wrangling.py

#===========================================================#
# METRO RIDERSHIP ANALYSIS - DATA WRANGLING
#
# NOTE: Project for GA Data Science class.
# In this file is, data wrangling, data visualization,
# and generating the compiled dataset for analysis
#===========================================================#
# CREATED BY: Lena Nguyen - March 15, 2015
#===========================================================#

import pandas as pd
import numpy as np
import matplotlib.pylab as plt
import os

# Set current working directory
os.chdir("/Users/Zelda/Data Science/GA/Project/Data")

#=======================================================#
# METRO DATA
#=======================================================#

# Data of daily ridership of metro rail from Open Data DC
# Note: Not disaggregated by station
# Source: http://www.opendatadc.org/dataset/wmata-metrorail-ridership-by-date
# Data scraped from visualization: http://planitmetro.com/ridership_cal/

#===========#
# READ DATA #
#===========#

# OPEN DATA DC METRO RIDERSHIP DATA
metro = pd.read_csv('metro.csv', header=False, names=['Date','Riders'])

# QUICK LOOK
metro.head(10)
metro.describe()
metro.dtypes

assert (metro.duplicated().sum() == 0)  # make sure there are no duplicate values
metro.isnull().sum()  # check for missing values

#======================#
# CLEAN/TRANSFORM DATA #
#======================#

# CLEAN DATA
metro[metro['Riders'] == 0]
#-- One value of 0 for October 29, 2012. Is this correct?
#-- Google of the date shows that the value is correct
#-- Whole metrorail system was shut down due to Hurricane Sandy on that day
#-- https://www.wmata.com/about_metro/news/PressReleaseDetail.cfm?ReleaseID=5362
#-- http://www.wmata.com/about_metro/news/PressReleaseDetail.cfm?ReleaseID=5363

# TRANSFORM DATES

# NOTE: More interested in weekday versus weekend ridership
# So I need to transform each into its corresponding day of the week
metro['Date'] = pd.to_datetime(metro.Date, format='%Y-%m-%d')
metro.dtypes
metro.set_index('Date', inplace=True)

# Year
metro['Year'] = metro.index.year
# Month
metro['Month'] = metro.index.month
# Day
metro['Day'] = metro.index.day

# Label day of week
metro['Weekday'] = metro.index.weekday  # Creates integer for day of week

weekend = metro[metro.Weekday.isin([5, 6])]
weekday = metro[~(metro.Weekday.isin([5, 6]))]

# Map the integer to names
metro['Weekday'] = metro.Weekday.map({  0:'Mon', 1:'Tue', 2:'Wed',
                                        3:'Thu', 4:'Fri', 5:'Sat',
                                        6:'Sun'})

#============#
# GRAPH DATA #
#============#

# General trend of ridership over the years
metro.groupby('Year').Total.mean().plot(kind='line',
                                        color='r',
                                        linewidth=2,
                                        title='Average Metro Ridership by Year')
plt.savefig('Average Ridership by Year.png')  # save plot to file

metro.groupby('Month').Total.mean().plot(kind='line',
                                        color='c',
                                        linewidth=2,
                                        title='Average Metro Ridership by Month')
plt.savefig('Average Ridership by Month.png')  # save plot to file

# General trend of ridership by days of the week
metro.groupby(['Year','Weekday']).Total.mean().unstack(0).plot(kind='line',
                                                                title='Average Metro Ridership by Day of the Week')
plt.legend(bbox_to_anchor=(1.05, 1), loc=2, borderaxespad=0.)  # moves legend to the side
plt.savefig('Average Ridership by Day of Week.png')

# Box plot of the same graph as above
# Shows average ridership by day of the week
metro.boxplot(column='Total', by='Weekday', sym=' ')
plt.xlabel('Day of Week')
plt.ylabel('Average number of riders')
plt.title('Average number of riders by Day')
plt.savefig('Average Ridership by Day of Week (Boxplot).png')

# Looking at ridership by month to see if there's any seasonal variation
# Hard to look at but you get the general trend. There's gotta be a prettier way
metro.groupby(['Year', 'Month']).Total.mean().unstack(0).plot(kind='bar')
plt.legend(bbox_to_anchor=(1.05, 1), loc=2, borderaxespad=0.)

