Skip to content

EXAMPLES.md

Latest commit

 

History

History
199 lines (150 loc) · 5.34 KB

EXAMPLES.md

File metadata and controls

199 lines (150 loc) · 5.34 KB

Example Workflows

This directory contains ready-to-use n8n workflows demonstrating FAIM forecast node capabilities.

How to Use

  1. In n8n, go to Workflows → Import from File
  2. Select any example JSON file
  3. Set up your FAIM API credentials
  4. Execute the workflow

Available Examples

1. Chronos2 Simple Forecast

File: chronos2-simple-forecast.json

What it does:

  • Takes a simple 1D time series (univariate)
  • Generates 10-step point forecast using Chronos2
  • Displays results with metadata and cost

Use case: Basic forecasting of a single metric (sales, traffic, etc.)

Data: Daily values over 20 days Model: Chronos2 (accurate, good for most use cases) Output: Single point forecast for next 10 periods

2. FlowState with Custom Parameters

File: flowstate-with-parameters.json

What it does:

  • Demonstrates multivariate forecasting (multiple features)
  • Uses FlowState model with custom parameters
  • Shows scale factor and prediction type (median vs mean)
  • Good for flexible, balanced forecasting

Use case: Forecasting multiple correlated metrics together

Data: 3 time series with 5 samples each, 2 features per sample Model: FlowState (flexible state-space model) Parameters:

  • Scale factor: 1.0 (normalization)
  • Prediction type: median (robust to outliers) Output: Point forecast for next 5 periods

3. Chronos2 Quantiles for Confidence Intervals

File: chronos2-quantiles-confidence.json

What it does:

  • Generates not just point forecasts, but confidence intervals
  • Extracts 5th, 25th, 50th, 75th, 95th percentiles
  • Shows 90% confidence interval (5th to 95th)
  • Includes median forecast (50th percentile)

Use case: Risk analysis, planning with uncertainty bounds

Data: Revenue data over 14 days Model: Chronos2 (LLM-based, excellent for quantiles) Output:

  • Median forecast (50th percentile)
  • 90% confidence interval (5th-95th percentile)
  • Intermediate quantiles (25th, 75th)

Extract Confidence Intervals node:

  • Demonstrates how to process quantile outputs
  • Shows Code node usage for data transformation
  • Easy to integrate into downstream analysis

4. Batch Processing - Multiple Time Series

File: batch-processing-multiple-series.json

What it does:

  • Forecasts 3 different time series in ONE API call
  • More efficient than calling API 3 times separately
  • Processes results and displays comparison
  • Shows cost savings of batch processing

Use case: Forecasting multiple products, regions, or metrics efficiently

Data: 3 separate time series (batch size = 3) Model: TiRex (transformer-based, good for batches) Output: Forecasts for all 3 series simultaneously

Performance benefit:

  • Single API call = single billing transaction
  • Lower latency than 3 sequential calls
  • Better resource utilization

Workflow Patterns

Pattern 1: Simple Processing

Data → FAIM Forecast → Display

Minimal setup, good for testing and exploration.

Pattern 2: Output Transformation

Data → FAIM Forecast → Code (transform output) → Display/Store

Use Code node to extract specific parts of the forecast for downstream use.

Pattern 3: Multiple Models Comparison

Data →
  ├→ Chronos2 Forecast
  ├→ FlowState Forecast
  └→ TiRex Forecast
    ↓
  Merge Results → Compare Metrics

Evaluate which model works best for your data.

Pattern 4: Batch Processing

Multiple Series Array → FAIM Forecast (batch) → Process Results

Efficient processing of many time series in single request.

Common Modifications

Change Input Data

Edit the Set node to use your own data:

{
  "myTimeSeries": [10, 12, 15, 14, 16, ...]
}

Change Forecast Horizon

In FAIM Forecast node, adjust Forecast Horizon parameter:

  • 7 for weekly
  • 24 for daily (next 24 hours)
  • 30 for monthly
  • 365 for yearly

Add More Models

Duplicate the FAIM Forecast node and change Model parameter to:

  • chronos2 (best accuracy)
  • flowstate (balanced)
  • tirex (fast)

Store Results

Add a database node after forecast to save results:

FAIM Forecast → PostgreSQL/MongoDB → Update your system

Troubleshooting Examples

Error: "Invalid API key"

  • Check credentials are properly set in n8n
  • Verify API key format: sk-...
  • Confirm key is active on faim.it.com dashboard

Error: "Request too large"

  • Reduce batch size or sequence length
  • Split into multiple smaller requests
  • Example: Instead of 1000 points, use 500 + 500

Slow Performance

  • Use batch processing (combine multiple series)
  • Reduce horizon if not needed
  • Use TiRex instead of Chronos2 for speed

High Costs

  • Batch multiple forecasts together
  • Use smaller horizons
  • Use point forecast instead of quantiles

Next Steps

  1. Customize for your data: Modify example workflows with your actual time series
  2. Add automation: Schedule workflows to run on a schedule (daily, weekly, etc.)
  3. Integrate with databases: Store forecasts in your system
  4. Build dashboards: Visualize forecasts using n8n's visualization nodes
  5. Monitor predictions: Track forecast accuracy over time

Support