Overview
When setting up a VRIFY AI Prediction Map, you will be guided through six key steps to analyze geological data and other relevant features to generate results.
This process includes:
Defining your area of interest.
Selecting the appropriate features (input data).
Specifying what you are targeting, and at what grade.
Reviewing the results and adjusting the parameters to align with your real-world expertise of the project.
Carefully configuring your experiments improves the accuracy of drill target predictions, ensuring it is tailored to your unique targets and objectives.
However, many of the parameters have default values. If you are not sure how to set the values, start with the defaults and make any necessary changes after the first pass.
Step-by-Step Instructions
View Panel Navigation Tips
At any time, click the compass on the top right of the view panel to view the birds-eye-view and centered perspective of your AOI.
Left-click and drag to spin the model.
Right-click and drag to move the model around without spinning it.
Zoom in and out by scrolling or pinching your touch pad.
Log into VRIFY.com
On the left side panel, click VRIFY AI.
Click + New Prediction Map on the top right of the screen.
Select the asset you want to use from the drop-down menu at the top of the page.
Next, at the top of the screen, input a name for your Prediction Map.
If a custom name is not entered, it will be auto-created with a unique map id, your company name, and the asset used.
From here, the instructions are broken down into the 6 key steps.
Jump to:
Select AOI
In this step, an AOI (Area of Interest) is set for this experiment. This lets you set the boundary for your features, ensuring predictions are only made within the specified AOI. Features outside of the specified AOI won’t be included in this experiment’s feature set.
In the Step 1 pop-out, select the AOI file you want to use from the Select an AOI drop-down menu.
Using the sliders, set the Height (px) and Width (px) for your AOI. As these settings are adjusted, the view panel will reflect the shape and density of your AOI grid.
The default is 512px. Check out this AOI Pixels Guide for more details on optimal pixelization.
This parameter impacts the resolution of the results. The higher the pixels, the higher the resolution.
Note: Choose a resolution that matches your data.
Generally, for smaller, well-defined AOIs, a higher resolution may produce better results. For broader, less defined AOIs with scattered coverage and is of lower resolution, a lower resolution might be more suitable.
If you need to create a new AOI shape, like setting an AOI in a different area, please reach out to your VRIFY AI contact for assistance in uploading a new shapefile.
Click Apply AOI to complete this step.
For more detailed information on Selecting an AOI see: Understanding: Select AOI and AOI Pixels Guide
Input Features
This section is where the features (exploration vectors) that will be used in the model are defined. The features available to you are the result of the data compilation process, and if you notice you are missing anything, please contact your dedicated VRIFY AI contact.
Start by selecting the features you want to use from the Input Features pop-out.
You can hold shift+click to select or deselect a range of features at one time.
Optionally, you can visualize the features available by clicking on the eye-icon next to the feature. This step won't affect the model output.
Selecting a Terrain Layer allows you to have a sustained visualization of whichever layer is selected, allowing you to overlay the other feature layers on top of.
You can apply a colour scale to the visualized layer to better visualize the features.
Adjust the Vertical Exaggeration slider at the bottom of the screen to exaggerate the layers if desired.
Next, set the Correlation Threshold.
Setting a correlation threshold helps prevent an imbalance of too much importance on duplication in the dataset by indicating how similar two features must be before one is omitted from the features list.
The default threshold is 0.90 (features that are 90% similar will be flagged as duplicate information, and one of the 2 will be removed).
If the threshold is set to 0, every feature selected will be used, regardless of similarity. If set to 100, only exact duplicate features will be removed.
Finally, click Apply to complete this step.
For more detailed information on Input Features see: Understanding: Input Features
Set Up Learning Data
This step is where the targeted elements are defined, and where the learning data is separated into positive and negative examples, for our 2-class classification algorithm. The threshold indicates the minimum grade the element must meet to be represented as a positive learning data point. Element(s) that do not reach the defined grade threshold will be represented as a negative learning point.
Aim for an even spread of positive (mineralized) and negative (unmineralized) data to ensure a balance in learning points, if your data allows for it. The system uses these learning points during training to recognize patterns in the data and make predictions on the underlying information.
In the Learning Data popout, start by selecting a file to use for your Learning Points. If this drop-down is not available, it means you have only 1 file and it is being used by default.
Select the Elevation Field from the dropdown menu.
This is used to predict the depth of the target occurrence.
The values in this drop-down menu are populated from the column headers in your AOI shapefile attribute file.
Then, set the Learning Data Filters. Start by selecting an Element from the drop-down menu.
By default, the filter is set to include elements greater than the corresponding target threshold. If necessary, toggle the switch to change it to less than.
Next, set a target threshold for the element. This is where you indicate the grade the element must meet to be represented as a positive learning data point.
The available target threshold range is displayed below the filter.
When setting this threshold, refer to the Learning Data Breakdown table at the bottom of the popout. This provides realtime ratios that indicate how balanced your positive and negative learning points are:
When targeting multiple commodities, you can specify whether your targets are combined using “AND” logic, where thresholds for each target must be met, or handled independently with “OR” logic, where at least one of the thresholds for any of the targets can be met. Click the "or" toggle to adjust.
Optionally, click Add to add another data filter (target element).
Click Proceed to set the parameters and complete this step.
