **0** means there's nothing to do right now. Use a **Branch** element to control the execution path.
|
||||
|
||||
![Branch element: Grid 3 Position 2A][28]
|
||||
|
||||
Branch element: Grid 3, Position 2A
|
||||
|
||||
Due to the fact that both **0** and **-1** are processed the same way, you need an additional Branch element on the right-most execution path to decide whether or not you should sell.
|
||||
|
||||
![Branch element: Grid 3 Position 3B][29]
|
||||
|
||||
Branch element: Grid 3, Position 3B
|
||||
|
||||
Grid 3 should now look like this:
|
||||
|
||||
![Workflow on Grid 3][30]
|
||||
|
||||
Workflow on Grid 3
|
||||
|
||||
### Execute orders
|
||||
|
||||
Since you cannot buy twice, you must keep a persistent variable between the cycles that indicates whether you have already bought.
|
||||
|
||||
You can do this with a **Stack element**. The Stack element is, as the name suggests, a representation of a file-based stack that can be filled with any Python data type.
|
||||
|
||||
You need to define that the stack contains only one Boolean element, which determines if you bought (**True**) or not (**False**). As a consequence, you have to preset the stack with one **False**. You can set this up, for example, in Grid 4 by simply passing a **False** to the stack.
|
||||
|
||||
![Forward a False-variable to the subsequent Stack element][31]
|
||||
|
||||
Forward a **False** variable to the subsequent Stack element
|
||||
|
||||
The Stack instances after the branch tree can be configured as follows:
|
||||
|
||||
![Configuration of the Stack element][32]
|
||||
|
||||
Configuring the Stack element
|
||||
|
||||
In the Stack element configuration, set **Do this with input** to **Nothing**. Otherwise, the Boolean value will be overwritten by a 1 or 0.
|
||||
|
||||
This configuration ensures that only one value is ever saved in the stack (**True** or **False**), and only one value can ever be read (for clarity).
|
||||
|
||||
Right after the Stack element, you need an additional **Branch** element to evaluate the stack value before you place the **Binance Order** elements.
|
||||
|
||||
![Evaluate the variable from the stack][33]
|
||||
|
||||
Evaluating the variable from the stack
|
||||
|
||||
Append the Binance Order element to the **True** path of the Branch element. The workflow on Grid 3 should now look like this:
|
||||
|
||||
![Workflow on Grid 3][34]
|
||||
|
||||
Workflow on Grid 3
|
||||
|
||||
The Binance Order element is configured as follows:
|
||||
|
||||
![Configuration of the Binance Order element][35]
|
||||
|
||||
Configuring the Binance Order element
|
||||
|
||||
You can generate the API and Secret keys on the Binance website under your account settings.
|
||||
|
||||
![Creating an API key in Binance][36]
|
||||
|
||||
Creating an API key in the Binance account settings
|
||||
|
||||
In this tutorial, every trade is executed as a market trade and has a volume of 10,000 TRX (~US$ 150 on March 2020). (For the purposes of this tutorial, I am demonstrating the overall process by using a Market Order. Because of that, I recommend using at least a Limit order.)
|
||||
|
||||
The subsequent element is not triggered if the order was not executed properly (e.g., a connection issue, insufficient funds, or incorrect currency pair). Therefore, you can assume that if the subsequent element is triggered, the order was placed.
|
||||
|
||||
Here is an example of output from a successful sell order for XMRBTC:
|
||||
|
||||
![Output of a successfully placed sell order][37]
|
||||
|
||||
Successful sell order output
|
||||
|
||||
This behavior makes subsequent steps more comfortable: You can always assume that as long the output is proper, the order was placed. Therefore, you can append a **Basic Operation** element that simply writes the output to **True** and writes this value on the stack to indicate whether the order was placed or not.
|
||||
|
||||
If something went wrong, you can find the details in the logging message (if logging is enabled).
|
||||
|
||||
![Logging output of Binance Order element][38]
|
||||
|
||||
Logging output from Binance Order element
|
||||
|
||||
### Schedule and sync
|
||||
|
||||
For regular scheduling and synchronization, prepend the entire workflow in Grid 1 with the **Binance Scheduler** element.
|
||||
|
||||
![Binance Scheduler at Grid 1, Position 1A][39]
|
||||
|
||||
Binance Scheduler at Grid 1, Position 1A
|
||||
|
||||
The Binance Scheduler element executes only once, so split the execution path on the end of Grid 1 and force it to re-synchronize itself by passing the output back to the Binance Scheduler element.
|
||||
|
||||
![Grid 1: Split execution path][40]
|
||||
|
||||
Grid 1: Split execution path
|
||||
|
||||
Element 5A points to Element 1A of Grid 2, and Element 5B points to Element 1A of Grid 1 (Binance Scheduler).
