Bokeh Application Using Quandl

Bokeh dashboards using Quandl datasets¶

Though I am not from financial industry, I can understand the need to track how precious commodities and stocks are performing over years. So I started looking into commodity/stock datasets available for creating useful charts/dashboard. One of my colleague told me about Quandl which has datasets realted to finance and economy sectors. So I set out to create a bokeh application using quandl data and this article outlines the steps performed to achieve it. In order to demonstrate how the app is built, I am going to choose a couple of stocks and commodities. You can extend it other quandl datasets.

About Quandl¶

You need to sign up for Quandl account which allows you to download free data and samples of premium data. Once you have signed up, you will be provided with API key and you can find the API key under account settings. To know more about API limits for a free account click here

You can use Quandl API in R, Python and through Excel plugin. I went for Python as Quandl has a package that you can install using pip. I won't go into the details as it is well documented by quandl's team.

Data Processing¶

main.py¶

Data can be accessed by calling the quandl api using the following code:

quandl.ApiConfig.api_key = "GIVE YOUR API KEY HERE"

Once you have set the api key you can quandl datatables using .get method

data1 = quandl.get("LBMA/SILVER", collapse="monthly")
data2 = quandl.get("LBMA/GOLD", collapse="monthly")

Importing tab_commodity() function from commodity_select.py

from scripts.commodity_select import tab_commodity

We will be doing a bit a data processing to make sure data is consistent across commodities. Some data processing steps performed are

Convert the output from quandl api get call into pandas dataframe

df1 = pd.DataFrame(data1)
df2 = pd.DataFrame(data2)

Fill NA values with 0

df1 = df1.fillna(0)
df2 = df2.fillna(0)

Resetting the index on the dataframe

df1 = df1.reset_index()
df2 = df2.reset_index()

Converting the Date column to datetime object

df1['Date'] = pd.to_datetime(df1['Date'])
df1['Date'] = pd.DatetimeIndex(df1['Date'])

df2['Date'] = pd.to_datetime(df2['Date'])
df2['Date'] = pd.DatetimeIndex(df2['Date'])

Selecting a subset of dataframe

df1 = df1[['Date','USD']]
df2 = df2[['Date','USD (PM)']]

Renaming the column in df2 and assigning new column to each dataframe

df2.rename(columns={'USD (PM)': 'USD'}, inplace=True)
df1['Commodity']='SILVER'
df2['Commodity']='GOLD'

Combining the dataframes

frames = [df1,df2]
df_commodity = pd.concat(frames)

Also lets create a list of commodities that will be passed to the function in commodity_select.py

tab2 = tab_commodity(commodities,df_commodity)

Tab1 - Precious Commodities¶

commodity_select.py¶

Here is the commodity_select.py which is used to created Commodity Trend tab.

First step is to import the libraries required for passing data to the plot and dropdown select

import pandas as pd
from bokeh.layouts import column
from bokeh.models import ColumnDataSource, Select, Panel, HoverTool
from bokeh.plotting import figure

Now lets define the function tab_commodity that will be called from main.py

def tab_commodity(commodities, df_commodity):
    p = figure(plot_width=960, plot_height=540, title="Commodity Trend",x_axis_type="datetime",sizing_mode="scale_width")
    p.background_fill_color="#f5f5f5"
    p.background_fill_alpha = 0.5
    p.grid.grid_line_color="white"
    p.xaxis.axis_label = 'Date'
    p.yaxis.axis_label = 'Value in USD'
    p.axis.axis_line_color = None

    hover = HoverTool()

    select_commodity = Select(title='Commodity:',value=commodities[0],options=commodities)
    x_val = df_commodity.loc[df_commodity['Commodity'] == select_commodity.value].Date.tolist()
    y_val = df_commodity.loc[df_commodity['Commodity'] == select_commodity.value].USD.tolist()
    source = ColumnDataSource(data=(dict(x=x_val,y=y_val)))