# Line graph of average metro ridership by month
metro.groupby(['Year', 'Month']).Total.mean().unstack(0).plot(kind='line',
                                                            title='Average Metro Ridership by Month')
plt.legend(bbox_to_anchor=(1.05, 1), loc=2, borderaxespad=0.)
# Mildly easier but not by much. Can see the general trend

# Effects of Snowmageddon (February 2010)
metro_snow = metro[(metro.Year.isin([2010])) & (metro.Month == 2)]


metro_snow.groupby(['Year', 'Day']).Total.sum().unstack(0).plot(kind='bar',
                                                                color='c',
                                                                figsize=(7, 9),
                                                                title='Effects of Snow Day on Metro Ridership')
plt.legend().set_visible(False)  # Hides legend
plt.savefig('Ridership for February 2010.png')
# ? How to turn the bars different lines different colors to highlight?

# Look at a regular month (July 2010)
metro_july = metro[(metro.Year.isin([2010])) & (metro.Month == 7)]

metro_july.groupby(['Year', 'Day']).Total.sum().unstack(0).plot(kind='bar',
                                                                color='r',
                                                                figsize=(7,9),
                                                                title='Daily Ridership in July 2010')
plt.legend().set_visible(False)  # Hides legend
plt.savefig('Ridership for July 2010.png')


# Graph weekend and weekday ridership on the same line
weekday.groupby('Month').Total.mean().plot(kind='line', color='b', linewidth=2)
weekend.groupby('Month').Total.mean().plot(kind='line', color='r', linewidth=2)
plt.xlabel('Month')
plt.ylabel('Average Number of Riders')
plt.title('Weekend and Weekday Ridership by Month')
plt.axis([1, 12, 100000, 800000])
plt.savefig('Weekday vs weekend Ridership.png')

#########################################################################

#=============================#
# MAY 2013 ENTRANCE/EXIT DATA #
#=============================#
#-- NOTE: This dataset will help us see how people are moving throughout
#-- the metrorail system in different times of days
#-- For example, is everyone coming from the MD/VA suburbs in the morning to DC?
#-- Not very important for the regression but interesting data

#-- NOTE: Late night data only for Saturday night (Labeled as Sunday).
#-- The number of riders seem too small from personal experience
#-- I plan to ignore late night data completely. Does not seem reliable.

# Data of rail ridership from May 2013
# This data will be mainly used to see how people are moving throught the Metrorail system
# Source: http://planitmetro.com/2014/08/28/data-download-may-2013-2014-metrorail-ridership-by-origin-and-destination/

# MAY 2013 RIDERSHIP DATA
may13 = pd.read_csv('Metro_May_2013_Data.csv', header=True,
                    names=['Holiday', 'Day', 'Entrance', 'Exit', 'Period', 'Riders'])

# QUICK LOOK
may13.head(10)
may13.describe()
may13.dtypes

#======================#
# CLEAN/TRANSFORM DATA #
#======================#

# Transform Riders column from object to integer
may13['Riders'] = may13['Riders'].map(lambda x: x.replace(',', ''))  # removes the , from the numbers
may13['Riders'] = may13.Riders.astype('int')  # makes them all ints
may13['Riders'].describe()

#============#
# GRAPH DATA #
#============#

may13.groupby(['Period']).Riders.sum().plot(kind='line', color='r',figsize=(7,9))
plt.title('Total Daily Ridership in May 2013')
plt.legend().set_visible(False)  # Hides legend

# AM PEAK
ampeak = may13[may13.Period.isin(['AM Peak'])]  # Make a data frame with the AM Peak obs

# Make a stacked bar graph with the number of riders by Entrance and Exit Stations
ampeak.groupby(['Exit']).Riders.sum().plot(kind='bar', color='b')
ampeak.groupby(['Entrance']).Riders.sum().plot(kind='bar', color='r')
plt.title('Number of Riders Entering/Exiting at Station during AM Peak in May 2013')
plt.legend()

# PM PEAK
pmpeak = may13[may13.Period.isin(['PM Peak'])]  # Make a data frame with the PM Peak obs
# Make a stacked bar graph with the number of riders by Entrance and Exit Stations

pmpeak.groupby(['Exit']).Riders.sum().plot(kind='bar', color='m', label='Exit')
pmpeak.groupby(['Entrance']).Riders.sum().plot(kind='bar', color='g', label='Entrance')
plt.title('Number of Riders Entering/Exiting at Station during PM Peak in May 2013')
plt.legend()

#########################################################################

#=======================================================#
# WEATHER DATA
#=======================================================#

# Daily weather data
# Source: NOAA, Washington Reagan National Airport Weather Station

# IMPORTANT: Variables are all measured in METRIC units.
# PRCP/SNOW is measured in millimiters .
# TMAX/TMIN are in tenths of degrees celsius

#===========#
# READ DATA #
#===========#

weather = pd.read_csv('Weather - Daily.csv')

# QUICK LOOK
weather.columns.values
weather.head(10)
weather.dtypes
weather.describe()
#-- 406 degrees Celsius (~771 deg F/hotter than hell) seems a bit high to be a max temp.
#-- TMAX and TMIN variable have to be divided by 10 to get degrees Celsius
#-- That must be what tenths of degrees celsius in documentation means

assert (weather.duplicated().sum() == 0)  # make sure there are no duplicate values
weather.isnull().sum()  # check for missing values

#======================#
# CLEAN/TRANSFORM DATA #
#======================#
# Documentation says that "-9999" means it's a missing value
# Will transform those to NaN values to see how many missing values there actually are
cols = list(weather.columns.values)
for col in cols:
    weather[col][weather[col] == -9999] = np.nan

weather.isnull().sum()  # check number of missing values
#-- WT columns have a 1 for a day when a certain weather type occurs
#-- Most weather don't occur that often
#-- Surprisingly, snow (WT18) occurs only 264 days from 2004-2014. Hail (WT05) occurs more often at 364 days
#-- Most common are Fog (WT01), Rain (WT16), Mist (WT13), Haze (WT08)

#-- Will drop some (maybe all?) WT columns
#-- For now, keep Hail (WT05), Rain (WT16), Snow (WT18), Thunder (WT03)

# REMOVE SOME COLUMNS
weather.drop(weather.columns[:2], axis=1, inplace=True)  # Removes Station/Station Name columns
weather.drop(weather.columns[6:14], axis=1, inplace=True)  # Removes WT columns

weather.columns.values  # assess the damage

weather.drop(weather.columns[7:12], axis=1, inplace=True)  # Removes more WT columns
weather.columns.values  # assess the damage
weather.drop(weather.columns[-2:], axis=1, inplace=True)  # Remove more WT columns
del weather['WT08']  # Last manual removal

weather.columns.values  # assess the damage

# CHANGE MISSING VALUES TO 0
# This will make the WT variables binary variables
cols = list(weather.columns.values)
for col in cols:
    weather[col][np.isnan(weather[col])] = 0

weather.isnull().sum()  # No missing values
weather.describe()

# SEPARATE DATE INTO YEAR, MONTH, DAY
weather['DATE'] = pd.to_datetime(weather.DATE, format='%Y%m%d')
weather.set_index('DATE', inplace=True)

# Year
weather['Year'] = weather.index.year
# Month
weather['Month'] = weather.index.month
# Day
weather['Day'] = weather.index.day