|
||||
|
||||
### Deploy
|
||||
|
||||
You can run the whole setup 24/7 on your local machine, or you could host it entirely on an inexpensive cloud system. For example, you can use a Linux/FreeBSD cloud system for about US$5 per month, but they usually don't provide a window system. If you want to take advantage of these low-cost clouds, you can use PythonicDaemon, which runs completely inside the terminal.
|
||||
|
||||
![PythonicDaemon console interface][41]
|
||||
|
||||
PythonicDaemon console
|
||||
|
||||
PythonicDaemon is part of the basic installation. To use it, save your complete workflow, transfer it to the remote running system (e.g., by Secure Copy [SCP]), and start PythonicDaemon with the workflow file as an argument:
|
||||
|
||||
|
||||
```
|
||||
`$ PythonicDaemon trading_bot_one`
|
||||
```
|
||||
|
||||
To automatically start PythonicDaemon at system startup, you can add an entry to the crontab:
|
||||
|
||||
|
||||
```
|
||||
`# crontab -e`
|
||||
```
|
||||
|
||||
![Crontab on Ubuntu Server][42]
|
||||
|
||||
Crontab on Ubuntu Server
|
||||
|
||||
### Next steps
|
||||
|
||||
As I wrote at the beginning, this tutorial is just a starting point into automated trading. Programming trading bots is approximately 10% programming and 90% testing. When it comes to letting your bot trade with your money, you will definitely think thrice about the code you program. So I advise you to keep your code as simple and easy to understand as you can.
|
||||
|
||||
If you want to continue developing your trading bot on your own, the next things to set up are:
|
||||
|
||||
* Automatic profit calculation (hopefully only positive!)
|
||||
* Calculation of the prices you want to buy for
|
||||
* Comparison with your order book (i.e., was the order filled completely?)
|
||||
|
||||
|
||||
|
||||
You can download the whole example on [GitHub][2].
|
||||
|
||||
--------------------------------------------------------------------------------
|
||||
|
||||
via: https://opensource.com/article/20/4/python-crypto-trading-bot
|
||||
|
||||
作者：[Stephan Avenwedde][a]
|
||||
选题：[lujun9972][b]
|
||||
译者：[译者ID](https://github.com/译者ID)
|
||||
校对：[校对者ID](https://github.com/校对者ID)
|
||||
|
||||
本文由 [LCTT](https://github.com/LCTT/TranslateProject) 原创编译，[Linux中国](https://linux.cn/) 荣誉推出
|
||||
|
||||
[a]: https://opensource.com/users/hansic99
|
||||
[b]: https://github.com/lujun9972
|
||||
[1]: https://opensource.com/sites/default/files/styles/image-full-size/public/lead-images/calculator_money_currency_financial_tool.jpg?itok=2QMa1y8c (scientific calculator)
|
||||
[2]: https://github.com/hANSIc99/Pythonic
|
||||
[3]: https://opensource.com/article/19/5/graphically-programming-pythonic
|
||||
[4]: https://tron.network/
|
||||
[5]: https://bitcoin.org/en/
|
||||
[6]: https://www.binance.com/
|
||||
[7]: https://www.investopedia.com/terms/e/ema.asp
|
||||
[8]: https://opensource.com/sites/default/files/uploads/1_ema-25.png (TRX/BTC 1-hour candle chart)
|
||||
[9]: https://en.wikipedia.org/wiki/Open-high-low-close_chart
|
||||
[10]: https://opensource.com/sites/default/files/uploads/2_data-mining-workflow.png (Data-mining workflow)
|
||||
[11]: https://opensource.com/sites/default/files/uploads/3_ohlc-query.png (Configuration of the OHLC query element)
|
||||
[12]: https://pandas.pydata.org/pandas-docs/stable/getting_started/dsintro.html#dataframe
|
||||
[13]: https://opensource.com/sites/default/files/uploads/4_edit-basic-operation.png (Basic Operation element set up to use Vim)
|
||||
[14]: https://opensource.com/sites/default/files/uploads/6_grid2-workflow.png (Technical analysis workflow in Grid 2)
|
||||
[15]: https://opensource.com/sites/default/files/uploads/7_technical-analysis-config.png (Configuration of the technical analysis element)
|
||||
[16]: https://opensource.com/sites/default/files/uploads/8_missing-decimals.png (Missing decimal places in output)
|
||||
[17]: https://opensource.com/sites/default/files/uploads/9_basic-operation-element.png (Workflow in Grid 2)
|
||||
[18]: https://opensource.com/sites/default/files/uploads/10_dump-extended-dataframe.png (Dump extended DataFrame to file)
|
||||
[19]: https://opensource.com/sites/default/files/uploads/11_load-dataframe-decimals.png (Representation with all decimal places)
|
||||
[20]: https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.DataFrame.iloc.html
|
||||
[21]: https://opensource.com/sites/default/files/uploads/12_trade-factor-decision.png (Buy/sell decision)
|
||||
[22]: https://opensource.com/sites/default/files/uploads/13_validation-function.png (Validation function)
|
||||
[23]: https://opensource.com/sites/default/files/uploads/14_brute-force-tf.png (Nested for loops for determining the buy and sell factor)
|
||||
[24]: https://opensource.com/sites/default/files/uploads/15_system-utilization.png (System utilization while brute forcing)
|
||||
[25]: https://opensource.com/sites/default/files/uploads/16_sort-profit.png (Sort profit with related trading factors in descending order)
|
||||
[26]: https://opensource.com/sites/default/files/uploads/17_sorted-trading-factors.png (Sorted list of trading factors and profit)
|
||||
[27]: https://opensource.com/sites/default/files/uploads/18_implemented-evaluation-logic.png (Implemented evaluation logic)
|
||||
[28]: https://opensource.com/sites/default/files/uploads/19_output.png (Branch element: Grid 3 Position 2A)
|
||||
[29]: https://opensource.com/sites/default/files/uploads/20_editbranch.png (Branch element: Grid 3 Position 3B)
|
||||
[30]: https://opensource.com/sites/default/files/uploads/21_grid3-workflow.png (Workflow on Grid 3)
|
||||
[31]: https://opensource.com/sites/default/files/uploads/22_pass-false-to-stack.png (Forward a False-variable to the subsequent Stack element)
|
||||
[32]: https://opensource.com/sites/default/files/uploads/23_stack-config.png (Configuration of the Stack element)
|
||||
[33]: https://opensource.com/sites/default/files/uploads/24_evaluate-stack-value.png (Evaluate the variable from the stack)
|
||||
[34]: https://opensource.com/sites/default/files/uploads/25_grid3-workflow.png (Workflow on Grid 3)
|
||||
[35]: https://opensource.com/sites/default/files/uploads/26_binance-order.png (Configuration of the Binance Order element)
|
||||
[36]: https://opensource.com/sites/default/files/uploads/27_api-key-binance.png (Creating an API key in Binance)
|
||||
[37]: https://opensource.com/sites/default/files/uploads/28_sell-order.png (Output of a successfully placed sell order)
|
||||
[38]: https://opensource.com/sites/default/files/uploads/29_binance-order-output.png (Logging output of Binance Order element)
|
||||
