Creating a New Inference Session in PrismRCL

Introduction #

After training your AI model with PrismRCL, the next step is to deploy it for inference on new datasets. This guide explains how to initiate an inference session using a pre-trained model and a dataset not previously used during the training or testing phases.

Requirements #

• Pre-trained Model: You must have a model that was previously trained with PrismRCL.
• Inference Dataset: Prepare a dataset specifically for inference. This dataset should not include data used in the training or testing phases.
• Consistency in Evaluation Method: Ensure the evaluation method and parameters used during inference match those used during training.

Command Overview #

The command to start an inference session is structured as follows:

C:\PrismRCL\PrismRCL.exe fractal rclticks=15 loadmodel=C:\PrismRCL\models\best_model.classify testdata=D:\deploy\data\dataset\inference-data intoText=C:\PrismRCL\output\inference_output.txt log=D:\PrismRCL\logfiles\job_folder stopwhendone

• fractal: The evaluation method, must match the method used during training.
• rclticks=15: Specifies the setting for the fractal evaluation method, should be identical to the training session.
• loadmodel: Path to the pre-trained model to be used for inference.
• testdata: Path to the inference dataset.
• intoText: Destination path for the output text file where the inference results will be saved. The output will list the image names and their predicted classes in two columns.
• log: Directory where logs and result files from the inference session will be stored.
• stopwhendone: Instructs PrismRCL to shut down automatically after the inference session concludes.

Steps for Creating an Inference Session #

1. Verify Model and Method Consistency: Ensure the pre-trained model and evaluation method parameters align with those used during its training.
2. Prepare the Inference Dataset: Assemble your inference dataset, ensuring it is formatted according to PrismRCL’s requirements and does not overlap with training or testing data.
3. Set Up the Command: Customize the provided command with the paths to your model, inference dataset, output file, and log directory.
4. Execute the Inference Session: Run the command in the command line. Ensure you are in the PrismRCL directory or have the PrismRCL executable in your system’s PATH.
5. Review the Inference Output: After completion, review the output text file to see the predictions made by your model. The log files can provide additional insights into the inference process.

Conclusion #

Launching an inference session with PrismRCL is a crucial step in deploying your AI model to make predictions on new data. By following the steps outlined in this article, you can effectively utilize your pre-trained model to classify unseen datasets, allowing you to evaluate its performance in real-world scenarios.