# Drop date column
weather.reset_index(drop=True)

# Convert from metric to imperial units for easier understanding
def convert(df, var):
    if var in ['TMAX', 'TMIN']:
        df[var] = df[var].map(lambda x: (((9.0/5.0)*(x/10.0) + 32)))
    elif var == 'PRCP':
        df[var] = df[var].map(lambda x: (x/254.0))
    elif var in ['SNWD', 'SNOW']:
        df[var] = df[var].map(lambda x: (x/25.4))

# Loop through all the variables that need converting
for x in weather.columns.values[:5]:
    convert(weather,x)

weather.describe()
# Those max/min temps look much more reasonable

#============#
# GRAPH DATA #
#============#

# Look at data numerically
weather.groupby('Year')['PRCP','SNWD', 'SNOW','TMAX','TMIN'].mean()  # Average by year
weather.groupby('Year')['PRCP','SNWD', 'SNOW','TMAX','TMIN'].max()  # Average by year

# Look at data numerically
weather.groupby('Month')['PRCP','SNWD', 'SNOW','TMAX','TMIN'].mean()  # Average by month
weather.groupby('Month')['PRCP','SNWD', 'SNOW','TMAX','TMIN'].max()  # Average by year

# Scatter matrix (to help check for collinearity)
pd.scatter_matrix(weather[['PRCP','SNWD', 'SNOW']])

# Total rain fail by month
weather.groupby('Month').PRCP.sum().plot(kind='bar',color='g')
plt.xlabel('Month')
plt.ylabel('Total Precipitation Amount (Inches)')
plt.title('Total Precipitation by Month')

weather.groupby('Month').SNWD.mean().plot(kind='bar',color='g')
plt.xlabel('Month')
plt.ylabel('Total Precipitation Amount (Inches)')
plt.title('Total Precipitation by Month')

# Average low/high temperature by month
weather.groupby('Month').TMAX.mean().plot(kind='line', color='b', label='Max Temp')
weather.groupby('Month').TMIN.mean().plot(kind='line', color='r', label='Min Temp')
plt.xlabel('Month')
plt.ylabel('Average Temperature (Deg F)')
plt.axis([1, 12, 0, 100])
plt.legend()

# Record max/min temp from 2004-2014 by month
weather.groupby('Month').TMAX.max().plot(kind='line', color='b', label='Max Temp')
weather.groupby('Month').TMIN.min().plot(kind='line', color='r', label='Min Temp')
plt.xlabel('Month')
plt.ylabel('Temperature (Deg F)')
plt.axis([1, 12, 0, 120])
plt.legend()

#============#
# MERGE DATA #
#============#

data = pd.merge(metro, weather, on=['Year', 'Month', 'Day'])  # Merge metro/weather data

data.columns.values
data.head(10)
assert len(data) == 4018  # should have 4018 obs in dataset

##########################################################################

#=======================================================#
# GAS PRICE DATA
#=======================================================#

# Monthly data of gas prices in Lower Atlantic Region from EIA
# Source URL: http://www.eia.gov/dnav/pet/pet_pri_gnd_dcus_r1z_m.htm

#===========#
# READ DATA #
#===========#

gas = pd.read_csv('Gas Prices.csv')
gas.columns = ['Date', 'Gas_Price']  # Rename columns

# QUICK LOOK
gas.head(10)
gas.describe
gas.dtypes

assert (gas.duplicated().sum() == 0)  # make sure there are no duplicate values
gas.isnull().sum()
# No missing data

#======================#
# CLEAN/TRANSFORM DATA #
#======================#

# Round price column to 2 decimal places to look like dollar prices
gas['Gas_Price'] = np.round(gas['Gas_Price'], decimals=2)

# Change date into month/year column
gas['Date'] = pd.to_datetime(gas.Date, format='%m/%d/%Y')
gas.dtypes
gas.set_index('Date', inplace=True)