[39]: https://opensource.com/sites/default/files/uploads/30_binance-scheduler.png (Binance Scheduler at Grid 1, Position 1A)
|
||||
[40]: https://opensource.com/sites/default/files/uploads/31_split-execution-path.png (Grid 1: Split execution path)
|
||||
[41]: https://opensource.com/sites/default/files/uploads/32_pythonic-daemon.png (PythonicDaemon console interface)
|
||||
[42]: https://opensource.com/sites/default/files/uploads/33_crontab.png (Crontab on Ubuntu Server)
|
||||
@ -0,0 +1,429 @@
|
||||
[#]: collector: (lujun9972)
|
||||
[#]: translator: (wyxplus)
|
||||
[#]: reviewer: ( )
|
||||
[#]: publisher: ( )
|
||||
[#]: url: ( )
|
||||
[#]: subject: (How to automate your cryptocurrency trades with Python)
|
||||
[#]: via: (https://opensource.com/article/20/4/python-crypto-trading-bot)
|
||||
[#]: author: (Stephan Avenwedde https://opensource.com/users/hansic99)
|
||||
|
||||
|
||||
如何使用 Python 来自动交易加密货币
|
||||
======
|
||||
|
||||
在本教程中，教你如何设置和使用 Pythonic 来编程。它是一个图形化编程工具，用户可以很容易地使用现成的函数模块创建 Python 程序。
|
||||
|
||||
![scientific calculator][1]
|
||||
|
||||
然而，不像纽约证券交易所这样的传统证券交易所一样，有一段固定的交易时间。对于加密货币而言，则是 7×24 小时交易，任何人都无法独自盯着市场。
|
||||
|
||||
在以前，我经常思考与加密货币交易相关的问题：
|
||||
|
||||
- 一夜之间发生了什么？
|
||||
- 为什么没有日志记录？
|
||||
- 为什么下单？
|
||||
- 为什么不下单？
|
||||
|
||||
通常的解决手段是当在你做其他事情时，例如睡觉、与家人在一起或享受空闲时光，使用加密交易机器人代替你下单。虽然有很多商业解决方案可用，但是我选择开源的解决方案，因此我编写了加密交易机器人 [Pythonic][2]。 正如去年 [我写过的文章][3] 一样，“ Pythonic 是一种图形化编程工具，它让用户可以轻松使用现成的功能模块来创建Python应用程序。” 最初它是作为加密货币机器人使用，并具有可扩展的日志记录引擎以及经过精心测试的可重用部件，例如调度器和计时器。
|
||||
|
||||
### 开始
|
||||
|
||||
本教程将教你如何开始使用 Pythonic 进行自动交易。我选择 [币安][6]Binance<ruby>[币安][6]<rt>Binance</rt></ruby> 交易所的 [波场][4]Tron<ruby>[波场][4]<rt>Tron</rt></ruby> 与 [比特币][3]Bitcoin<ruby>[比特币][3]<rt>Bitcoin</rt></ruby>
|
||||
|
||||
交易对为例。我之所以选择这些加密货币，是因为它们彼此之间的波动性大，而不是出于个人喜好。
|
||||
|
||||
机器人将根据 [指数移动平均][7] （EMAs）来做出决策。
|
||||
|
||||
![TRX/BTC 1-hour candle chart][8]
|
||||
|
||||
TRX/BTC 1 小时 K 线图
|
||||
|
||||
EMA 指标通常是指加权移动平均线，可以对近期价格数据赋予更多权重。尽管移动平均线可能只是一个简单的指标，但我能熟练使用它。
|
||||
|
||||
上图中的紫色线显示了 EMA-25 指标（这表示要考虑最近的 25 个值）。
|
||||
|
||||
机器人监视当前的 EMA-25 值（t0）和前一个 EMA-25 值（t-1）之间的差距。如果差值超过某个值，则表示价格上涨，机器人将下达购买订单。如果差值低于某个值，则机器人将下达卖单。
|
||||
|
||||
差值将是做出交易决策的主要指标。在本教程中，它称为交易参数。
|
||||
|
||||
### 工具链
|
||||
|
||||
|
||||
|
||||
将在本教程使用如下工具：
|
||||
|
||||
- 币安专业交易视图（已经有其他人做了数据可视化，所以不需要重复造轮子）
|
||||
- Jupyter Notebook：用于数据科学任务
|
||||
- Pythonic：作为整体框架
|
||||
- PythonicDaemon ：作为终端运行（仅适用于控制台和 Linux）
|
||||
|
||||
|
||||
|
||||
### 数据挖掘
|
||||
|
||||
为了使加密货币交易机器人尽可能能做出正确的决定，以可靠的方式获取资产的美国线（[OHLC][9]）数据是至关重要。你可以使用 Pythonic 的内置元素，还可以根据自己逻辑来对其进行扩展。
|
||||
|
||||
一般的工作流程：
|
||||
|
||||
1. 与币安时间同步
|
||||
2. 下载 OHLC 数据
|
||||
3. 从文件中把 OHLC 数据加载到内存
|
||||
4. 比较数据集并扩展更新数据集
|
||||
|
||||
|
||||
|
||||
这个工作流程可能有点夸张，但是它能使得程序更加健壮，甚至在停机和断开连接时，也能平稳运行。
|
||||
|
||||
一开始，你需要 **币安 OHLC 查询**Binance OHLC Query<ruby>**币安 OHLC 查询**<rt>Binance OHLC Query</rt></ruby> 元素和一个 **基础操作**Basic Operation<ruby>**基础操作**<rt>Basic Operation</rt></ruby> 元素来执行你的代码。
|
||||
|
||||
![Data-mining workflow][10]
|
||||
|
||||
数据挖掘工作流程
|
||||
|
||||
OHLC 查询设置为每隔一小时查询一次 **TRXBTC** 资产对（波场/比特币）。
|
||||
|
||||
![Configuration of the OHLC query element][11]
|
||||
|
||||
配置 OHLC 查询元素
|
||||
|
||||
其中输出的元素是 [Pandas DataFrame][12]。你可以在 **基础操作** 元素中使用 **输入**input<ruby>**输入**<rt>input</rt></ruby> 变量来访问 DataFrame。其中，将 Vim 设置为 **基础操作** 元素的默认代码编辑器。
|
||||
|
||||
