Skip to content

Latest commit

 

History

History
96 lines (72 loc) · 3.9 KB

README.md

File metadata and controls

96 lines (72 loc) · 3.9 KB

FML

FML - Arizona Zervas (10/10 would recommend)

Running Instructions

Two terminals are required to run this program. To run the frontend, first, in frontend/src/environments, create a duplicate of the file environment.prod.ts in the same directory, but rename the duplicate to environment.ts. Next, cd into the "frontend" directory and run ng serve. To run the backend, cd into the backend and run the run.py python script with the argument "run" as such: python run.py run. Prior to running the script, ensure that all necessary dependencies have been installed by running pip install -r requirements.txt through pip. Please note that this project was made to be compatible with Python 3.

Once both the frontend and backend are running, navigate to localhost:4200 on your browser (preferably Chrome) to access the app.

ML

LSTM

Overview

Using Long-Short Term Memory, we are training a model to see a large set of data of both massaged prices and other indicies.

Creating Data for LSTM

from gen_lstm_data import gen_data

eq = 'VSLR' ## Ticker of the stock you want to look at, be sure you have added this to the data/equities folder

train_split = 0.8 ## Percent of data to use for training data, rest to test data

days = 500 ## How many days you want to be using of the data. As of now, there is no check on size, so if you put too large of a number, you may get index OB

look_back = 19 ## How many days you want to use to predict the next one. This will be the second dimension of your X data

label_range = 5 ## How many days you want to look at price change over. This will determine how the labels are calculated

verbose = False ## Prints out some shapes and other relevant things if you are curious.

X_train, y_train, X_test, y_test = gen_data(eq, train_split, days, look_back, label_range, verbose)

## X_train.shape = (days * train_split - (look_back + 1), look_back, 25)

## y_test.shape = (days * train_split - (look_back + 1), 10)

## X_test.shape = (days * (1 - train_split) - (look_back + 1), look_back, 25)

## y_test.shape = (days * (1 - train_split) - (look_back + 1), 10)

Features

Vector Description
Vector 0 Volumes
Vector 1 Prices
Vector 2 SMA
Vector 3 EMA
Vector 4 Wilder MA
Vector 5 Upper Bolinger Band
Vector 6 Lower Bolinger Band
Vector 7 Accumulative Swing Index
Vector 8 Average True Range
Vector 9 Balance of Power
Vector 10 Gopalakrishnan Range Index
Vector 11 Price - Pivot Point
Vector 12 Pring's Know Sure Thing - SMA(Pring's Know Sure Thing)
Vector 13 MACD - SMA(MACD)
Vector 14 d KST * d TRIX
Vector 15 TRIX - MA(TRIX)
Vector 16 RSI
Vector 17 MA(OHLC/4, 1)
Vector 18 MA(OHLC/4, 3)
Vector 19 MA(OHLC/4, 5)
Vector 20 MA(OHLC/4, 7)
Vector 21 MA(OHLC/4, 9)
Vector 22 West Texas
Vector 23 Wilshire US Real Estate
Vector 24 SNP

Labels

Label
-10% <= (p_1 - p_0)/p_0 < - 7%
- 7% <= (p_1 - p_0)/p_0 < - 5%
- 5% <= (p_1 - p_0)/p_0 < - 3%
- 3% <= (p_1 - p_0)/p_0 < - 1%
- 1% <= (p_1 - p_0)/p_0 < 1%
1% <= (p_1 - p_0)/p_0 < 3%
3% <= (p_1 - p_0)/p_0 < 5%
5% <= (p_1 - p_0)/p_0 < 7%
7% <= (p_1 - p_0)/p_0 < 9%
9% <= (p_1 - p_0)/p_0 < 100%

Where p_1 is the price at the predicted day and p_0 is the price at predicted day - label_range

Signal Distributions

Overview

This algorithm takes in a series of buy/sell signals from a variety of market indicators, determines how to combine these signals into an overall buy/sell signal and then compares the strength of these signals across equities.