single_preprocessing_use_case_function.py

#I'll just recylce the function we already built and slice the last observation

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
import datetime as dt
from sklearn.preprocessing import MinMaxScaler

def single_preprocessing_use_case(ticker='SPLK', drop_col='Close', delay=15, lag=0):
    df = pd.read_pickle("./data/data_pkls/nasdaq100.pkl") #read data
    #Filter dataframe by this specific ticker
    df = df[ticker]
    #Drop missing values since there is no good way to estimate stock values that are not tracked
    df.dropna(inplace=True)
    #Drop Close column since it Adj Close column is calcualted using this information
    #so it would be allowing the model to cheat by already knowing the answers
    df.drop(columns={drop_col}, inplace=True)

    if drop_col == 'Close':
        organize_col = 'Adj Close'
    else:
        organize_col = 'Close'
    #Now just organize the columns 
    df = df.loc[:, ['Open', 'High', 'Low', 'Volume', organize_col]]
    #Now let's scale our data
    #First let's split data into X and Y
    X = df.loc[:, ['Open', 'High', 'Low', 'Volume']]
    y = df['Adj Close']
    scaler = MinMaxScaler() #create scaler object
    scaled_data = scaler.fit_transform(X) #fit transform data
    #Split data into samples and reshape X
    samples = [ scaled_data[i:i+delay] for i in range(0, (y.shape[0]), delay)]
    if len(samples[-1] < delay):
        samples = np.array(samples[:-1])
    else:
        samples = np.array(samples)
    #Reshape y target also
    new_y = [ y[i+delay+lag] for i in range(0, (y.shape[0]-(delay+lag)), delay)]
    new_y = np.array(new_y)

    #slice the last observation for this ticker
    samples = samples[-1]
    new_y = new_y[-1]
    return samples.reshape(1, delay, 4), new_y, ticker, delay, lag