02-data.md

---
title: Working with Data
teaching: 20
exercises: 10
questions:
- "How should I work with numeric data in Python?"
- "What's the recommended way to handle and analyse tabular data?"
- "How can I import tabular data for analysis in Python and export the results?"
objectives:
- "handle and summarise numeric data with Numpy."
- "filter values in their data based on a range of conditions."
- "load tabular data into a Pandas dataframe object."
- "describe what is meant by the data type of an array/series, and the impact this has on how the data is handled."
- "add and remove columns from a dataframe."
- "select, aggregate, and visualise data in a dataframe."
keypoints:
- "Specialised third-party libraries such as Numpy and Pandas provide powerful objects and functions that can help us analyse our data."
- "Pandas dataframe objects allow us to efficiently load and handle large tabular data."
- "Use the `pandas.read_csv` and `pandas.write_csv` functions to read and write tabular data."
---

## plan

- Toby currently scheduled to lead this session
- Numpy
  - arrays
  - masking
  - aside about data types and potential hazards
  - reading data from a file (with note that more will come later on this topic)
  - link to existing image analysis material
- Pandas
  - when an array just isn't enough
  - DataFrames - re-use material from [Software Carpentry][swc-python-gapminder]?
    - ideally with a more relevant example dataset... [maybe a COVID one](https://data.europa.eu/euodp/en/data/dataset/covid-19-coronavirus-data/resource/260bbbde-2316-40eb-aec3-7cd7bfc2f590)
    - include an aside about I/O - reading/writing files (pandas (the `.to_*()` methods and highlight some: `csv`, `json`, `feather`, `hdf`), numpy, `open()`, (?) bytes vs strings, (?) encoding)
  - Finish with example of `df.plot()` to set the scene for plotting section

## Numpy

## Reading data to a numpy array

We'll use the popular image analysis package scikit-image,
to read two example images into numpy arrays.

~~~
from skimage.io import imread
raw = imread('cilliated_cell.png')
nuclei = imread('cilliated_cell_nuclei.png')
# if you want to see what these images look like - we can use matplotlib (more to come later!)
import matplotlib.pyplot as plt
plt.imshow(raw, cmap='gray')
plt.imshow(nuclei)
~~~
{: .language-python }

> ## Exploring Image Arrays
>
> * What are the dimensions of these arrays?
> * What data type are these arrays?
> * What is the minimum and maximum value of these arrays?
>
> > ## Solution
> > ~~~
> > print(raw.shape)
> > print(raw.dtype)
> > print(np.max(raw))
> > print(np.min(raw))
> > ~~~
> > {: .language-python }
> {: .solution }
{: .challenge }

> ## Masking arrays
>
> The nuclei image contains a binary segmentation i.e.:
>
> * 1 = nuclei
> * 0 = not nuclei
>
> 1. Find the median value of the raw image within the nuclei
> 2. Create a new version of raw where all values outside the nuclei are 0
>
> > ## Solution
> > ~~~
> > # 1
> > pixels_in_nuclei = raw[nuclei == 1]
> > print(np.median(pixels_in_nuclei))
> >
> > # 2
> > new_image = raw.copy()
> > new_image[nuclei == 0] = 0
> > plt.imshow(new_image, cmap='gray')
> > ~~~
> > {: .language-python }
> {: .solution }
{: .challenge }


## Pandas

> ## Working with Filtered Data
>
> 1. On what date were the most cases reported in Germany so far?
> 2. What was the mean number of cases reported per day in Germany in April 2020?
> 3. Is this higher or lower than the mean for March 2020?
> 4. On how many days in March was the number of cases in Germany higher than the mean for April?
>
> > ## Solution
> > ~~~
> > # 1
> > mask_germany = covid_cases['countryterritoryCode'] == 'DEU'
> > id_max = covid_cases[mask_germany]['cases'].idxmax()
> > print(covid_cases.iloc[id_max]['dateRep'])
> >
> > # 2
> > mask_april = (covid_cases['year'] == 2020) & (covid_cases['month'] == 4)
> > mean_april = covid_cases[mask_germany & mask_april]['cases'].mean()
> > print(mean_april)
> >
> > # 3
> > mask_march = (covid_cases['year'] == 2020) & (covid_cases['month'] == 3)
> > mean_march = covid_cases[mask_germany & mask_march]['cases'].mean()
> > print(mean_march)
> > print("Mean cases per day was {} in April than in March 2020.".
> >       format(["lower", "higher"][mean_april > mean_march]))
> >
> > # 4
> > mask_higher_mean_april = (covid_cases['cases'] > mean_april)
> > selection = covid_cases[mask_germany & mask_march & mask_higher_mean_april]
> > nbr_days = len(selection)   # Assume clean data
> > print(nbr_days)
> > ~~~
> > {: .language-python }
> {: .solution }
{: .challenge }

{% include links.md %}