# Year
gas['Year'] = gas.index.year

# Month
gas['Month'] = gas.index.month
# Quarterly assignment
gas['Quarter'] = [((x-1)//3)+1 for x in gas['Month']]

#============#
# GRAPH DATA #
#============#

# General trend of gas price over the years (1993-2015)
gas.groupby('Year').Gas_Price.mean().plot(kind='line', color='b', linewidth=2)
plt.title('Average Gas Price by Year')
plt.xlabel('Year')
plt.ylabel('Average gas price (USD/Gallon)')
plt.axis([1993, 2014, 0.50, 4.00])
plt.savefig('Average Gas Price by Year.png')
#-- Graph shows continuous price increase since 1993, a short slight dip around 2009,
#-- Gas prices have been going down since Fall 2014

# See if there are seasonal differences in gas prices (Unlikely)
gas.groupby(['Quarter', 'Year']).Gas_Price.mean().unstack(0).plot(kind='bar', figsize=(7,9))
plt.title('Average Gas Price by Quarter')
plt.xlabel('Year')
plt.ylabel('Average Gas Price (USD/Gallon)')
plt.legend().set_visible(False)  # Hides legend
## As suspected, no seasonal differences in gas prices that occur annually

#============#
# MERGE DATA #
#============#

del gas['Quarter']  # delete quarter variable before merge
data = pd.merge(data, gas, on=['Year', 'Month'])  # Merge in gas data

data.columns.values
data.head(10)
assert len(data) == 4018  # should have 4018 obs in dataset

##########################################################################

#=======================================================#
# UNEMPLOYMENT DATA
#=======================================================#

# Monthly Unemployment data for the DC metro area, not adjusted for seasonality
# Source: http://www.bls.gov/eag/eag.dc_washington_md.htm

#===========#
# READ DATA #
#===========#

labor = pd.read_csv('Unemployment.csv')

# QUICK LOOK
labor.dtypes
labor.head(10)
labor.describe()

assert (labor.duplicated().sum() == 0)  # check for duplicate values
labor.isnull().sum()  # Check for missing values

#======================#
# CLEAN/TRANSFORM DATA #
#======================#

# Change month abbreviation to integers for easier merge
monthDict = {'Jan':1, 'Feb':2, 'Mar':3,
             'Apr':4, 'May':5, 'Jun':6,
             'Jul':7, 'Aug':8, 'Sep':9,
             'Oct':10, 'Nov':11, 'Dec':12}

labor['Month'] = labor.Month.map(monthDict)

# Check it worked
labor.head(10)
labor.dtypes

#============#
# GRAPH DATA #
#============#

# General trend of unemployment rate over the years (2000-2015)
labor.groupby('Year').Unemp_Rate.mean().plot(kind='line', color='b', linewidth=2)
plt.title('Average Unemployment Rate by Year')
plt.xlabel('Year')
plt.ylabel('Average Unemployment Rate')
plt.savefig('Average Unemployment Rate by Year.png')
#-- This smooths out the seasonal effects. Unemployment rate doubled in 2008.
#-- This was due to the financial crisis in 2008 due to the housing bubble bursting

#============#
# MERGE DATA #
#============#

data = pd.merge(data, labor, on=['Year', 'Month'])  # Merge in unemployment data

data.columns.values
data.head(10)
assert len(data) == 4018  # should have 4018 obs

##########################################################################

#=======================================================#
# FEDERAL GOVERNMENT CLOSING DATA
#=======================================================#

# Data with dates of federal holidays from 1997 - 2020
# Source: http://catalog.data.gov/dataset/federal-holidays
# Also added the government shut down days in October 2013 and snow days from OPM website

#===========#
# READ DATA #
#===========#

holiday = pd.read_csv('holidays.csv')

holiday.head(10)  # quick look at data
holiday.describe()


assert (holiday.duplicated().sum() == 0)  # make sure there are no duplicate values
holiday.isnull().sum()  # check for missing values

