ModelProxy

(static) dispose(params, args, data) → {void}

Parameters:

Name	Type	Description
params	Object	Not used by this function.
args	Object	Not used by this function.
data	Object	Not used by this function.

Source:

• modules/analyze/components/nn.js, line 719

Returns:

Type

void

Example

model.dispose();

(async, static) predict(options) → {Promise.<Array>}

Parameters:

Name	Type	Description
options	Object	Prediction options Properties Name Type Attributes Description params Object <optional> Prediction parameters Properties Name Type Attributes Description inputShape Array <optional> Override input shape data Array Input data for prediction

Source:

• modules/analyze/components/nn.js, line 610

Returns:

Type

Promise.<Array>

Examples

// Dense model: Predict streamflow from current conditions
const model = await hydro.analyze.nn.createModel({
  params: { type: 'dense' },
  args: { inputShape: [3], units: [16, 1] }
});
await model.train({ params: { epochs: 50 }, data: [X_train, y_train] });

// Predict for new conditions
const newConditions = [[6, 22, 14]]; // [precip, temp, humidity]
const flowPrediction = await model.predict({ data: newConditions });
console.log(`Predicted flow: ${flowPrediction[0][0].toFixed(2)} m³/s`);

// LSTM: Forecast next day flow (auto-reshaping)
const lstmModel = await hydro.analyze.nn.createModel({
  params: { type: 'lstm' },
  args: { timeSteps: 7, features: 1, units: [32], outputUnits: 1 }
});
await lstmModel.train({ params: { epochs: 50 }, data: [sequences, targets] });

// 1D input automatically reshaped to [1, 7, 1] for LSTM
const last7Days = [100, 105, 98, 110, 115, 108, 112];
const nextDayPrediction = await lstmModel.predict({ data: last7Days });
console.log(`Tomorrow's flow: ${nextDayPrediction[0][0].toFixed(2)} m³/s`);

// Batch predictions: Multiple scenarios
const scenarios = [
  [5, 20, 15],  // Scenario 1: light rain
  [15, 22, 18], // Scenario 2: heavy rain
  [2, 25, 12]   // Scenario 3: minimal rain
];
const predictions = await model.predict({ data: scenarios });
scenarios.forEach((scenario, i) => {
  console.log(`Scenario ${i+1}: ${predictions[i][0].toFixed(2)} m³/s`);
});

// Classification: Flood risk category
const classifier = await hydro.analyze.nn.createModel({
  params: { type: 'dense' },
  args: { inputShape: [4], units: [32, 16, 3], outputActivation: 'softmax' }
});
// Train with flood categories: 0=Normal, 1=Warning, 2=Danger

const currentConditions = [[120, 8, 25, 85]]; // [flow, precip, temp, saturation]
const probabilities = await classifier.predict({ data: currentConditions });
const category = probabilities[0].indexOf(Math.max(...probabilities[0]));
const categories = ['Normal', 'Warning', 'Danger'];
console.log(`Flood Risk: ${categories[category]} (${(probabilities[0][category]*100).toFixed(1)}%)`);

(static) save(params, args, data) → {Promise.<Object>}

Parameters:

Name	Type	Description
params	Object	Contains: name (filename for the model).
args	Object	Not used by this function.
data	Object	Not used by this function.

Source:

• modules/analyze/components/nn.js, line 698

Returns:

Type

Promise.<Object>

Example

await model.save({
  params: { name: 'my-trained-model' }
});

(async, static) train(options) → {Promise.<Object>}

Parameters:

Name	Type	Description
options	Object	Training options Properties Name Type Attributes Description params Object <optional> Training parameters Properties Name Type Attributes Default Description epochs number <optional> 50 Number of training epochs batchSize number <optional> 32 Batch size for training validationSplit number <optional> 0.2 Validation data fraction (0-1) shuffle boolean <optional> true Shuffle training data onEpochEnd function <optional> Callback after each epoch data Object | Array Training data: [inputs, outputs] or {inputs, outputs}

Source:

• modules/analyze/components/nn.js, line 497

Returns:

Type

Promise.<Object>

Examples

// Basic training: Streamflow prediction from rainfall
const model = await hydro.analyze.nn.createModel({
  params: { type: 'dense' },
  args: { inputShape: [3], units: [16, 8, 1] }
});

const inputs = [[5, 20, 15], [3, 18, 12], [8, 25, 18]]; // [precip, temp, humidity]
const outputs = [[120], [95], [145]]; // streamflow

const history = await model.train({
  params: { epochs: 100, batchSize: 16 },
  data: [inputs, outputs]
});
console.log(`Final loss: ${history.loss[history.loss.length - 1]}`);

// LSTM training: Time series forecasting
const lstmModel = await hydro.analyze.nn.createModel({
  params: { type: 'lstm' },
  args: { timeSteps: 7, features: 2, units: [32], outputUnits: 1 }
});

// Training data: 7 days of [rainfall, temperature] → next day flow
const sequences = [
  [[[10,20],[12,22],[8,21],[15,23],[5,19],[7,18],[9,20]]], // Input sequence
  [[125]]  // Target: next day flow
];

await lstmModel.train({
  params: { epochs: 50, validationSplit: 0.2 },
  data: sequences
});

// Training with callbacks and monitoring
await model.train({
  params: {
    epochs: 100,
    batchSize: 32,
    onEpochEnd: (epoch, logs) => {
      console.log(`Epoch ${epoch}: loss=${logs.loss.toFixed(4)}, val_loss=${logs.val_loss?.toFixed(4)}`);
      // Early stopping logic could go here
    }
  },
  data: [X_train, y_train]
});

// Object format (alternative to array format)
await model.train({
  params: { epochs: 50 },
  data: {
    inputs: flowObservations,
    outputs: nextDayFlow
  }
});