Data processing with Pandas course for the CM Hub at Imperial College
2 × 2 hour classes
- use Jupyter notebooks to perform simple Pandas data analysis,
- apply fundamental components of Pandas syntax including data selection and grouping,
- create programs designed to process example data and display simple statistics,
- interpret common errors and use these to help debug a program.
- Students are expected to be familiar with basic Python.
- Students are welcome to bring their own fully charged laptops to these sessions although there are computers in situ.
- If using your own laptop, pleasure ensure you have your preferred Python environment installed (we recommend Anaconda), including the modules
pandas,xlrd,statsmodels,matplotlib. The instructor will use Jupyter Notebooks for demonstrations.
Timestamps are in the descriptions and correspond to the numbering in this document.
- Part 1: https://www.youtube.com/watch?v=ozxAW2mm7A0
- Part 2: https://www.youtube.com/watch?v=MZuni39oxr0
- Playlist: https://www.youtube.com/playlist?list=PLcRp2oAQBugtlo4ZjnuzhOCS21SXba4Zq
- Pandas is a Python package. You need Python working on your system.
- On your own machine:
pip install pandaspip install xlrd(required for pd.read_excel() to work)pip install statsmodels
- On the college computers:
conda install pandasconda install xlrdconda install statsmodels- Restart the kernel (click Kernel > Restart)
Pros of Pandas:
- Is part of Python and Python is great
- Good for large data sets
- Powerful tools for dealing with broken data (etc.)
Cons of Pandas:
- You need to know Python
- No graphical interface
- You need a few extra packages
- Eats shoots and leaves
In short: The setup cost is often worthwhile to allow you powerful data manipulation features, which you can then connect to some useful Python code.
- Join in. Open a Jupyter notebook by typing
jupyter notebookin the terminal. - Jupyter notebook specific: We're going to want to display our graphs inside the notebook, rather than externally. Do to that, type
%matplotlib inline. - We're going to be using Python 3. If you've got Python 2, that's OK, just start by typing
from __future__ import print_functionprint("hello world")
Test it works!
import pandas as pdimport matplotlib.pyplot as plt
OK let's go
data = pd.read_csv("A1_mosquito_data.csv")data(fine within a Jupyter notebook)print(data)(otherwise)data.head()data.tail()data.sample()data.describe()
Let's pick out things:
data['year']data[['year','mosquitos']]data[0:2]
Your turn:
- Print the last 3 rows
- Print the first two rows, but only the year and rainfall columns
- Try to print a single row
Join in:
data[1:2]data.iloc[1]data.iloc[1:3]data['temperature'][0])(so you can select a single row this way)data['temperature'][data['temperature'] > 75]data['temperature'][data['year'] > 2005]data.mean()data.mean().mean()data.mean(1))(not very useful here)data.max()data.idxmin()data['temperature'].min()
Your turn:
- Print the mean number of mosquitos in the years where the rainfall was more than 200mm.
- Now do the same for when the rainfall was less than 200mm.
- Print the standard deviation of the temperatures (you may have to Google!)
These work in the usual Python way:
Join in:
for index, row in data.iterrows():
temp_in_f = row['temperature']
temp_in_c = (temp_in_f - 32) * 5 / 9.0
print(temp_in_c)- We can add in:
if temp_in_c > 27:
print("Hotter than 27C!")Your turn:
- Import the data from
A2_mosquito_data.csvintodata2, determine the mean temperature, and loop over the temperature values. For each value print out whether it is greater than the mean, less than the mean, or equal to the mean.
Join in:
mosquitos_vs_year = data[['year','mosquitos']]mosquitos_vs_year.plot(kind='line')mosquitos_vs_year.plot(kind='line',x='year')mosquitos_vs_year.plot(kind='bar',x='year')
Your turn:
- Using the
data2dataset, plot the number of mosquitos against the rainfall. What needs fixing?
Join in:
mosquitos_vs_rainfall = data[['rainfall','mosquitos']]mosquitos_vs_rainfall.plot(kind='scatter',x='rainfall',y='mosquitos')plt.xlabel('Rainfall (mm)')plt.ylabel('Mosquitos')plt.title('Mosquitoes like water')
Your turn:
- Plot the number of mosquitos against the temperature.
Join in:
plt.savefig('beautiful_graph.pdf') (directly after plot command in same cell)
Let's group the data by temperature and then plot the mean number of mosquitoes for each temperature (to the nearest degree).
Join in:
mosquito_data_only = data[['temperature','mosquitos']]mosquito_data_only.groupby('temperature').mean().plot(kind='line')
Your turn:
- Do the same with the larger data file.
Binning sorts data into intervals, or 'bins'.
Join in:
bins = [0,200,250,300]
labels = ['dry','normal','wet']
pd.cut(data['rainfall'],bins,labels=labels).value_counts().plot(kind='pie')We can add columns:
Join in:
data['temperature_celsius'] = (data['temperature']-32)*5/9.data.head()
We can sort data:
Join in:
data.sort_values('temperature')data.sort_values(['temperature','rainfall'])data.sort_values('temperature',ascending=False)data.mean().sort_values()
- If you want to continue from your previous Jupyter notebook and have managed to re-open it, you will have to click Cell > Run All.
- To continue in a new Jupyter notebook, place in the top cell
%matplotlib inlineand in the second cell:
import pandas as pd
import matplotlib.pyplot as plt
data = pd.read_csv("A1_mosquito_data.csv")
data2 = pd.read_csv("A2_mosquito_data.csv")data.hist()
And the scatter matrix:
from pandas.plotting import scatter_matrix
scatter_matrix(data[['temperature','rainfall','mosquitos']])Correlation matrix:
data.corr()
import statsmodels.api as sm
x = data["rainfall"]
y = data["mosquitos"]
X = sm.add_constant(x) # add y-intercept
model = sm.OLS(y, X).fit() # Ordinary Least Squares
predictions = model.predict(X)
model.summary()In
coef std err t P>|t| [0.025 0.975]
------------------------------------------------------------------------------
const 49.8413 3.223 15.465 0.000 42.409 57.273
rainfall 0.6592 0.015 44.965 0.000 0.625 0.693
we see mosquitos = 0.6592*rainfall + 49.8413
To get them out:
model.params['const']model.params['rainfall']model.pvaluesetc
We can ignore missing or NaN values:
data.sum(0, skipna=False)
- Import from Excel
exam_results.xlsx
You're the director of undergraduate studies. You have to do a few things.
- You need to give prizes to the five students taking Physics with the top mean marks over all four modules. Which students get the prizes?
- The staff member running the tutor group with the highest mean mark gets a beer. Which group's tutor gets the beer?
- You need to report the mean mark for each course to the faculty. List the four courses by order of mean mark. Plot these on a bar chart so they can understand it.
- Scores above 70% are a 'first'. Scores between 60 and 69% are an 'upper second', between 50 and 59% a 'lower second', between 40 and 49% a 'third', and 39% and below is a fail. For Quantum Mechanics, plot a pie chart showing the number of students who fall in each of these categories.
- Students on the Physics programme pass the year if they score more than 40% on three out of four modules. Otherwise they fail. How many students failed? Loop through the failing students, printing out a personalised statement (imagine that you will code it so Python emails it to them) telling them they've failed.
- Rumour has it the scores for Lab Work have been made up. Create a scatter matrix for the four courses. What does this tell you?
- Do a linear regression analysis to come up with a linear model for the Waves score based on the Relativity score.
If you're taking this course through the Graduate School, please fill out the feedback form.
Mosquitos example adapted from Software Carpentry. CC BY 4.0.
The rest of the work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International Licence.