#======================#
# CLEAN/TRANSFORM DATA #
#======================#

# Keep only days where federal government was closed for relevant years (2004-2014)
holiday = holiday[(holiday['Year'] > 2003) & (holiday['Year'] < 2015)]
holiday.head(10)  # check it worked

holiday['Holiday'] = 1  # mark which days where federal govt was closed for merge
holiday.head()  # check it worked

#============#
# MERGE DATA #
#============#

data = pd.merge(data, holiday, how='outer', on=['Year', 'Month', 'Day'])  # Merge in holidays data

data.head(10)
assert len(data) == 4018  # Should be 4018 obs

# Replace NaN values with 0 to make it a binary variable
data['Holiday'].fillna(value=0, inplace=True)
data['Holiday'].isnull().sum()  # check it worked
data['Holiday'] = data['Holiday'].astype('int8')  # Turn into integer

#=======================================================#
# CAPITAL BIKESHARE DATA
#=======================================================#

# Data summarizing the capital bikeshare ridership data
# This dataset of number of registered/casual was compiled from their trip history data
# Source: https://www.capitalbikeshare.com/trip-history-data

# Casual useers will be a proxy for tourism
# CaBi started in 2010 so there is only data from 2010-2014

cabi = pd.read_csv('CaBi.csv')

cabi.head(10)  # quick look at data
cabi.describe()

assert (cabi.duplicated().sum() == 0)  # make sure there are no duplicate values
cabi.isnull().sum()

#============#
# MERGE DATA #
#============#

data = pd.merge(data, cabi, how='outer', on=['Year', 'Month', 'Day'])  # Merge in bikeshare data

data.head(10)
assert len(data) == 4018  # Should be 4018 obs

#=======================================================#
# NUMBER OF TRAINS DATA
#=======================================================#

# Using the data from the WMATA website:
# http://www.wmata.com/rail/frequency.cfm
# and a bit of math.
# See the following spreadsheet for the math part:
# https://docs.google.com/spreadsheets/d/1sXMg_2tFk22iIffE610YcJFJAdwMj2NgVwkaKZR2Evo/edit?usp=sharing

#===========#
# READ DATA #
#===========#

cars = pd.read_csv('Metro_cars.csv')
cars.columns = ['Weekday', 'Cars']  # rename columns
cars.head(10)

# Change Weekday variable be the 3 letter abbreviation of day name
# so that it looks like the metro data for merge
cars['Weekday'] = cars['Weekday'].apply(lambda x: x[0:3])

cars.head(10)  # make sure it worked

#============#
# MERGE DATA #
#============#

data = pd.merge(data, cars, on='Weekday')  # Merge into data

data.sort(['Year', 'Month', 'Day'], ascending=True, inplace=True)  # sort by date
data.index = range(0,len(data.index))  # Re-indexing after sort
data.head(10)  # Checked it worked
data.columns.values  # Check that the column names are all correct

#=================================#
# CLEAN IRREGULAR TRAIN SCHEDULES #
#=================================#

#- HOLIDAY SCHEDULE
#- Holidays that fall on a weekday run on a Saturday schedule except for major holidays
#- (ie Christmas), where it runs on a Sunday schedule. Let's just change it all to
#- same number of trains as Saturday schedule for simplicity's sake

weekday = ['Mon', 'Tue', 'Wed', 'Thu', 'Fri']

# Change number of trains from weekday number to Saturday number (89)
data['Cars'][(data['Holiday'] == 1) & (data['Weekday'].isin(weekday))] = 89

#- JULY 4TH SCHEDULE
#- According to the WMATA website:
#- July 4th runs on regular holiday schedule from 7 AM-6 PM
#- And then on "near rush hour schedule" from 6 PM until midnight
#- I'm assuming near rush hour is 8 minutes per train
#- The math works out to 100 trains for July 4th

