Visualization Template: Double-axis Time Series Plot with Auxiliary Lines/Bands

viz1_ts_plot.py

# Matplotlib version: 3.5.2 (My Local Jupyter Notebook)
# Seaborn version: 0.11.2 (My Local Jupyter Notebook)
# Python version: 3.7.4 (My Local Jupyter Notebook)

import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
import matplotlib.ticker as ticker 
from datetime import datetime
%matplotlib inline
sns.set(style="white",context="talk")

def double_axis_line_plot(df_1, col_x, col_1_y, line_1_label, df_2, col_2_y, line_2_label, title='TITLE', xlabel = 'XLABEL', y_1_label = 'LEFT_YLABEL',y_2_label = 'RIGHT_YLABEL', figsize=(14, 6)):
    '''
    This lineplot function visualizes the (time series) trend of key metrics col_1_y versus col_2_y over col_x.
    col_1_y corresponds to the left y-axis and col_2_y is ties to the right y-axis.
    I also keep the alternative to add a shaded area / auxiliary lines to the graph.
    
    Input:
        df_1: pandas dataframe. The dataframe for the first variable.
        col_x: string. The name of the column in df_1 as the x-axis
        col_1_y: string. The name of the column in df_1 as the left x-axis 
        line_1_label: string. The label for col_1_y that appears as legends
        
        df_2: pandas dataframe. The dataframe for the second variable.
        col_2_y: string. The name of the column in df_2 as the right x-axis 
        line_2_label: string. The label for col_2_y that appears as legends  
        
        title: string. The title of the diagram
        xlabel: string. The name of the x-axis
        y_1_label: string. The title of the left y-axis
        y_2_label: string. The title of the right y-axis
        figsize: tuple. The width and height of the figure.
    Output:
        fig: figure object
    '''
    fig, ax = plt.subplots(nrows=1, ncols=1, figsize = figsize)
    
    # Lineplot for the first variable (left axis)
    sns.lineplot(
        x = col_x, y = col_1_y,
        data = df_1, ax = ax,
        color = 'royalblue',
        label = line_1_label,
        # marker = "o", markersize = 6,
    )
    ax.set_xlabel(xlabel,fontsize = 16)
    ax.set_ylabel(y_1_label,fontsize = 16)
    # Set the legend location and fontsize
    ax.legend(loc=2,fontsize=14)
    # Display 2 times fewer ticks on the x-axis to avoid the overlapping mess
    ax.xaxis.set_major_locator(ticker.MultipleLocator(2))
    # Rotate the x-ticks to save some space
    plt.setp(ax.get_xticklabels(), fontsize = 14, rotation=45, ha="right", rotation_mode="anchor")
    
    # Lineplot for the second variable (right axis)
    ax_y = ax.twinx()
    sns.lineplot(
        x = col_x, 
        y = col_2_y, 
        data = df_2,
        ax = ax_y, 
        label = line_2_label,
        color = 'black',
        alpha = 0.4,
        # linestyle = 'dashed',
        # marker='o',markersize= 6,
    )
    # Display 2 times fewer ticks on the x-axis to avoid the overlapping mess
    ax_y.xaxis.set_major_locator(ticker.MultipleLocator(2))
    # Rotate the x-ticks to save some space
    plt.setp(ax_y.get_xticklabels(), fontsize = 14, rotation=45, ha="right", rotation_mode="anchor")
    ax_y.set_ylabel(y_2_label,fontsize = 16)
    # Make it a dashed lineplot (https://stackoverflow.com/questions/51963725/how-to-plot-a-dashed-line-on-seaborn-lineplot) 
    ax_y.lines[0].set_linestyle("--")
    # Set the legend location and fontsize
    ax_y.legend(loc=1,fontsize = 14)

    # Add auxiliary lines and auxiliary shaded area (if needed)
    ax.axvspan(
        xmin = datetime.strptime('2019/03/20', '%Y/%m/%d'), 
        xmax = datetime.strptime('2019/03/23', '%Y/%m/%d'), 
        color = 'darkgrey',
        alpha=0.2)
    # ax.axhline(y = 0.5, ls = "--", c = "black",alpha = 0.6) #
    # ax.axvline(x = datetime.strptime('2019/09/28', '%Y/%m/%d'), ls = "--", c = "black",alpha = 0.6) 
    
    # Format the graph-level features 
    ax.set_title(title, fontsize=18, pad=10) # pad paramter increases the space between title and graph (https://stackoverflow.com/questions/16419670/increase-distance-between-title-and-plot-in-matplolib)
    # Faint the grid lines
    ax.grid(linestyle="--", alpha=0.4)
    # ax.tick_params(top=False)    
    plt.show()
    
    return fig

# ====================================================================================
# Illustration
# Load the data
# The first dataset: NYC public taxis records from seaborn in March, 2019
df = sns.load_dataset("taxis")
df['pickup'] = list(map(lambda x:datetime.strptime(x, '%Y-%m-%d %H:%M:%S'),df.pickup))
df['date'] = list(map(lambda x:x.date(),df.pickup))

taxis = df.groupby(by='date').agg(
    num_rides = pd.NamedAgg(column='pickup', aggfunc='count'),
    avg_price = pd.NamedAgg(column='total', aggfunc='mean'),
    max_price = pd.NamedAgg(column='total', aggfunc='max'),
    med_price = pd.NamedAgg(column='total', aggfunc=np.median),
    avg_distance = pd.NamedAgg(column='distance', aggfunc='mean')
).sort_values(by='date').reset_index()
taxis.date = pd.to_datetime(taxis.date)
taxis = taxis[1:] # Remove the abnormal date - Feb 28th

# The second dataset: NYC public weather data from National Oceanic and Atmospheric Administration in March, 2019
weather = pd.read_csv("https://raw.githubusercontent.com/mintaow/MyMediumWork/main/data/climate_data_nyc_noaa.csv")
weather.date = pd.to_datetime(weather.date)


fig = double_axis_line_plot(
    df_1 = taxis, 
    col_x = 'date', 
    col_1_y = 'num_rides', 
    line_1_label = 'Number of Taxis Rides in NYC', 
    df_2 = weather, 
    col_2_y = 'precip', 
    line_2_label = 'Precipitation (Inch)', 
    title='Understanding Demand Fluctuation: Daily Number of Taxis Rides in NYC', 
    xlabel = 'Date (March, 2019)', 
    y_1_label = '# Rides',
    y_2_label = 'Inch', 
    figsize=(12, 6)
)

# fig.savefig(
#     fname = "../double_axis_time_series_plot_with_band_taxis.png", # path&filename for the output
#     dsi = 300, # make it a high-resolution graph
#     bbox_inches='tight' # sometimes default savefig method cuts off the x-axis and x-ticks. This param avoids that 
# )