![Basic Operation element set up to use Vim][13]
|
||||
|
||||
使用 Vim 编辑基础操作元素
|
||||
|
||||
具体代码如下：
|
||||
|
||||
|
||||
```
|
||||
import pickle, pathlib, os
|
||||
import pandas as pd
|
||||
|
||||
outout = None
|
||||
|
||||
if isinstance(input, pd.DataFrame):
|
||||
file_name = 'TRXBTC_1h.bin'
|
||||
home_path = str(pathlib.Path.home())
|
||||
data_path = os.path.join(home_path, file_name)
|
||||
|
||||
try:
|
||||
df = pickle.load(open(data_path, 'rb'))
|
||||
n_row_cnt = df.shape[0]
|
||||
df = pd.concat([df,input], ignore_index=True).drop_duplicates(['close_time'])
|
||||
df.reset_index(drop=True, inplace=True)
|
||||
n_new_rows = df.shape[0] - n_row_cnt
|
||||
log_txt = '{}: {} new rows written'.format(file_name, n_new_rows)
|
||||
except:
|
||||
log_txt = 'File error - writing new one: {}'.format(e)
|
||||
df = input
|
||||
|
||||
pickle.dump(df, open(data_path, "wb" ))
|
||||
output = df
|
||||
```
|
||||
|
||||
首先，检查输入是否为 DataFrame 元素。然后在用户的家目录（**〜/ **）中查找名为 **TRXBTC_1h.bin** 的文件。如果存在，则将其打开，执行新代码段（**try** 部分中的代码），并删除重复项。如果文件不存在，则触发异常并执行 **except** 部分中的代码，创建一个新文件。
|
||||
|
||||
只要启用了复选框 **日志输出**log output<ruby>**日志输出**<rt>log output</rt></ruby>，你就可以使用命令行工具 **tail** 查看日志记录：
|
||||
|
||||
|
||||
```
|
||||
`$ tail -f ~/Pythonic_2020/Feb/log_2020_02_19.txt`
|
||||
```
|
||||
|
||||
出于开发目的，现在跳过与币安时间的同步和计划执行，这将在下面实现。
|
||||
|
||||
### 准备数据
|
||||
|
||||
下一步是在单独的 网格Grid<ruby>网格<rt>Grid</rt></ruby> 中处理评估逻辑。因此，你必须借助 **返回元素**Return element<ruby>**返回元素**<rt>Return element</rt></ruby> 将 DataFrame 从网格 1 传递到网格 2 的第一个元素。
|
||||
|
||||
在网格 2 中，通过使 DataFrame 通过 **基础技术分析**Basic Technical Analysis<ruby>**基础技术分析**<rt>Basic Technical Analysis</rt></ruby> 元素，将 DataFrame 扩展包含 EMA 值的一列。
|
||||
|
||||
![Technical analysis workflow in Grid 2][14]
|
||||
|
||||
在网格 2 中技术分析工作流程
|
||||
|
||||
配置技术分析元素以计算 25 个值的 EMAs。
|
||||
|
||||
![Configuration of the technical analysis element][15]
|
||||
|
||||
配置技术分析元素
|
||||
|
||||
当你运行整个程序并开启 **技术分析**Technical Analysis<ruby>**技术分析**<rt>Technical Analysis</rt></ruby> 元素的调试输出时，你将发现 EMA-25 列的值似乎都相同。
|
||||
|
||||
![Missing decimal places in output][16]
|
||||
|
||||
输出中精度不够
|
||||
|
||||
这是因为调试输出中的 EMA-25 值仅包含六位小数，即使输出保留了 8 个字节完整精度的浮点值。
|
||||
|
||||
为了能进行进一步处理，请添加 **基础操作** 元素：
|
||||
|
||||
![Workflow in Grid 2][17]
|
||||
|
||||
网格 2 中的工作流程
|
||||
|
||||
使用 **基础操作** 元素，将 DataFrame 与添加的 EMA-25 列一起转储，以便可以将其加载到 Jupyter Notebook中；
|
||||
|
||||
![Dump extended DataFrame to file][18]
|
||||
|
||||
将扩展后的 DataFrame 存储到文件中
|
||||
|
||||
### 评估策略
|
||||
|
||||
在 Juypter Notebook 中开发评估策略，让你可以更直接地访问代码。要加载 DataFrame，你需要使用如下代码：
|
||||
|
||||
![Representation with all decimal places][19]
|
||||
|
||||
用全部小数位表示
|
||||
|
||||
你可以使用 [**iloc**][20] 和列名来访问最新的 EMA-25 值，并且会保留所有小数位。
|
||||
|
||||
你已经知道如何来获得最新的数据。上面示例的最后一行仅显示该值。为了能将该值拷贝到不同的变量中，你必须使用如下图所示的 **.at** 方法方能成功。
|
||||
|
||||
你也可以直接计算出你下一步所需的交易参数。
|
||||
|
||||
![Buy/sell decision][21]
|
||||
|
||||
买卖决策
|
||||
|
||||
### 确定交易参数
|
||||
|
||||
如上面代码所示，我选择 0.009 作为交易参数。但是我怎么知道 0.009 是决定交易的一个好参数呢？ 实际上，这个参数确实很糟糕，因此，你可以直接计算出表现最佳的交易参数。
|
||||
|
||||
假设你将根据收盘价进行买卖。
|
||||
|
||||
![Validation function][22]
|
||||
|
||||
回测功能
|
||||
|
||||
在此示例中，**buy_factor** 和 **sell_factor** 是预先定义好的。因此，发散思维用直接计算出表现最佳的参数。
|
||||
|
||||
![Nested for loops for determining the buy and sell factor][23]
|
||||
|
||||
嵌套的 _for_ 循环，用于确定购买和出售的参数
|
||||
