combining one hot encoded category data and continuous data examples in TensorFlow JS - categorical-data

Question

When using both one-hot encoded category data and numerically variable data how do you combine these different types of data to fit the model?
Previously using TensorflowJS, and this is just my side project that I am using to learn more about google's tf, I have used only one-hot encoded data for both the feature and label values. All the features were the same shape as were the labels. I am simply building a prediction model for features that are one-hot encoded names such as
"name1" = [1,0,0]
"name2" = [0,1,0]
...
with outcomes that are one-hot encoded too such as
"outcome1" = [1,0,0]
"outcome2" = [0,1,0]
...
This worked pretty well and I was able to build a lovely model that when tested worked quite well.
However once I wanted to do some feature engineering and use a numerically continuous piece of data I could not see how to combine the category and continuous data.
What I am wanting to do is use an additional non category piece of variable data. So my features would be
"name1" = [1,0,0]
"name2" = [0,1,0]
variablePieceOfInformation = 10
I do not initially see a way within tensor flow of using these pieces of data together.
Initially all my features were identically sized one hot encoded category names and my labels were one hot encoded outcomes that I wanted to predict.
Now I want to use an additional piece of information that is continuously variable and I do not see examples on how to combine the two types.
What I was expecting was that there may be a way of combining category (onehot encoded) and continuously variable data.
my git repo
import getTrainingData, { TrainingData, getNames } from "./getModelData";
import save from "./saveModel";
import { numAllTimeTeams } from "#gvhinks/epl-constants";
const createModel = async (): Promise<tf.Sequential> => {
const model: tf.Sequential = tf.sequential({
name: "predict"
});
model.add(tf.layers.dense({inputShape: [2,numAllTimeTeams], units: 3, useBias: true, name:"teams_layer"}));
model.add(tf.layers.flatten());
model.add(tf.layers.dense({units: 3, useBias: true, name: "results_layer"}));
model.compile({optimizer: tf.train.adam(0.001), loss: 'meanSquaredError'});
const { labelValues, featureValues } = await getTrainingData();
const numTeamsInLeague: number = (featureValues[0][0]).length;
const featureTensors = tf.tensor3d(featureValues, [featureValues.length, 2, numTeamsInLeague], 'int32');
const labelTensors = tf.tensor2d(labelValues, [labelValues.length, 3], 'int32');
const fitArgs = {
batchSize: 500,
epochs: 200,
verbose: 0
};
await model.fit(featureTensors, labelTensors, fitArgs);
return model;
};
export { getTrainingData, createModel as default, TrainingData, getNames, save };```

Answer 1

Combining a tensor to have both categorical and continuous data is in itself not a complicated task. One needs to concat (using tf.concat) two tensors one with a categorical data and the second with a numerical data.
With such an input, the prediction can only be either a categorical or a numerical data, but not both - the reason being that the type of the data will influence the choice of the error function.