Skip to main content

DeepLearning

Deep Learning Node

The Deep Learning node enables you to build, train, and deploy neural networks for advanced AI predictions without any coding. This powerful node automatically creates sophisticated machine learning models that can recognize patterns, make predictions, and classify data with high accuracy.

What This Node Does​

The Deep Learning node processes your data through artificial neural networks - computer systems that learn patterns similar to how the human brain works. It can predict future sales, classify customer behavior, detect anomalies, or recognize patterns in any structured data you provide.

Key Capabilities:

  • Pattern Recognition: Automatically discovers complex relationships in your data
  • Predictive Analytics: Forecasts future trends and outcomes
  • Classification: Categorizes data into different groups or types
  • Custom Model Training: Builds AI models specifically for your business needs
  • Real-time Predictions: Applies trained models to new data instantly

Business Applications​

Sales Forecasting​

Business Situation: A retail company wants to predict monthly sales based on historical data, seasonality, and marketing campaigns.

What You'll Configure:

  • Use the Deep Learning Editor to design your neural network
  • Set output option to "Return result merged with input" to keep original data
  • Train the model using 2+ years of historical sales data
  • Deploy for real-time sales predictions

Business Value: Improves inventory planning accuracy by 40% and reduces stockouts by 25%.

Customer Behavior Analysis​

Business Situation: An e-commerce platform needs to predict which customers are likely to make repeat purchases.

What You'll Configure:

  • Design a classification network in the Deep Learning Editor
  • Choose "Return result only" for clean prediction outputs
  • Train on customer transaction history and engagement metrics
  • Set up automated customer scoring

Business Value: Increases targeted marketing effectiveness by 60% and customer retention by 35%.

Quality Control Automation​

Business Situation: A manufacturing company wants to automatically detect product defects from sensor data.

What You'll Configure:

  • Create an anomaly detection neural network
  • Use custom training files if you have pre-trained models
  • Set up real-time quality monitoring
  • Configure alerts for detected anomalies

Business Value: Reduces defective products by 80% and saves $50,000 monthly in quality control costs.

Configuration Parameters​

Deep Learning Editor​

  • Field Name: Deep Learning Editor (Button)
  • Type: Button that opens visual neural network designer
  • Default Value: Available when editor is enabled
  • Simple Description: Opens a visual interface where you design your neural network architecture by dragging and dropping layers
  • When to Change This: Use this to create custom neural networks for specific business problems
  • Business Impact: Proper network design is crucial for model accuracy - well-designed networks can achieve 90%+ prediction accuracy

Custom DNN Files Section​

Use Custom DNN Files​

  • Field Name: customFiles
  • Type: Toggle switch (On/Off)
  • Default Value: Off
  • Simple Description: Enables you to use pre-trained neural network files instead of training from scratch
  • When to Change This:
    • On: When you have existing trained models or want to use industry-standard pre-trained networks
    • Off: When building and training new models from your data
  • Business Impact: Using pre-trained models can reduce training time from hours to minutes and leverage proven architectures

Pipeline Pickle Filename​

  • Field Name: filePathPipelinePikle
  • Type: Text field
  • Default Value: Empty
  • Expected Format: Filename ending in .pkl (e.g., "customer_pipeline.pkl")
  • Simple Description: The filename of your data preprocessing pipeline that prepares raw data for the neural network
  • When to Change This: When you have a custom data preprocessing pipeline that transforms your business data into the format needed by your neural network
  • Business Impact: Proper data preprocessing can improve model accuracy by 20-30%

DNN Training Filename​

  • Field Name: filePathNeuralTraining
  • Type: Text field
  • Default Value: Empty
  • Expected Format: Filename ending in .hdf5 (e.g., "sales_model.hdf5")
  • Simple Description: The filename of your pre-trained neural network model
  • When to Change This: When you want to use an existing trained model instead of training a new one
  • Business Impact: Reusing proven models ensures consistent performance and eliminates training time

Output Configuration​

Output Format Option​

  • Field Name: outTransformId
  • Type: Dropdown menu with options:
    • Return result merged with input: Combines prediction results with your original data, keeping all original columns plus adding prediction columns
    • Return result only: Returns only the prediction results without original data, creating a clean output focused solely on predictions
  • Default Value: Return result only
  • Simple Description: Controls how the neural network results are formatted in your workflow output
  • When to Change This:
    • Merged with input: When you need both original data and predictions for reporting or further analysis
    • Result only: When you only need the predictions for decision-making or triggering other workflow actions
  • Business Impact: Proper output formatting ensures downstream workflow nodes receive data in the expected format

Action Buttons​

Train Model Button​

  • Function: Starts the neural network training process using your configured settings and data
  • When to Use: After designing your network and configuring all parameters
  • What Happens: The system trains your neural network on your data, which may take several minutes to hours depending on data size and network complexity
  • Business Impact: Training creates the AI model that will make predictions on new data

Log Button​

  • Function: Opens the training log to view progress, performance metrics, and any error messages
  • When to Use: During training to monitor progress, or after training to review model performance
  • What Happens: Displays detailed information about training progress, accuracy metrics, and system messages
  • Business Impact: Helps optimize model performance and troubleshoot any training issues

Step-by-Step Configuration​

Setting Up Basic Deep Learning​

  1. Adding the Node:

    • Drag the Deep Learning node from the AI/ML section onto your workflow canvas
    • Connect it to your data source node using the arrow connector

  2. Designing Your Neural Network:

    • Click the "Deep Learning Editor" button to open the visual designer
    • Drag neural network layers from the toolbox onto the design canvas
    • Connect layers by drawing arrows between them
    • Configure each layer's parameters using the property panels
    • Save your network design and close the editor

  3. Configuring Output Format:

    • In the "Output" section, click the dropdown menu
    • Choose "Return result merged with input" to keep original data with predictions
    • Or select "Return result only" for clean prediction outputs

  4. Training Your Model:

    • Click the "Train Model" button to start training
    • Monitor progress by clicking the "Log" button
    • Wait for training to complete (indicated in the log)

Using Pre-Trained Models​

  1. Enable Custom Files:

    • Toggle "Use Custom DNN Files" to On
    • This reveals additional configuration fields

  2. Specify Model Files:

    • Enter your pipeline file name in "Pipeline Pickle Filename" (e.g., "data_pipeline.pkl")
    • Enter your model file name in "DNN Training Filename" (e.g., "trained_model.hdf5")
    • Ensure these files are uploaded to your TheoBuilder file storage

  3. Configure and Deploy:

    • Set your preferred output format
    • Test the configuration with sample data
    • Deploy your workflow for real-time predictions

Industry-Specific Applications​

Financial Services​

Common Challenge: Detecting fraudulent transactions in real-time while minimizing false positives that block legitimate customers.

How This Node Helps: Creates sophisticated fraud detection models that analyze transaction patterns, user behavior, and risk factors simultaneously.

Configuration Recommendations:

  • Design a multi-layer classification network in the Deep Learning Editor
  • Use "Return result only" output for clean fraud scores
  • Train on historical transaction data with known fraud labels
  • Set up real-time scoring for incoming transactions

Results: Financial institutions achieve 95% fraud detection accuracy with 60% fewer false positives, saving millions in fraud losses.

Healthcare Organizations​

Common Challenge: Predicting patient readmission risk to improve care coordination and reduce costs.

How This Node Helps: Analyzes patient data, medical history, and treatment patterns to identify high-risk patients requiring additional care.

Configuration Recommendations:

  • Create a regression network for risk scoring
  • Use "Return result merged with input" to maintain patient context
  • Train on electronic health records and readmission outcomes
  • Implement HIPAA-compliant data handling procedures

Results: Hospitals reduce readmission rates by 23% and improve patient outcomes while saving $2M annually in penalties.

Manufacturing Operations​

Common Challenge: Predicting equipment failures before they cause costly production downtime.

How This Node Helps: Processes sensor data, maintenance records, and operational parameters to forecast equipment failures days or weeks in advance.

Configuration Recommendations:

  • Design a time-series prediction network
  • Use custom preprocessing pipelines for sensor data normalization
  • Train on historical sensor data and failure records
  • Set up automated maintenance scheduling based on predictions

Results: Manufacturing plants achieve 85% reduction in unplanned downtime and 40% lower maintenance costs.

Retail and E-commerce​

Common Challenge: Optimizing inventory levels across multiple locations while minimizing stockouts and overstock situations.

How This Node Helps: Predicts demand patterns considering seasonality, promotions, weather, and local events for accurate inventory planning.

Configuration Recommendations:

  • Create multi-output regression networks for different product categories
  • Use "Return result merged with input" to maintain product details
  • Train on sales history, promotional data, and external factors
  • Implement automated reordering based on predictions

Results: Retailers improve inventory turnover by 30% while reducing stockouts by 45%, increasing profitability by 15%.

Best Practices​

Model Design Guidelines​

  • Start Simple: Begin with basic network architectures and add complexity gradually
  • Data Quality: Ensure your training data is clean, complete, and representative
  • Validation: Always test your model on data it hasn't seen during training
  • Regular Updates: Retrain models periodically as business conditions change

Performance Optimization​

  • Monitor Training: Use the Log button to track training progress and identify issues early
  • Batch Processing: For large datasets, process predictions in batches rather than individual records
  • Resource Management: Schedule intensive training during off-peak hours
  • Version Control: Keep track of different model versions and their performance metrics

Business Integration​

  • Stakeholder Alignment: Ensure model outputs align with business decision-making processes
  • Threshold Setting: Establish clear decision thresholds for classification and scoring models
  • Feedback Loops: Implement systems to capture model performance in real-world applications
  • Documentation: Maintain clear records of model configurations and business rules

The Deep Learning node transforms complex business challenges into automated, intelligent solutions that continuously improve over time. By leveraging advanced neural networks through an intuitive interface, organizations can implement enterprise-grade AI without requiring specialized technical expertise.