Spaces:
Sleeping
Sleeping
File size: 8,681 Bytes
9ffa007 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 | """
Project: Optimizing Stock Trading Strategy With Reinforcement Learning
Authors: Amey Thakur & Mega Satish
Reference: https://github.com/Amey-Thakur/OPTIMIZING-STOCK-TRADING-STRATEGY-WITH-REINFORCEMENT-LEARNING
License: MIT
Description:
This script contains the Main Application logic served via Streamlit.
It loads the pre-trained Q-Learning model (model.pkl), processes user-selected
stock data, simulates the trading strategy on unseen data, and visualizes
the portfolio performance using interactive Plotly charts.
"""
import numpy as np
import pandas as pd
from pandas._libs.missing import NA
import streamlit as st
import time
import plotly.graph_objects as go
import pickle as pkl
# ==========================================
# 1. Data Processing Logic
# ==========================================
# @st.cache(persist=True)
def data_prep(data, name):
"""
Prepares the dataset for the selected stock ticker.
Args:
data (pd.DataFrame): The raw dataset.
name (str): The specific stock name selected by the user.
Returns:
pd.DataFrame: A clean dataframe with computed Moving Averages (5-day & 1-day).
"""
df = pd.DataFrame(data[data['Name'] == name])
df.dropna(inplace=True)
df.reset_index(drop=True, inplace=True)
# Calculate Moving Averages (Technical Indicators)
# These indicators form the basis of the State Space for the RL agent.
df['5day_MA'] = df['close'].rolling(5).mean()
df['1day_MA'] = df['close'].rolling(1).mean()
# Handle initial NaN values
df.loc[:4, '5day_MA'] = 0
return df
# ==========================================
# 2. Agent Logic (Inference)
# ==========================================
# @st.cache(persist=True)
def get_state(long_ma, short_ma, t):
"""
Determines the current state of the market based on MA crossovers.
Returns a tuple (Trend, Position) matching the Q-Table structure used during training.
"""
if short_ma < long_ma:
if t == 1:
return (0, 1) # Bearish, Cash
else:
return (0, 0) # Bearish, Stock
elif short_ma > long_ma:
if t == 1:
return (1, 1) # Bullish, Cash
else:
return (1, 0) # Bullish, Stock
return (0, 1) # Default
# @st.cache(persist=True)
def trade_t(num_of_stocks, port_value, current_price):
"""
Checks if a trade (Buy) is financially feasible.
"""
if num_of_stocks >= 0:
if port_value > current_price:
return 1 # Can Buy
else: return 0
else:
if port_value > current_price:
return 1
else: return 0
# @st.cache(persist=True)
def next_act(state, qtable, epsilon, action=3):
"""
Decides the next action based on the trained Q-Table.
During inference (testing), epsilon is typically 0 (pure exploitation),
meaning the agent always chooses the optimal action learned during training.
"""
if np.random.rand() < epsilon:
action = np.random.randint(action)
else:
action = np.argmax(qtable[state])
return action
# @st.cache(persist=True)
def test_stock(stocks_test, q_table, invest):
"""
Runs a simulation of the trading strategy on the selected stock.
Args:
stocks_test (pd.DataFrame): The stock data to test on.
q_table (np.array): The loaded reinforcement learning model.
invest (int): Initial investment amount.
Returns:
list: A time-series list of net worth values over the simulation period.
