#!/usr/bin/env python
# coding: utf-8

# ## Pandas demo
# 
# #### Henry Schreiner
# 
# A few basic imports

# In[ ]:


import pandas as pd
import numpy as np
import matplotlib.pyplot as plt


# ### MPG dataset
# 
# Let's read a CSV file from the web *directly* into Pandas.

# In[ ]:


data = pd.read_csv('https://raw.githubusercontent.com/mwaskom/seaborn-data/master/mpg.csv')
data


# Info about the dataframe:

# In[ ]:


data.info()


# Detailed memory usage:
# <!--
# data.memory_usage(deep=True)
# -->

# In[ ]:





# You can ask for statistical information per column or often per dataframe:
# 
# <!--
# data.mean()
# -->

# In[ ]:





# Plots are easy; try `.plot`, `.scatter`, or `.hist`:
# 
# <!--
# data.mpg.plot();
# data.plot.scatter('weight', 'horsepower');
# data.mpg.hist(bins=30);
# -->

# In[ ]:





# In[ ]:





# In[ ]:





# You can select subsets, such as `mpg > 42`:
# 
# <!--
# data[data.mpg > 42]
# -->

# In[ ]:





# Or you can use `groupby('cylinders')` to work on per-cylinder groups: 
# 
# <!--
# data.groupby('cylinders').mpg.hist();
# -->

# In[ ]:





# If you want legends, it's no longer one line, but still simple:

# In[ ]:


for name, grp in data.groupby('cylinders').mpg:
    grp.hist(label=name)
plt.legend();


# The category type is better for data.origin, and saves memory too!
# 
# <!--
# data.origin.memory_usage(deep=True)
# data.origin.astype('category').memory_usage(deep=True)
# -->

# In[ ]:


data.origin.astype('category')


# In[ ]:





# In[ ]:





# You can select using operators or isin:
# 
# <!--
# data[~data.origin.isin(['usa', 'japan'])]
# -->

# In[ ]:





# Now, let's convert the name into make and model:
# 
# <!--
# makemodel = cars.name.str.split(n=1, expand=True)
# -->

# In[ ]:


cars = pd.read_csv('https://raw.githubusercontent.com/mwaskom/seaborn-data/master/mpg.csv',
                  dtype={"origin":"category"})

...
# cars["make"], cars["model"] = makemodel[0].astype('category'), makemodel[1]
# del cars["name"]


# We can put make and model together again:
# 
# <!--
# cars.make.str.upper() + " " + cars.model.str.title()
# -->

# In[ ]:





# Math: To convert mpg to liters per 100 kilometers:
# 
# $$
# lp100km = \frac{1}{mpg} \cdot 62.1371 \frac{\mathrm{miles}}{\mathrm{100 km}} \cdot 3.78541 \frac{\mathrm{liter}}{\mathrm{gallon}}
# $$
# 
# <!--
# 1 / cars.mpg * 62.1371 * 3.78541
# cars.mpg.apply(lambda x: 1/x * 62.1371 * 3.78541)
# -->

# In[ ]:





# In[ ]:





# Also see:
# 
# 
# * 10 minutes to Pandas: https://pandas.pydata.org/pandas-docs/stable/getting_started/10min.html
# * 100 tips: https://www.dataschool.io/python-pandas-tips-and-tricks
# * My blog: https://iscinumpy.gitlab.io

# Not covered above:
# * Fantastic date/time support, including holidays
# * Resampling, interpolation
# * Multi-indexing
# * Support for many input formats, such as HTML, the clipbord, Excel files, and more

# In[ ]: