Merge pull request #10 from utorque/claude/tirex-neural-forecasting-01BcwXd3oLCprUTzmRvNMuV1 (968585cc) · Commits · Barthelet Thibault / horloml

forecast_app/app.py

+79 −0

Original line number	Diff line number	Diff line
		@@ -12,6 +12,7 @@ from datetime import datetime
		from io import BytesIO
		import pandas as pd
		import numpy as np
		import torch

		try:
		from openpyxl import Workbook
		@@ -20,6 +21,12 @@ try:
		except ImportError:
		EXCEL_AVAILABLE = False

		try:
		from tirex import load_model, ForecastModel
		TIREX_AVAILABLE = True
		except ImportError:
		TIREX_AVAILABLE = False

		app = Flask(__name__)
		app.secret_key = 'supply-chain-forecast-secret-key-2024'
		app.config['SESSION_TYPE'] = 'filesystem'
		@@ -29,6 +36,17 @@ DATA_DIR = os.path.join(os.path.dirname(__file__), 'data')
		TRAINING_DATA_PATH = os.path.join(DATA_DIR, 'sim2_supply_chain_data_training.json')
		TEST_DATA_PATH = os.path.join(DATA_DIR, 'sim2_supply_chain_data_test.json')

		# TiRex model (loaded lazily on first request)
		_tirex_model = None


		def get_tirex_model():
		"""Load TiRex model lazily (first time only)"""
		global _tirex_model
		if _tirex_model is None and TIREX_AVAILABLE:
		_tirex_model = load_model("NX-AI/TiRex", backend="torch")
		return _tirex_model


		def load_training_data():
		"""Load the 10-year training dataset"""
		@@ -572,6 +590,67 @@ def results():
		test_data=test_data)


		@app.route('/api/tirex_forecast', methods=['GET'])
		def tirex_forecast():
		"""Generate AI predictions using TiRex model"""
		if not TIREX_AVAILABLE:
		return jsonify({'error': 'TiRex library not available. Please install: pip install tirex-ts torch'}), 500

		try:
		# Load TiRex model (lazy loading)
		model = get_tirex_model()
		if model is None:
		return jsonify({'error': 'Failed to load TiRex model'}), 500

		# Load training data
		training_data = load_training_data()

		# Convert to DataFrame format for processing
		train_rows = []
		for month_data in training_data['historical_data']:
		for watch_data in month_data['watches']:
		train_rows.append({
		'year': month_data['year'],
		'month': month_data['month'],
		'watch_id': watch_data['watch_id'],
		'demand': watch_data['demand']
		})

		train_df = pd.DataFrame(train_rows)

		# Generate predictions for each watch
		predictions = {}

		for watch_id in train_df['watch_id'].unique():
		watch_data = train_df[train_df['watch_id'] == watch_id].sort_values(['year', 'month'])
		demands = watch_data['demand'].values

		# Convert to torch tensor (1 series, full history)
		context = torch.tensor(demands, dtype=torch.float32).unsqueeze(0)

		# Forecast 12 months ahead
		with torch.no_grad():
		quantiles, mean = model.forecast(context=context, prediction_length=12)

		# Extract mean predictions
		mean_preds = mean.squeeze().numpy()

		predictions[str(watch_id)] = {}
		for month_idx in range(12):
		pred = max(0, round(float(mean_preds[month_idx])))
		predictions[str(watch_id)][month_idx + 1] = pred

		return jsonify({
		'success': True,
		'predictions': predictions
		})

		except Exception as e:
		return jsonify({
		'error': f'Failed to generate predictions: {str(e)}'
		}), 500


		@app.route('/reset')
		def reset():
		"""Clear session and start over"""

forecast_app/requirements.txt

+3 −0

Original line number	Diff line number	Diff line
		@@ -2,3 +2,6 @@ Flask==3.0.0
		numpy==2.3.4
		Flask-Session==0.5.0
		openpyxl==3.1.2
		pandas==2.2.0
		torch==2.5.1
		tirex-ts

forecast_app/templates/predict.html

+68 −0

Original line number	Diff line number	Diff line
		@@ -84,6 +84,12 @@
		<span>Fill with Last Year</span>
		</button>
		</div>
		<div class="control">
		<button type="button" class="button is-primary" onclick="fillWithAI()" id="aiButton">
		<span class="icon"><i class="fas fa-brain"></i></span>
		<span>Fill with AI Model</span>
		</button>
		</div>
		</div>

		<hr>
		@@ -98,6 +104,14 @@
		</div>
		</div>
		</form>

		<!-- About TiRex AI Model Link -->
		<div class="has-text-centered" style="margin-top: 2rem; margin-bottom: 1rem;">
		<a href="https://github.com/NX-AI/tirex" target="_blank" class="button is-small is-light">
		<span class="icon"><i class="fas fa-info-circle"></i></span>
		<span>About TiRex AI Model</span>
		</a>
		</div>
		</div>
		</section>

		@@ -185,6 +199,60 @@ function fillFromYear10() {
		});
		}

		async function fillWithAI() {
		const button = document.getElementById('aiButton');
		const originalText = button.innerHTML;

		// Disable button and show loading state
		button.disabled = true;
		button.innerHTML = '<span class="icon"><i class="fas fa-spinner fa-pulse"></i></span><span>Loading AI Predictions...</span>';

		try {
		const response = await fetch('/api/tirex_forecast');
		const data = await response.json();

		if (!response.ok) {
		throw new Error(data.error \|\| 'Failed to fetch predictions');
		}

		if (data.success && data.predictions) {
		// Fill in the predictions
		watches.forEach(watch => {
		const inputs = getInputs(watch.id);
		const watchPredictions = data.predictions[watch.id.toString()];

		if (watchPredictions) {
		for (let month = 1; month <= 12; month++) {
		if (inputs[month - 1] && watchPredictions[month] !== undefined) {
		inputs[month - 1].value = watchPredictions[month];
		}
		}
		}
		});

		// Show success message
		button.innerHTML = '<span class="icon"><i class="fas fa-check"></i></span><span>AI Predictions Loaded!</span>';
		button.classList.remove('is-primary');
		button.classList.add('is-success');

		setTimeout(() => {
		button.innerHTML = originalText;
		button.classList.remove('is-success');
		button.classList.add('is-primary');
		button.disabled = false;
		}, 2000);
		} else {
		throw new Error('Invalid response format');
		}
		} catch (error) {
		console.error('Error fetching AI predictions:', error);
		alert('Failed to load AI predictions: ' + error.message + '\n\nPlease ensure TiRex is installed: pip install tirex torch');

		button.innerHTML = originalText;
		button.disabled = false;
		}
		}

		// Add paste functionality for each watch column
		watches.forEach((watch, watchIdx) => {
		const columnHeader = document.querySelector(`.watch-column-${watch.id}`);