Lesser-Known Pandas Tricks

栏目: IT技术 · 发布时间: 4年前

内容简介:pandas needs no introduction as it became the de facto tool for data analysis in Python. As a Data Scientist, I use pandas daily and I am always amazed by how many functionalities it has. In this post, I am going to show you 5 pandas tricks that I learned

5 lesser-known pandas tricks

Lesser-Known Pandas Tricks

Pandas provides high-performance, easy-to-use data structures and data analysis tools for the Python

pandas needs no introduction as it became the de facto tool for data analysis in Python. As a Data Scientist, I use pandas daily and I am always amazed by how many functionalities it has. In this post, I am going to show you 5 pandas tricks that I learned recently and using them helps me to be more productive.

For pandas newbies — Pandas provides high-performance, easy-to-use data structures and data analysis tools for the Python programming language. The name is derived from the term “panel data”, an econometrics term for data sets that include observations over multiple time periods for the same individuals.

To run the examples download this Jupyter notebook .

To Step Up Your Pandas Game, read:

1. Date Ranges

When fetching the data from an external API or a database, we many times need to specify a date range. Pandas got us covered. There is a data_range function, which returns dates incremented by days, months or years, etc.

Let’s say we need a date range incremented by days.

date_from = "2019-01-01"
date_to = "2019-01-12"
date_range = pd.date_range(date_from, date_to, freq="D")
date_range

Let’s transform the generated date_range to start and end dates, which can be passed to a subsequent function.

for i, (date_from, date_to) in enumerate(zip(date_range[:-1], date_range[1:]), 1):
date_from = date_from.date().isoformat()
date_to = date_to.date().isoformat()
print("%d. date_from: %s, date_to: %s" % (i, date_from, date_to))
1. date_from: 2019-01-01, date_to: 2019-01-02
2. date_from: 2019-01-02, date_to: 2019-01-03
3. date_from: 2019-01-03, date_to: 2019-01-04
4. date_from: 2019-01-04, date_to: 2019-01-05
5. date_from: 2019-01-05, date_to: 2019-01-06
6. date_from: 2019-01-06, date_to: 2019-01-07
7. date_from: 2019-01-07, date_to: 2019-01-08
8. date_from: 2019-01-08, date_to: 2019-01-09
9. date_from: 2019-01-09, date_to: 2019-01-10
10. date_from: 2019-01-10, date_to: 2019-01-11
11. date_from: 2019-01-11, date_to: 2019-01-12

2. Merge with indicator

Merging two datasets is the process of bringing two datasets together into one, and aligning the rows from each based on common attributes or columns.

One of the arguments of the merge function that I’ve missed is the indicator argument. Indicator argument adds a _merge column to a DataFrame, which tells you “where the row came from”, left, right or both DataFrames. The _merge column can be very useful when working with bigger datasets to check the correctness of a merge operation.

left = pd.DataFrame({"key": ["key1", "key2", "key3", "key4"], "value_l": [1, 2, 3, 4]})

Lesser-Known Pandas Tricks

right = pd.DataFrame({"key": ["key3", "key2", "key1", "key6"], "value_r": [3, 2, 1, 6]})

Lesser-Known Pandas Tricks

df_merge = left.merge(right, on='key', how='left', indicator=True)

Lesser-Known Pandas Tricks

The _merge column can be used to check if there is an expected number of rows with values from both DataFrames.

df_merge._merge.value_counts()both          3
left_only 1
right_only 0
Name: _merge, dtype: int64

3. Nearest merge

When working with financial data, like stocks or cryptocurrencies, we may need to combine quotes (price changes) with actual trades. Let’s say that we would like to merge each trade with a quote that occurred a few milliseconds before it. Pandas has a function merge_asof, which enables merging DataFrames by the nearest key (timestamp in our example). The datasets quotes and trades are taken from pandas example

The quotes DataFrame contains price changes for different stocks. Usually, there are many more quotes than trades.

quotes = pd.DataFrame(
[
["2016-05-25 13:30:00.023", "GOOG", 720.50, 720.93],
["2016-05-25 13:30:00.023", "MSFT", 51.95, 51.96],
["2016-05-25 13:30:00.030", "MSFT", 51.97, 51.98],
["2016-05-25 13:30:00.041", "MSFT", 51.99, 52.00],
["2016-05-25 13:30:00.048", "GOOG", 720.50, 720.93],
["2016-05-25 13:30:00.049", "AAPL", 97.99, 98.01],
["2016-05-25 13:30:00.072", "GOOG", 720.50, 720.88],
["2016-05-25 13:30:00.075", "MSFT", 52.01, 52.03],
],
columns=["timestamp", "ticker", "bid", "ask"],
)
quotes['timestamp'] = pd.to_datetime(quotes['timestamp'])

Lesser-Known Pandas Tricks

The trades DataFrame contains trades of different stocks.