|
||||
这要跑 81 个循环（9x9），在我的机器（Core i7 267QM）上花费了几分钟。
|
||||
|
||||
![System utilization while brute forcing][24]
|
||||
|
||||
在暴力运算时系统的利用率
|
||||
|
||||
在每个循环之后，它将 **buy_factor**，**sell_factor** 元组和生成的 **利润**profit<ruby>**利润**<rt>profit</rt></ruby> 元组追加到 **trading_factors** 列表中。按利润降序对列表进行排序。
|
||||
|
||||
![Sort profit with related trading factors in descending order][25]
|
||||
|
||||
将利润与相关的交易参数按降序排序
|
||||
|
||||
当你打印出列表时，你会看到 0.002 是最好的参数。
|
||||
|
||||
![Sorted list of trading factors and profit][26]
|
||||
|
||||
交易要素和收益的有序列表
|
||||
|
||||
当我在 2020 年 3 月写下这篇文章时，价格的波动还不足以呈现出更理想的结果。我在 2 月份得到了更好的结果，但即使在那个时候，表现最好的交易参数也在 0.002 左右。
|
||||
|
||||
### 分割执行路径
|
||||
|
||||
现在开始新建一个网格以保持逻辑清晰。使用 **返回** 元素将带有 EMA-25 列的 DataFrame 从网格 2 传递到网格 3 的 0A 元素。
|
||||
|
||||
在网格 3 中，添加 **基础操作** 元素以执行评估逻辑。这是该元素中的代码：
|
||||
|
||||
![Implemented evaluation logic][27]
|
||||
|
||||
实现评估策略
|
||||
|
||||
如果输出 **1** 表示你应该购买，如果输出 **2** 则表示你应该卖出。 输出 **0** 表示现在无需操作。使用 **分支**Branch<ruby>**分支**<rt>Branch</rt></ruby> 元素来控制执行路径。
|
||||
|
||||
![Branch element: Grid 3 Position 2A][28]
|
||||
|
||||
Branch 元素：网格 3，2A 位置
|
||||
|
||||
|
||||
|
||||
因为 **0** 和 **-1** 的处理流程一样，所以你需要在最右边添加一个分支元素来判断你是否应该卖出。
|
||||
|
||||
![Branch element: Grid 3 Position 3B][29]
|
||||
|
||||
分支元素：网格 3，3B 位置
|
||||
|
||||
网格 3 应该现在如下图所示：
|
||||
|
||||
![Workflow on Grid 3][30]
|
||||
|
||||
网格 3 的工作流程
|
||||
|
||||
### 下单
|
||||
|
||||
由于无需在一个周期中购买两次，因此必须在周期之间保留一个持久变量，以指示你是否已经购买。
|
||||
|
||||
你可以利用 **栈**Stack<ruby>**栈**<rt>Stack</rt></ruby> 元素来实现。顾名思义，栈元素表示可以用任何 Python 数据类型来放入的基于文件的栈。
|
||||
|
||||
你需要定义栈仅包含一个布尔类型，该布尔类型决定是否购买了（**True**）或（**False**）。因此，你必须使用 **False** 来初始化栈。例如，你可以在网格 4 中简单地通过将 **False** 传递给栈来进行设置。![Forward a False-variable to the subsequent Stack element][31]
|
||||
|
||||
将 **False** 变量传输到后续的栈元素中
|
||||
|
||||
在分支树后的栈实例可以进行如下配置：
|
||||
|
||||
![Configuration of the Stack element][32]
|
||||
|
||||
设置栈元素
|
||||
|
||||
在栈元素设置中，将 **Do this with input** 设置成 **Nothing**。否则，布尔值将被 1 或 0 覆盖。
|
||||
|
||||
该设置确保仅将一个值保存于栈中（**True** 或 **False**），并且只能读取一个值（为了清楚起见）。
|
||||
|
||||
在栈元素之后，你需要另外一个 **分支** 元素来判断栈的值，然后再放置 **币安订单**Binance Order<ruby>**币安订单**<rt>Binance Order</rt></ruby> 元素。
|
||||
|
||||
![Evaluate the variable from the stack][33]
|
||||
|
||||
判断栈中的变量
|
||||
|
||||
将币安订单元素添加到分支元素的 **True** 路径。网格 3 上的工作流现在应如下所示：
|
||||
|
||||
![Workflow on Grid 3][34]
|
||||
|
||||
网格 3 的工作流程
|
||||
|
||||
币安订单元素应如下配置：
|
||||
|
||||
![Configuration of the Binance Order element][35]
|
||||
|
||||
编辑币安订单元素
|
||||
|
||||
你可以在币安网站上的帐户设置中生成 API 和密钥。
|
||||
|
||||
![Creating an API key in Binance][36]
|
||||
|
||||
在币安账户设置中创建一个 API key
|
||||
|
||||
在本文中，每笔交易都是作为市价交易执行的，交易量为10,000 TRX（2020 年 3 月约为 150 美元）（出于教学的目的，我通过使用市价下单来演示整个过程。因此，我建议至少使用限价下单。）
|
||||
|
||||
如果未正确执行下单（例如，网络问题、资金不足或货币对不正确），则不会触发后续元素。因此，你可以假定如果触发了后续元素，则表示该订单已下达。
|
||||
|
||||
这是一个成功的 XMRBTC 卖单的输出示例：
|
||||
|
||||
![Output of a successfully placed sell order][37]
|
||||
|
||||
成功卖单的输出
|
||||
|
||||
该行为使后续步骤更加简单：你可以始终假设只要成功输出，就表示订单成功。因此，你可以添加一个 **基础操作** 元素，该元素将简单地输出 **True** 并将此值放入栈中以表示是否下单。
|
||||
|
||||
如果出现错误的话，你可以在日志信息中查看具体细节（如果启用日志功能）。
|
||||
|
||||
![Logging output of Binance Order element][38]
|
||||
|
||||
币安订单元素中的输出日志信息
|
||||
|
||||
### 调度和同步
|
||||
|
||||
对于日程调度和同步，请在网格 1 中将整个工作流程置于 **币安调度器**Binance Scheduler<ruby>**币安调度器**<rt>Binance Scheduler</rt></ruby> 元素的前面。
|
||||
