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

# # Makeover Monday, 9 April 2018
# 
# Analysis and visualisation of Arctic sea ice extent for [Makeover Monday](http://www.makeovermonday.co.uk).
# 
# Data from [US National Snow & Ice data centre](https://nsidc.org/) via [Makeover Monday](http://www.makeovermonday.co.uk/data/).

# In[1]:


import collections
from datetime import datetime
import matplotlib as mpl
import matplotlib.pyplot as plt
import matplotlib.cm as cm
get_ipython().run_line_magic('matplotlib', 'inline')

import numpy as np
import pandas as pd
import scipy.stats


# # Read the dataset
# Rename the columns while we're here.

# In[2]:


sea_ice_raw = pd.read_excel('Arctic Sea Ice Extent.xlsx')
sea_ice_raw.columns = ['date', 'extent']
sea_ice_raw.head()


# In[3]:


sea_ice_raw.tail()


# In[4]:


sea_ice_raw.dtypes


# In[5]:


sea_ice_raw.extent.plot()


# # Prepare for reshaping
# Annotate each entry in the dataset with the year, month, and day of year.

# In[6]:


sea_ice_raw['year'] = sea_ice_raw['date'].dt.year
sea_ice_raw['doy'] = sea_ice_raw['date'].dt.dayofyear
sea_ice_raw['month'] = sea_ice_raw['date'].dt.month
sea_ice_raw.head()


# # One column per year, one row per day
# Create a DataFrame with one column per year, and one row per day number.
# 
# Plot the data, with each line being a separate year. Pick out the first and last years in the data.

# In[7]:


sea_ice_day_year = pd.pivot_table(sea_ice_raw,index='doy',columns='year',values='extent')


# In[8]:


ax = sea_ice_day_year.loc[:, 1989:2016].plot(legend=None, label='_nolegend_', color='lightgrey')
sea_ice_day_year[1988].plot(color='blue', ax=ax, legend=True)
sea_ice_day_year[2017].plot(color='red', ax=ax, legend=True)


# # One column per year, one row per month
# Aggregate with median, to give an overall value for each month.

# In[9]:


sea_ice_month_year = pd.pivot_table(sea_ice_raw,index='month',columns='year',values='extent', aggfunc=np.median)


# In[10]:


sea_ice_month_year


# ## Medians for each month
# These are the median values of ice extent for each month, over the whole dataset.

# In[11]:


sea_ice_month_year.median(axis=1)


# Now find the difference of each month's extent from the median for that month over all years.

# In[12]:


sea_ice_month_year_diff = sea_ice_month_year.subtract(sea_ice_month_year.median(axis=1), axis=0)
sea_ice_month_year_diff


# In[13]:


sea_ice_month_year_diff.loc[:, 1988:2017]


# ## Quick plot of each year
# How each year varies against the median values for each month.

# In[14]:


sea_ice_month_year_diff.plot(legend=None, subplots=True, layout=(7, 6), sharey=True, figsize=(15, 15), kind='bar');


# Plot just the last 30 complete years.

# In[15]:


sea_ice_month_year_diff.loc[:, 1988:2017].plot(legend=None, subplots=True, layout=(6, 5), sharey=True, figsize=(15, 15), kind='bar');


# Similar plot, with each month in a subplot, trend over years.

# In[16]:


sea_ice_month_year_diff.T.plot(legend=None, subplots=True, layout=(3, 4), sharey=True, figsize=(15, 15), kind='line');


# ## Colouring the data
# Pandas doesn't make it easy to have each bar in a chart have a different colour depending on its value. Therefore, we'll create a separate dataframe with the colour that each data point should be plotted with.
# 
# To start with, normalise the data, so that the data ranges from 0 to 1.

# In[17]:


sid_max = sea_ice_month_year_diff.max().max()
sid_min = sea_ice_month_year_diff.min().min()
sea_ice_month_year_dnorm = (sea_ice_month_year_diff - sid_min) / (sid_max - sid_min)
sea_ice_month_year_dnorm


# Now use the `matplotlib` color map to find the colour of each value.
# 
# Note that we're doing `1 - value` to make the map scale from blue at high values to magenta at low ones, the opposite of what it normally does.

# In[18]:


cmap = cm.get_cmap('cool')
sea_ice_month_year_colour = (1 - sea_ice_month_year_dnorm).applymap(cmap)
sea_ice_month_year_colour


# Quick check it works for one chart.

# In[19]:


plt.bar(sea_ice_month_year_diff.index, sea_ice_month_year_diff[2017], color=sea_ice_month_year_colour[2017])


# Now replot the data for each year, but using the colours we manually specified.

# In[29]:


# sea_ice_month_year_diff.loc[:, 1988:2017].plot(legend=None, subplots=True, layout=(6, 5), sharey=True, figsize=(15, 15), kind='bar', color=sea_ice_month_year_colour);
f, axes = plt.subplots(6, 5, sharey=True, figsize=(15, 15), facecolor='darkslateblue')
plt.suptitle('Changes in sea ice extent vs 30-year average\nEach bar is one month', color='lightsteelblue', fontsize=24)
for r, row in enumerate(axes):
    for c, ax in enumerate(row):
        yr = 1988 + r * 5 + c
        ax.bar(sea_ice_month_year_diff.index, 
               sea_ice_month_year_diff[yr], 
               color=sea_ice_month_year_colour[yr])
        ax.set_frame_on(False)
        ax.set_axis_off()
        ax.set_title(str(yr), color='lightsteelblue')
#         ax.set_facecolor('darkslateblue')


# In[30]:


f.savefig("changes-per-year.png", facecolor=f.get_facecolor(), transparent=True)


# And again, plot for each month, one bar per year. 
# 
# Restrict to just the past 30 years of full data.

# In[21]:


# sea_ice_month_year_diff.loc[:, 1988:2017].plot(legend=None, subplots=True, layout=(6, 5), sharey=True, figsize=(15, 15), kind='bar', color=sea_ice_month_year_colour);
f, axes = plt.subplots(3, 4, sharey=True, figsize=(15, 15), facecolor='darkslateblue')
for r, row in enumerate(axes):
    for c, ax in enumerate(row):
        mt = r * 4 + c + 1
        ax.bar(sea_ice_month_year_diff.loc[:, 1988:2017].T.index, 
               sea_ice_month_year_diff.loc[:, 1988:2017].T[mt], 
               color=sea_ice_month_year_colour.loc[:, 1988:2017].T[mt])
        ax.set_frame_on(False)
        ax.set_axis_off()
        ax.set_title(datetime(1900, mt, 1).strftime('%B'), color='lightsteelblue')


# In[ ]: