CAI NEURAL API - conversion from TensorFlow

[Kryvich 165]

If you have experience with CAI NEURAL API, you can help me.

I need to create and train a neural network in Delphi like in TensorFlow, using the linear regression method::

import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense

# Model Definition
model = Sequential()
model.add(Dense(64, input_dim=14, activation='relu'))  # Input layer + first hidden layer
model.add(Dense(128, activation='relu'))               # Second hidden layer
model.add(Dense(64, activation='relu'))                # Third Hidden Layer
model.add(Dense(46, activation='linear'))              # Output layer

# Compiling the model
model.compile(optimizer='adam', loss='mean_squared_error')

# Train the model ( X_train and y_train - training data)
model.fit(X_train, y_train, epochs=50, batch_size=10, validation_split=0.2)

# Prediction for a new order
predicted_materials = model.predict(new_order)
print(predicted_materials)

I created a CAI NEURAL API model, prepared training data, and started training.

  NN := TNNet.Create();
  NFit := TNeuralFit.Create();
  CreateWorkMatPairLists(80{%}, 10{%}, 10{%}, TrainingPairs, ValidationPairs, TestPairs);
  NN.AddLayer([
    TNNetInput.Create(WORKS_COUNT),
    TNNetFullConnectReLU.Create(64),
    TNNetFullConnectReLU.Create(128),
    TNNetFullConnectReLU.Create(64),
    TNNetFullConnectLinear.Create(MATERIALS_COUNT)
  ]);

  NFit.InitialLearningRate := 0.001;
  NFit.LearningRateDecay := 0.01;
  NFit.L2Decay := 0.00001;
  NFit.Fit(NN, TrainingPairs, ValidationPairs, TestPairs, {batchsize=}10, {epochs=}50);

But I don't know how to set up the optimizer correctly (TensorFlow uses Adam). 

And how to set the loss function - mean squared error.

Does the CAI NEURAL API have such functionality, or do I need to write it myself?

Edited 5 hours ago by Kryvich