trades = pd.DataFrame(
[
["2016-05-25 13:30:00.023", "MSFT", 51.95, 75],
["2016-05-25 13:30:00.038", "MSFT", 51.95, 155],
["2016-05-25 13:30:00.048", "GOOG", 720.77, 100],
["2016-05-25 13:30:00.048", "GOOG", 720.92, 100],
["2016-05-25 13:30:00.048", "AAPL", 98.00, 100],
],
columns=["timestamp", "ticker", "price", "quantity"],
)
trades['timestamp'] = pd.to_datetime(trades['timestamp'])

Lesser-Known Pandas Tricks

We merge trades and quotes by tickers, where the latest quote can be 10 ms behind the trade. If a quote is more than 10 ms behind the trade or there isn’t any quote, the bid and ask for that quote will be null (AAPL ticker in this example).

pd.merge_asof(trades, quotes, on="timestamp", by='ticker', tolerance=pd.Timedelta('10ms'), direction='backward')

Lesser-Known Pandas Tricks

4. Create an Excel report

Lesser-Known Pandas Tricks

Pandas (with XlsxWriter library) enables us to create an Excel report from the DataFrame. This is a major time saver — no more saving a DataFrame to CSV and then formatting it in Excel. We can also add all kinds of charts , etc.

df = pd.DataFrame(pd.np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]]), columns=["a", "b", "c"])

The code snippet below creates an Excel report. To save a DataFrame to the Excel file, uncomment the writer.save() line.

report_name = 'example_report.xlsx'
sheet_name = 'Sheet1'
writer = pd.ExcelWriter(report_name, engine='xlsxwriter')
df.to_excel(writer, sheet_name=sheet_name, index=False)
# writer.save()

As mentioned before, the library also supports adding charts to the Excel report. We need to define the type of the chart (line chart in our example) and the data series for the chart (the data series needs to be in the Excel spreadsheet).

# define the workbook
workbook = writer.book
worksheet = writer.sheets[sheet_name]
# create a chart line object
chart = workbook.add_chart({'type': 'line'})
# configure the series of the chart from the spreadsheet
# using a list of values instead of category/value formulas:
# [sheetname, first_row, first_col, last_row, last_col]
chart.add_series({
'categories': [sheet_name, 1, 0, 3, 0],
'values': [sheet_name, 1, 1, 3, 1],
})
# configure the chart axes
chart.set_x_axis({'name': 'Index', 'position_axis': 'on_tick'})
chart.set_y_axis({'name': 'Value', 'major_gridlines': {'visible': False}})
# place the chart on the worksheet
worksheet.insert_chart('E2', chart)
# output the excel file
writer.save()

5. Save the disk space

When working on multiple Data Science projects, you usually end up with many preprocessed datasets from different experiments. Your SSD on a laptop can get cluttered quickly. Pandas enables you to compress the dataset when saving it and then reading back in compressed format.

Let’s create a big pandas DataFrame with random numbers.

df = pd.DataFrame(pd.np.random.randn(50000,300))

Lesser-Known Pandas Tricks

When we save this file as CSV, it takes almost 300 MB on the hard drive.

df.to_csv('random_data.csv', index=False)

With a single argument compression='gzip' , we can reduce the file size to 136 MB.

df.to_csv('random_data.gz', compression='gzip', index=False)

It is also easy to read the gzipped data to the DataFrame, so we don’t lose any functionality.

df = pd.read_csv('random_data.gz')

Conclusion

These tricks help me daily to be more productive with pandas. Hopefully, this blog post showed you a new pandas function, that will help you to be more productive.

What’s your favorite pandas trick?


以上就是本文的全部内容,希望本文的内容对大家的学习或者工作能带来一定的帮助,也希望大家多多支持 码农网

查看所有标签

猜你喜欢:

本站部分资源来源于网络,本站转载出于传递更多信息之目的,版权归原作者或者来源机构所有,如转载稿涉及版权问题,请联系我们

重构(影印版)

重构(影印版)

Martin Fowler / 中国电力出版社 / 2003-7-1 / 49.00元

随着对象技术应用越来越普及,软件开发社区出现了一个新的问题。缺乏经验的开发者编写出了大批设计较差的程序,导致这些应用程序非常低效,且难于维护和扩展。本书除了讨论重构的各种技巧之外,还提供了超过70个可行重构的详细编目,对如何应用它们给出了有用的提示;并以step by step的形式给出了应用每一种重构的指南;而且用实例展示了重构的工作原理。这些示例都是用Java语言写成的,但其中的思想却可以运用......一起来看看 《重构(影印版)》 这本书的介绍吧!

随机密码生成器
随机密码生成器

多种字符组合密码

HEX HSV 转换工具
HEX HSV 转换工具

HEX HSV 互换工具

HSV CMYK 转换工具
HSV CMYK 转换工具

HSV CMYK互换工具