    p.line('x','y',source=source,color='#ebbd5b',alpha=0.5,line_width=2)

    hover.tooltips = [("Date", "@x{%F}"),("Value", "$@{y}")]
    hover.formatters={
        'x' : 'datetime'
        }
    hover.mode = 'vline'
    p.tools.append(hover)

    def callback(attr,old,new):
        new_commodity = select_commodity.value
        x_new = df_commodity.loc[df_commodity['Commodity'] == new_commodity].Date.tolist()
        y_new = df_commodity.loc[df_commodity['Commodity'] == new_commodity].USD.tolist()
        source_new = ColumnDataSource(data=(dict(x=x_new,y=y_new)))
        source.data = source_new.data

    select_commodity.on_change('value',callback)
    layout = column(select_commodity,p)
    tab2 = Panel(child = layout, title = 'Commodity Performance')

    return tab2

Tab2 - Stocks¶

main.py¶

Quandl has data on stocks which can be used to create dashboards that are useful for financial institutions. To get the data from quandl, we are going to use get_table method. Below is the code block for doing it

from scripts.Stocks_Select import stock_tab

tickerlist = ['AAPL','IBM','MSFT','GOOG']

df = quandl.get_table('WIKI/PRICES', qopts = { 'columns': ['ticker','date', 'close'] }, 
                       ticker = tickerlist, date = { 'gte': '2016-01-01', 'lte': '2016-12-31' })

df = df[['ticker','date','close']]
df['date'] = pd.DatetimeIndex(df['date'])

Calling the stock_tab function from Stocks_Select.py

tab1 = stock_tab(tickerlist, df)

Stocks_Select.py¶

import pandas as pd
from bokeh.layouts import column
from bokeh.models import ColumnDataSource, Select, Panel, HoverTool
from bokeh.plotting import figure

def stock_tab(tickerlist, df):
    p = figure(plot_width=960, plot_height=540, x_axis_type="datetime",sizing_mode="scale_width")
    p.background_fill_color="#f5f5f5"
    p.background_fill_alpha = 0.5
    p.grid.grid_line_color="white"
    p.xaxis.axis_label = 'Date'
    p.yaxis.axis_label = 'Value in USD'
    p.axis.axis_line_color = None

    hover = HoverTool()
    hover.tooltips = [("Date", "@x{%F}"),("Value", "$@{y}")]
    hover.formatters={'x':'datetime'}
    hover.mode = 'vline'
    p.tools.append(hover)

    def callback(attr,old,new):
        ticker_new = ticker.value
        x = df.loc[df['ticker'] == ticker_new].date.tolist()
        y = df.loc[df['ticker'] == ticker_new].close.tolist()
        source_new = ColumnDataSource(data=dict(x=x,y=y))
        source.data = source_new.data
        ticker.value = ticker_new

    ticker = Select(title="Ticker:",value=tickerlist[0],options=tickerlist)
    x = df.loc[df['ticker'] == ticker.value].date.tolist()
    y = df.loc[df['ticker'] == ticker.value].close.tolist()
    ticker.on_change('value',callback)
    source = ColumnDataSource(data=dict(x=x, y=y))
    p.line('x','y', line_width=2, color='navy', alpha=0.5,source=source)
    layout = column(ticker, p)
    tab = Panel(child = layout, title = 'Stock Performance')
    return tab

Final steps¶

Combine tab1 and tab2

tabs = Tabs(tabs = [tab1,tab2])

Pass tabs to curdoc

curdoc().add_root(tabs)

All the code blocks above should be added to main.py. The directory format for the project should follow the structure outlined in the link here.

The final output looks like

In [24]:

from IPython.display import HTML

HTML('<iframe width="560" height="315" src="https://www.youtube.com/embed/jj_GTBJcNr0" frameborder="0" allow="accelerometer; encrypted-media; gyroscope" allowfullscreen></iframe>')

Out[24]:

links

social