SandTank-ML provides an application for learning and seeing how ML behaves compared to a physics-based simulator such as ParFlow.
In order to run this application you will need to have Docker installed along with a set of pre-trained ML models that have been stored in a google drive under Hydroframe-ML > Sandtank-ML > data.
# Customize these
PORT=9000
DATA=/path_to_gdrive_content/
docker run --rm \
-it \
-p ${PORT}:80 \
-v ${DATA}:/data \
hydroframe/sandtank:mlJust point your browser to $SERVER_NAME
Then to choose a given lesson plan (json file in your $DATA directory), just provide the name of the file without the extension in the URL. There are sample lesson plans in a google drive under Hydroframe-ML > Sandtank-ML > data.
For example, after downloadingdropout.json, point your browser to http://localhost:9000/?name=dropout.
Lesson plans are .json files in your $DATA directory following the structure below. Samples are available in this repos /data folder.
Set the range of timesteps available to the simulator. (Note, -1 is for debugging purposes)
"time": [-1, 5],Select which charts and images are available for comparison, as well as whether to show their show/hide controls.
"moduleSelector": {
"show": true,
"values": ["selection", "prediction", "diff", "hist", "error", "stats"]
},Select the contrast range for the image diffing the simulation vs an ai model.
"diffScaling": {
"show": true,
"value": 0.5,
"min": 0.1,
"step": 0.1,
"max": 1
}Select whether to show pressure, or saturation, or to show a button switching betwen them.
"usePressure": {
"show": true,
"value": false
}
Select whether to let the user add/remove models to/from the comparison. Also, set which models will appear in the comparison at startup.
"addRemoveAI": {
"show": true,
"defaultModels": [
{
"model": "RegressionPressure",
"training":"full",
"learningRate": "lr4",
"dropOut":"ndp",
"epoch":"e1"
},
]
}Configure labels for permeability image. Labels might be in a legend, or overlayed on the image following the mouse. values supplies text to overwrite default numeric labels. Setting a value to null will hide that value in the legend.
"usePermeabilityLabels": {
"showOverlay": true,
"showLegend": true,
"values": {
"0": "A River",
"1": "Sand",
"0.6": "Compacted Sand",
"0.05": "Dense Clay"
}
},If you prefer to only label the minimum and maximum values, you may instead use rangeLabels.
"usePermeabilityLabels": {
"showOverlay": true,
"showLegend": true,
"values": {
"0": null,
"rangeLabels": {
"low": "Low permeability",
"high": "High permeability"
}
}
},Configure labels for pressure or saturation images. Labels might be in a legend, or overlayed on the image following the mouse. values supplies text to overwrite default numeric labels. Setting a value to null will hide that value in the legend.
"useWaterLabels": {
"showOverlay": true,
"showLegend": true,
"saturationValues": {
"-1": null,
"rangeLabels": {
"low": "Unsaturated",
"high" : "Saturated"
}
},
"pressureValues": {
"-1": null,
"rangeLabels": {
"low": "Low pressure",
"high" : "High pressure"
}
}
},If you prefer to only label the minimum and maximum values, you may instead use rangeLabels.
"usePermeabilityLabels": {
"showOverlay": true,
"showLegend": true,
"values": {
"0": null,
"rangeLabels": {
"low": "Low permeability",
"high": "High permeability"
}
}
},Select whether a label should follow the mouse along the diff image. Also, set label values to show instead of numbers. values supplies text to overwrite default numeric labels. Setting a value to null will hide that value in the legend.
"useDiffLabels": {
"show": true,
"values": {
"0": "Accurate",
"1": "Error"
}
}Whether to show the button that toggles legend visibility, and the initial value for that button.
"useLegend": {
"show": true,
"value": true
},Select whether to skip the first epoch in learning stats sets with more than one epoch. First epochs are often outliers, distoring the graph. A value can be set, or a button can be shown.
"useSkipInitial" : {
"show": true,
"value": true
}Add guidance for Sandy Loam, our character, to give the user as they learn to use the application. Every step must have a name, and some names are special. These names will highlight parts of the app, so Sandy can point to them as she explains:
"walkthrough": {
"steps": [
{
"name": "start",
"guidance": "If you like, I can guide you through using Sandtank ML."
},
}The application will look for files in your $DATA directory corresponding to the uri pattern. That pattern is filled by your parameters, which will be shown in your specified order.
"uriPattern": "${model}://${training}-${learningRate}-${dropOut}-${epoch}.out"
"order": ["model", "training", "learningRate", "dropOut", "epoch"],
"parameters": {
"model": {
"label": "Model",
"items": [{ "text": "Pressure", "value": "RegressionPressure" }]
},
"training": {
"label": "Training set",
"items": [
{ "text": "All", "value": "full" },
{ "text": "Wet", "value": "wet" },
{ "text": "Dry", "value": "dry" }
]
},
"learningRate": {
"label": "Learning Rate",
"items": [
{ "text": "0.0001", "value": "lr4" }
]
},
"dropOut": {
"label": "Use drop out",
"items": [
{ "text": "No", "value": "ndp" }
]
},
"epoch": {
"label": "Number of trainings",
"items": [
{ "value": "e1", "text": "One time" }
]
}