{
"cells": [
{
"cell_type": "markdown",
"id": "9ddfd4df",
"metadata": {},
"source": [
"# Neuro-Symbolic AI Forecasting — Symbolic Seam\n",
"**Research Prototype: Constraint-Aware Forecasting with Neural + Symbolic Integration**\n",
"\n",
"## Research Objective\n",
"\n",
"Demonstrate how symbolic constraints (knowledge rules) can reduce unrealistic forecasts from a neural model under market stress, while documenting the accuracy tradeoff.**Pipeline:** Data → Knowledge Graph → Neural Baseline (System 1) → Domain Constraints (System 2) → Symbolic Projection → Stress Test → Evaluation\n",
"\n",
"\n",
"**Key Finding:** The symbolic layer achieves 95% violation reduction under stress but at the cost of slightly higher RMSE on normal data. This is a safety-first design, not an accuracy-first design."
]
},
{
"cell_type": "markdown",
"id": "c4882cfc",
"metadata": {},
"source": [
"## 1. Setup & GPU Config"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "2d14c2df",
"metadata": {
"execution": {
"iopub.execute_input": "2026-07-03T20:14:02.426722Z",
"iopub.status.busy": "2026-07-03T20:14:02.426341Z",
"iopub.status.idle": "2026-07-03T20:14:12.193955Z",
"shell.execute_reply": "2026-07-03T20:14:12.193310Z",
"shell.execute_reply.started": "2026-07-03T20:14:02.426693Z"
},
"trusted": true
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Device: cuda | GPUs available: 2\n"
]
}
],
"source": [
"%pip install yfinance cvxpy -q\n",
"\n",
"import torch, torch.nn as nn\n",
"import numpy as np, pandas as pd\n",
"import matplotlib.pyplot as plt\n",
"import cvxpy as cp\n",
"import warnings, os\n",
"warnings.filterwarnings(\"ignore\")\n",
"\n",
"# Use both GPUs if available (Kaggle T4 x2)\n",
"device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
"n_gpus = torch.cuda.device_count()\n",
"print(f\"Device: {device} | GPUs available: {n_gpus}\")\n",
"\n",
"SEED = 42\n",
"np.random.seed(SEED); torch.manual_seed(SEED)\n",
"\n",
"os.makedirs(\"data\", exist_ok=True)\n",
"os.makedirs(\"models\", exist_ok=True)\n",
"os.makedirs(\"results\", exist_ok=True)"
]
},
{
"cell_type": "markdown",
"id": "68d552da",
"metadata": {},
"source": [
"## 2. Data Exploration (Notebook 01)"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "69738e6b",
"metadata": {
"execution": {
"iopub.execute_input": "2026-07-03T20:14:18.467620Z",
"iopub.status.busy": "2026-07-03T20:14:18.467026Z",
"iopub.status.idle": "2026-07-03T20:14:19.898040Z",
"shell.execute_reply": "2026-07-03T20:14:19.897301Z",
"shell.execute_reply.started": "2026-07-03T20:14:18.467589Z"
},
"trusted": true
},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"[*********************100%***********************] 1 of 1 completed"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"(1006, 5)\n"
]
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"\n"
]
},
{
"data": {
"text/html": [
"\n",
"\n",
"
\n",
" \n",
" \n",
" | Price | \n",
" Open | \n",
" High | \n",
" Low | \n",
" Close | \n",
" Volume | \n",
"
\n",
" \n",
" | Ticker | \n",
" AAPL | \n",
" AAPL | \n",
" AAPL | \n",
" AAPL | \n",
" AAPL | \n",
"
\n",
" \n",
" | Date | \n",
" | \n",
" | \n",
" | \n",
" | \n",
" | \n",
"
\n",
" \n",
" \n",
" \n",
" | 2023-12-22 | \n",
" 192.995392 | \n",
" 193.222829 | \n",
" 190.810137 | \n",
" 191.433090 | \n",
" 37149600 | \n",
"
\n",
" \n",
" | 2023-12-26 | \n",
" 191.442997 | \n",
" 191.719862 | \n",
" 190.671728 | \n",
" 190.889267 | \n",
" 28919300 | \n",
"
\n",
" \n",
" | 2023-12-27 | \n",
" 190.335512 | \n",
" 191.334202 | \n",
" 188.951173 | \n",
" 190.988113 | \n",
" 48087700 | \n",
"
\n",
" \n",
" | 2023-12-28 | \n",
" 191.967044 | \n",
" 192.481228 | \n",
" 191.007900 | \n",
" 191.413315 | \n",
" 34049900 | \n",
"
\n",
" \n",
" | 2023-12-29 | \n",
" 191.729722 | \n",
" 192.224126 | \n",
" 189.584012 | \n",
" 190.375061 | \n",
" 42672100 | \n",
"
\n",
" \n",
"
\n",
"