|
||||
![Binance Scheduler at Grid 1, Position 1A][39]
|
||||
|
||||
在网格 1，1A 位置的币安调度器
|
||||
|
||||
由于币安调度器元素只执行一次，因此请在网格 1 的末尾拆分执行路径，并通过将输出传递回币安调度器来强制让其重新同步。
|
||||
|
||||
![Grid 1: Split execution path][40]
|
||||
|
||||
网格 1：拆分执行路径
|
||||
|
||||
5A 元素指向 网格 2 的 1A 元素，并且 5B 元素指向网格 1 的 1A 元素（币安调度器）。
|
||||
|
||||
### 部署
|
||||
|
||||
你可以在本地计算机上全天候 7×24 小时运行整个程序，也可以将其完全托管在廉价的云系统上。例如，你可以使用 Linux/FreeBSD 云系统，每月约 5 美元，但通常不提供图形化界面。如果你想利用这些低成本的云，可以使用 PythonicDaemon，它能在终端中完全运行。
|
||||
|
||||
![PythonicDaemon console interface][41]
|
||||
|
||||
PythonicDaemon 控制台
|
||||
|
||||
PythonicDaemon 是基础程序的一部分。要使用它，请保存完整的工作流程，将其传输到远程运行的系统中（例如，通过安全拷贝协议Secure Copy<ruby>安全拷贝协议<rt>Secure Copy</rt></ruby> [SCP]），然后把工作流程文件作为参数来启动 PythonicDaemon：
|
||||
|
||||
|
||||
```
|
||||
`$ PythonicDaemon trading_bot_one`
|
||||
```
|
||||
|
||||
为了能在系统启动时自启 PythonicDaemon，可以将一个条目添加到 crontab 中：
|
||||
|
||||
|
||||
```
|
||||
`# crontab -e`
|
||||
```
|
||||
|
||||
![Crontab on Ubuntu Server][42]
|
||||
|
||||
在 Ubuntu 服务器上的 Crontab
|
||||
|
||||
### 下一步
|
||||
|
||||
正如我在一开始时所说的，本教程只是自动交易的入门。对交易机器人进行编程大约需要 10％ 的编程和 90％ 的测试。当涉及到让你的机器人用金钱交易时，你肯定会对编写的代码再三思考。因此，我建议你编码时要尽可能简单和易于理解。
|
||||
|
||||
|
||||
|
||||
如果你想自己继续开发交易机器人，接下来所需要做的事：
|
||||
|
||||
- 收益自动计算（希望你有正收益！）
|
||||
- 计算你想买的价格
|
||||
- 比较你的预订单（例如，订单是否填写完整？）
|
||||
|
||||
|
||||
|
||||
你可以从 [GitHub][2] 上获取完整代码。
|
||||
|
||||
--------------------------------------------------------------------------------
|
||||
|
||||
via: https://opensource.com/article/20/4/python-crypto-trading-bot
|
||||
|
||||
作者：[Stephan Avenwedde][a]
|
||||
选题：[lujun9972][b]
|
||||
译者：[wyxplus](https://github.com/wyxplus)
|
||||
校对：[校对者ID](https://github.com/校对者ID)
|
||||
|
||||
本文由 [LCTT](https://github.com/LCTT/TranslateProject) 原创编译，[Linux中国](https://linux.cn/) 荣誉推出
|
||||
|
||||
[a]: https://opensource.com/users/hansic99
|
||||
[b]: https://github.com/lujun9972
|
||||
[1]: https://opensource.com/sites/default/files/styles/image-full-size/public/lead-images/calculator_money_currency_financial_tool.jpg?itok=2QMa1y8c "scientific calculator"
|
||||
[2]: https://github.com/hANSIc99/Pythonic
|
||||
[3]: https://opensource.com/article/19/5/graphically-programming-pythonic
|
||||
[4]: https://tron.network/
|
||||
[5]: https://bitcoin.org/en/
|
||||
[6]: https://www.binance.com/
|
||||
[7]: https://www.investopedia.com/terms/e/ema.asp
|
||||
[8]: https://opensource.com/sites/default/files/uploads/1_ema-25.png "TRX/BTC 1-hour candle chart"
|
||||
[9]: https://en.wikipedia.org/wiki/Open-high-low-close_chart
|
||||
[10]: https://opensource.com/sites/default/files/uploads/2_data-mining-workflow.png "Data-mining workflow"
|
||||
[11]: https://opensource.com/sites/default/files/uploads/3_ohlc-query.png "Configuration of the OHLC query element"
|
||||
[12]: https://pandas.pydata.org/pandas-docs/stable/getting_started/dsintro.html#dataframe
|
||||
[13]: https://opensource.com/sites/default/files/uploads/4_edit-basic-operation.png "Basic Operation element set up to use Vim"
|
||||
[14]: https://opensource.com/sites/default/files/uploads/6_grid2-workflow.png "Technical analysis workflow in Grid 2"
|
||||
[15]: https://opensource.com/sites/default/files/uploads/7_technical-analysis-config.png "Configuration of the technical analysis element"
|
||||