For more detailed information on Setting up Learning Data see: Understanding: Set Up Learning Data
Embed Visual Features
This step is where you configure which Vision Transformer Model will be used for your predictions. The Vision Transformer will conduct a dimensional reduction on the input features, much like a PCA would do. The number of epochs determine the number of times your dataset passes through the algorithm to conduct what is called local fine-tuning of the mode (referred to as the number of Epochs). This step is important to adjust the global model to the local geological domain specific to your project area.
The Vision Transformer Models available are for specific mineral system, including a Master Model that can be used if there is no model available for your desired mineral system.
Review the visualized data points in the viewer, which will show green (mineralized/positive data point) and red (unmineralized/negative data point) based on the grade thresholds set in the Set Up Learning Data step.
Check the data looks visually correct with the threshold you have set for your chosen elements. The displayed results are based on drilling and surface samples. Return to the previous step to adjust the parameters if necessary.
In the Embed Visual Features popout, start by selecting a Vision Transformer Model from the drop-down menu.
Choose the vision transformer that corresponds with the mineral system you are targeting.
If there is not a model for the mineral system you are targeting, choose “Master_Model.pt”.
Set the No. of Epochs to indicate the number of full passes through your training data, in the local fine-tuning of your model. The default is 100.
The model will run until the number of epochs is reached, or until the loss function levels out (meaning the model is no longer increasing in accuracy); whichever comes first.
Next, click Generate Embeddings to run the dimension reduction exercise, generating a new set of rasters for your input features.
Optional step: Through the 3D Layers Menu, review the raster files generated in this step by clicking the eye-icon to preview it. Set your desired colour scheme.
This provides valuable insight into the generated embeddings, offering a clearer understanding of the structure.
The base displacement scale allows you to adjust the intensity of surface detail, helping you more accurately visualize and represent the textures of each raster.
Click Proceed.
For more detailed information on Embedding Visual Features see: Understanding: Embed Visual Features
Build Predictive Models
In this step, a two-class classifier is used to predict the probability of mineralization occurrences within your AOI. The classifier is trained from your embedded rasters and your selected learning points. The parameters in this section are used to group your learning points into clusters that allow us to train/validate the model appropriately. Advanced options will also be available in order to overfit or underfit your predictive model.
In the Predictive Modelling popout, there are two parameters you can set; Minimum Cluster Size and Minimum Samples. With these parameters we are controlling the size and spatial extent of the train and test groups used for validation of the model.
Start by setting the Minimum Cluster Size.
The default is 100.
The Minimum Cluster Size indicates how far the clustering algorithm (that groups points together) is going to look to find the results for each cluster of data.
Next, set the Minimum samples.
The default is 10.
This sets the minimum number of samples needed to create a cluster of data.
Next, click Generate Model.
Note: There are advanced settings which are set to our recommended defaults. Please work with the VRIFY team if there are settings you are wanting to manage.
For more detailed information on Building Predictive Models see: Understanding: Build Predictive Models
Generate Targets
This final step is where your exploration targets are generated. You'll start by reviewing the results from the predictive model, adjusting the parameters as needed, generating SHAP labels, then review the results again. You can repeat this cycle as many times as necessary to refine your targets.
The view panel will display a visualization of the generated prospectivity score through a range of shades which represent the VRIFY Prospectivity Scores (VPS). This helps to refine your targets prior to exporting your final results. Prospectivity is represented by this colour scale:
When Step 5 is complete, a series of result graphs will display at the bottom of the screen.
Using your industry knowledge and familiarity with the project, evaluate the results, ensuring they fit within established geological patterns and project specifics for informed decision-making. For more details on interpreting the results graphs, check out this article.
Next, you can make adjustments to optimize how your prospectivity score results are clustered into defined exploration targets. The view panel will initially display targets based on the default parameters for the Target Threshold, Minimum Cluster Size, and Minimum Samples.
Target threshold: this controls the threshold of the prospectivity score that determines your targets. The view panel will update to display only targets that are at or above the set threshold, to help visualize your targets.
Move the slider to the right to display only areas with higher prospectivity scores, or conversely, farther to the left to include moderate prospectivity scores. Keep in mind that a score below 0.5 amends to a negative learning point, which means there are more chances this area is barren than it is mineralized.
Minimum Cluster Size: indicates the minimum size a cluster must be to be considered a target.
Minimum samples: indicates the minimum samples needed to create a cluster for a target.
Once your Target Threshold is set, we move on to generate our Target Clusters. When the Minimum Cluster Size or Minimum Samples configurations are adjusted, Target Clusters will be generated in the View Panel.
Each Target Cluster is colour coded.
Adjusting the Minimum Cluster Size and Minimum Samples sliders directly impact the number of Target Clusters.
Continue making adjustments as needed to the Target Threshold, Max Distance, and Min samples parameters, then click Generate SHAP Labels to apply your adjusted thresholds. This step will generate one label per target cluster.
Proceed to Export Results from the Results panel at the bottom of the screen.
To have your VRIFY AI 3D Model added to a VRIFY Presentation Deck, reach out to your dedicated VRIFY AI contact.
For more detailed information on Generating Targets see: Understanding: Generate Targets
Still have questions?
Reach out to your dedicated VRIFY AI contact or email Support@VRIFY.com for more information.