# Change number of trains for July 4th
data['Cars'][(data['Month'] == 7) & (data['Day'] == 4)] = 100

#- PRESIDENTIAL INAUGURATION
# In this dataset, there are three presidential inaguaration in this dataset.
# GW Bush - January 20, 2005; Obama - January20, 2009 and 2013
# According to the WMATA website, they run on a 4 AM - 9 PM rush hour
# And then regular holiday service after 9 PM
# Math works out to 190 cars
inaug_year = [2005, 2009, 2013]

# Change number of trains for Inauguration days
data['Cars'][(data['Year'].isin(inaug_year)) & (data['Month'] == 1) & (data['Day'] == 20)] = 190

#=========================#
# FINAL DATA MANIPULATION #
#=========================#

# Bikeshare did not exist before 2010 so will fill NaN values with 0
# for all the models
data['Registered'].fillna(value=0, inplace=True)
data['Casual'].fillna(value=0, inplace=True)
data.isnull().sum()  # check it worked

# Add binary variable of months into original data frame to use months as features
data = pd.concat([data, pd.get_dummies(data['Month'], prefix='Month')], axis=1)
data = pd.concat([data, pd.get_dummies(data['Weekday'])], axis=1)
data.columns.values  # Check it worked

# Create ridership per train variable
data['RidersPC'] = data['Riders']/data['Cars']
data.head()  # Check it worked

#=========================#
# EXPORT COMPILED DATASET #
#=========================#

data.to_csv('model_data.csv', index=False)

#=======================================================#
# SCATTER PLOTS
#=======================================================#

# Scatter matrix with weather data and Rider per Train
wdata = ['RidersPC','PRCP','SNWD','SNOW','TMAX','TMIN']
wscatter = data[wdata]

pd.scatter_matrix(wscatter)

#===============#
# WEEKDAY PLOTS #
#===============#

weekday = ['Mon', 'Tue', 'Wed', 'Thu', 'Fri']
weekday = data[data.Weekday.isin(weekday)]

# Scatter plot with Unemployment Rate
plt.scatter(weekday.Unemp_Rate, weekday.RidersPC, alpha=.8, color='r')
plt.xlabel("Unemployment Rate")
plt.ylabel("Riders per Train")
plt.show()
#- In general, the employment rate seems to not really have much of a relationship
#- with total number of riders

# Scatter plot with total number of people employed
plt.scatter(weekday.Employment, weekday.RidersPC, alpha=.8, color='b')
plt.xlabel("Total number of people employed (Number of person, in thousands)")
plt.ylabel("Riders per Train")
plt.show()
#- Slight upward trend but not a clear relationship

# Scatter plot with gas prices
plt.scatter(weekday.Gas_Price, weekday.RidersPC, alpha=.8, color='b')
plt.xlabel("Gas Price in Lower Atlantic Area")
plt.ylabel("Riders per Train")
plt.show()

# Scatter plot with number of trips taken by registered CaBi riders
plt.scatter(weekday.Registered, weekday.RidersPC, alpha=.8, color='r')
plt.xlabel("Total Number of Trips by Registered Riders")
plt.ylabel("Riders per Train")
plt.show()
# Does not appear to be a clear relationship

plt.scatter(weekday.Casual, weekday.RidersPC, alpha=.8, color='b')
plt.xlabel("Total Number of Trips by Casual Riders")
plt.ylabel("Riders per Train")
plt.show()
# Does not appear to be a clear relationship

# Bar graph of average holiday travel versus Regular day
data.groupby('Holiday').RidersPC.mean().plot(kind='bar',color='b')
plt.xlabel('Holiday')
plt.ylabel('Riders Per Train')
plt.show()
# Quite the difference in number of riders per train
# Same graph but for weekdays. Difference should be more pronounced.
weekday.groupby('Holiday').RidersPC.mean().plot(kind='bar',color='r')
plt.xlabel('Holiday')
plt.ylabel('Riders Per Train')
plt.show()