[16]: https://opensource.com/sites/default/files/uploads/8_missing-decimals.png "Missing decimal places in output"
|
||||
[17]: https://opensource.com/sites/default/files/uploads/9_basic-operation-element.png "Workflow in Grid 2"
|
||||
[18]: https://opensource.com/sites/default/files/uploads/10_dump-extended-dataframe.png "Dump extended DataFrame to file"
|
||||
[19]: https://opensource.com/sites/default/files/uploads/11_load-dataframe-decimals.png "Representation with all decimal places"
|
||||
[20]: https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.DataFrame.iloc.html
|
||||
[21]: https://opensource.com/sites/default/files/uploads/12_trade-factor-decision.png "Buy/sell decision"
|
||||
[22]: https://opensource.com/sites/default/files/uploads/13_validation-function.png "Validation function"
|
||||
[23]: https://opensource.com/sites/default/files/uploads/14_brute-force-tf.png "Nested for loops for determining the buy and sell factor"
|
||||
[24]: https://opensource.com/sites/default/files/uploads/15_system-utilization.png "System utilization while brute forcing"
|
||||
[25]: https://opensource.com/sites/default/files/uploads/16_sort-profit.png "Sort profit with related trading factors in descending order"
|
||||
[26]: https://opensource.com/sites/default/files/uploads/17_sorted-trading-factors.png "Sorted list of trading factors and profit"
|
||||
[27]: https://opensource.com/sites/default/files/uploads/18_implemented-evaluation-logic.png "Implemented evaluation logic"
|
||||
[28]: https://opensource.com/sites/default/files/uploads/19_output.png "Branch element: Grid 3 Position 2A"
|
||||
[29]: https://opensource.com/sites/default/files/uploads/20_editbranch.png "Branch element: Grid 3 Position 3B"
|
||||
[30]: https://opensource.com/sites/default/files/uploads/21_grid3-workflow.png "Workflow on Grid 3"
|
||||
[31]: https://opensource.com/sites/default/files/uploads/22_pass-false-to-stack.png "Forward a False-variable to the subsequent Stack element"
|
||||
[32]: https://opensource.com/sites/default/files/uploads/23_stack-config.png "Configuration of the Stack element"
|
||||
[33]: https://opensource.com/sites/default/files/uploads/24_evaluate-stack-value.png "Evaluate the variable from the stack"
|
||||
[34]: https://opensource.com/sites/default/files/uploads/25_grid3-workflow.png "Workflow on Grid 3"
|
||||
[35]: https://opensource.com/sites/default/files/uploads/26_binance-order.png "Configuration of the Binance Order element"
|
||||
[36]: https://opensource.com/sites/default/files/uploads/27_api-key-binance.png "Creating an API key in Binance"
|
||||
[37]: https://opensource.com/sites/default/files/uploads/28_sell-order.png "Output of a successfully placed sell order"
|
||||
[38]: https://opensource.com/sites/default/files/uploads/29_binance-order-output.png "Logging output of Binance Order element"
|
||||
[39]: https://opensource.com/sites/default/files/uploads/30_binance-scheduler.png "Binance Scheduler at Grid 1, Position 1A"
|
||||
[40]: https://opensource.com/sites/default/files/uploads/31_split-execution-path.png "Grid 1: Split execution path"
|
||||
[41]: https://opensource.com/sites/default/files/uploads/32_pythonic-daemon.png "PythonicDaemon console interface"
|
||||
[42]: https://opensource.com/sites/default/files/uploads/33_crontab.png "Crontab on Ubuntu Server"
|
||||
Loading…
Reference in New Issue
Block a user