"""
num_stocks = 0
epsilon = 0 # No exploration during testing/inference
net_worth = [invest]
np.random.seed()
for dt in range(len(stocks_test)):
long_ma = stocks_test.iloc[dt]['5day_MA']
short_ma = stocks_test.iloc[dt]['1day_MA']
close_price = stocks_test.iloc[dt]['close']
# Determine Current State
t = trade_t(num_stocks, net_worth[-1], close_price)
state = get_state(long_ma, short_ma, t)
# Agent chooses action
action = next_act(state, q_table, epsilon)
if action == 0: # Buy
num_stocks += 1
to_append = net_worth[-1] - close_price
net_worth.append(np.round(to_append, 1))
elif action == 1: # Sell
num_stocks -= 1
to_append = net_worth[-1] + close_price
net_worth.append(np.round(to_append, 1))
elif action == 2: # Hold
to_append = net_worth[-1] + close_price # Mark-to-market valuation
net_worth.append(np.round(to_append, 1))
# Check for next state existence
try:
next_state = get_state(stocks_test.iloc[dt+1]['5day_MA'], stocks_test.iloc[dt+1]['1day_MA'], t)
except:
break
return net_worth
# ==========================================
# 3. Streamlit Interface
# ==========================================
def fun():
# Reading the Dataset
# Ensure all_stocks_5yr.csv is in the working directory
data = pd.read_csv('all_stocks_5yr.csv')
names = list(data['Name'].unique())
names.insert(0, "<Select Names>")
st.title("Optimizing Stock Trading Strategy With Reinforcement Learning")
st.sidebar.title("Choose Stock and Investment")
st.sidebar.subheader("Choose Company Stocks")
# User Input: Select Stock
stock = st.sidebar.selectbox("(*select one stock only)", names, index=0)
if stock != "<Select Names>":
stock_df = data_prep(data, stock)
# Sidebar Checkbox: Plot Data Trend
if st.sidebar.button("Show Stock Trend", key=1):
fig = go.Figure()
fig.add_trace(go.Scatter(
x=stock_df['date'],
y=stock_df['close'],
mode='lines',
name='Stock_Trend',
line=dict(color='cyan', width=2)
))
fig.update_layout(
title='Stock Trend of ' + stock,
xaxis_title='Date',
yaxis_title='Price ($) '
)
st.plotly_chart(fig, use_container_width=True)
# Simple heuristic for trend feedback
if stock_df.iloc[500]['close'] > stock_df.iloc[0]['close']:
original_title = '<b><p style="font-family:Play; color:Cyan; font-size: 20px;">NOTE:<br>Stock is on a solid upward trend. Investing here might be profitable.</p>'
st.markdown(original_title, unsafe_allow_html=True)
else:
original_title = '<b><p style="font-family:Play; color:Red; font-size: 20px;">NOTE:<br> Stock does not appear to be in a solid uptrend. Better not to invest here; instead, pick different stock.</p>'
st.markdown(original_title, unsafe_allow_html=True)
# Sidebar Checkbox: Investment Simulation
st.sidebar.subheader("Enter Your Available Initial Investment Fund")
invest = st.sidebar.slider('Select a range of values', 1000, 1000000)
if st.sidebar.button("Calculate", key=2):
# Load Pre-trained Model
try:
# Using 'model.pkl' as standardized
q_table = pkl.load(open('model.pkl', 'rb'))
except FileNotFoundError:
st.error("Model file 'model.pkl' not found. Please ensure the model is trained.")
return
# Run Simulation
net_worth = test_stock(stock_df, q_table, invest)
net_worth = pd.DataFrame(net_worth, columns=['value'])
# Plot Results
fig = go.Figure()
fig.add_trace(go.Scatter(
x=net_worth.index,
y=net_worth['value'],
mode='lines',
name='Net_Worth_Trend',
line=dict(color='cyan', width=2)
))
fig.update_layout(
title='Change in Portfolio Value Day by Day',
xaxis_title='Number of Days since Feb 2013 ',
yaxis_title='Value ($) '
)
st.plotly_chart(fig, use_container_width=True)
original_title = '<b><p style="font-family:Play; color:Cyan; font-size: 20px;">NOTE:<br> Increase in your net worth as a result of a model decision.</p>'
st.markdown(original_title, unsafe_allow_html=True)
if __name__ == '__main__':
fun()
# Dummy chart for layout purposes if needed, otherwise optional
# chart_data = pd.DataFrame(np.random.randn(20, 3), columns=['a', 'b', 'c'])
|