# TD Ameritrade API Historical Data Linear Regression

I’m not sure if anyone is interested in the stock market on Codecademy but I used the TD Ameritrade API to get historical data and applied the linear regression model on it that I just learned in the Learn the basics of Machine Learning course. It utilizes pandas and the plotly charting library to visualize the data. Pandas and Plotly must be installed for this to work. You also need a developer account on TD Ameritrade Developer which is free. You would then create an app on there and that’s where you get the apikey. This API is great and free, it’s just not documented too effectively. The next thing I want to do is see if I can apply this to the streaming data API but that’s mostly documented in JS and I prefer Python. In the linear regression function I’ve been running different learning rates and number of iterations to find the sweet spot. I’m trying to figure out what would be good numbers for 10 day 5 minute data. It’s pretty resource intensive and takes a few minutes to run and may not give you any results if the numbers get too large.

``````import requests
import pandas as pd
import plotly.graph_objects as go
from config import apiKey
import datetime

class TestApp:
def __init__(self):
self.data = None
self.df = None
self.df2 = None
self.y = None

def get_historical_price_data(self, symbol, periodType, period, frequencyType, frequency, endDate=None, startDate=None, needExtendedHoursData=False):

parameters = {
'apikey': apiKey,
'periodType': periodType, # day, month, year, ytd
'period': period,   # day: 1, 2, 3, 4, 5, 10*
# month: 1*, 2, 3, 6
# year: 1*, 2, 3, 5, 10, 15, 20
# ytd: 1*
'frequencyType': frequencyType, # day: minute*
# month: daily, weekly*
# year: daily, weekly, monthly*
# ytd: daily, weekly*
'frequency': frequency, # minute: 1*, 5, 10, 15, 30
# daily: 1*
# weekly: 1*
# monthly: 1*
'needExtendedHoursData': needExtendedHoursData
}

epoch = datetime.datetime.utcfromtimestamp(0)

def unix_time_milliseconds(datetime):
return (datetime - epoch).total_seconds() * 1000

# If endDate and startDate are provided period will be removed from parameters as required by api
if endDate and startDate:
parameters.pop('period')
parameters.update({'endDate': int(unix_time_milliseconds(datetime.datetime.strptime(endDate, '%Y%m%d %H:%M:%S'))), 'startDate': int(unix_time_milliseconds(datetime.datetime.strptime(startDate, '%Y%m%d %H:%M:%S')))})

self.data = requests.get(endpoint, params=parameters).json()

self.df = pd.DataFrame(self.data['candles'], columns=['datetime', 'open', 'high', 'low', 'close', 'volume'])

self.df2 = self.df

return self.data

def show_candlestick_chart(self):
self.df.iloc[:, 0] = pd.to_datetime(self.df['datetime'], unit='ms').dt.tz_localize('UTC').dt.tz_convert('US/Eastern')
fig = go.Figure(
[
go.Candlestick(x=self.df['datetime'], open=self.df['open'], high=self.df['high'], low=self.df['low'], close=self.df['close']),
go.Scatter(x=self.df['datetime'], y = self.y)
]
)
fig.update_xaxes(rangebreaks=[dict(bounds=[16, 9.5], pattern='hour'),
dict(bounds=['sat', 'mon'])])
fig.show()

def linear_regression(self, learning, iterations):
# 1 & 2 Day Five Minute - learning_rate = 0.0001, num_iterations = 500000
# 3 to 5 Day Five Minute - learning_rate = 0.00001, num_iterations = 1000000

# list of x values - ms since epoch
datetime_list = self.df['datetime'].to_list()
new_x = [x for x in range(self.df['datetime'].size)]
# list of y values - closing price
closing_prices_list = self.df['close'].to_list()

# m is the current gradient guess
# b is the current intercept guess
diff = 0
# gradient of loss as intercept changes = -(2 / number of points) * sum of (y point - (current gradient guess * x point + intercept guess))
for x, closing_price in zip(new_x, closing_prices_list):
diff += (closing_price - (m * x + b))
b_gradient = (-1 * (2 / len(new_x)) * diff)

diff = 0
for x, closing_price in zip(new_x, closing_prices_list):
diff += (x * (closing_price - (m * x + b)))
m_gradient = (-1 * (2 / len(new_x)) * diff)

b = (b_current - (learning_rate * b_gradient))
m = (m_current - (learning_rate * m_gradient))
return [b, m]

b = 0
m = 0
for iteration in range(num_iterations):
b, m = step_gradient(b, m, learning_rate)
return [b, m]

self.y = [m * x + b for x in new_x]

test = TestApp()

test.get_historical_price_data('SPY', 'day', 10, 'minute', 5)
# print(test.df)
test.linear_regression(0.000001, 5000000)
test.show_candlestick_chart()
print(test.y)

``````

I don’t really understand python but this api is so cool. Been wanting to make a stock scanner to look for covered call opportunities and my sticking point was finding good data that wasn’t too difficult for my beginner skill set exclusively knowing JS for now. This will make my project possible as I already trade through TD Ameritrade and just need to get a developer account. This is dope