How to I get the class probabilities from a model

TensorFlow General Discussion
,
1

I compute a keras model as follows;

        skf = StratifiedKFold(n_splits=n_splits, shuffle=False)
        for k, (train_index, test_index) in enumerate(skf.split(inputs, targets)):
            print(f"trait {trait_idx + 1}: {trait_dict[trait]}, fold: {k}")
            x_train, x_test = inputs[train_index], inputs[test_index]
            y_train, y_test = targets[train_index], targets[test_index]
            # converting to one-hot embedding
            y_train = tf.keras.utils.to_categorical(y_train, num_classes=n_classes)
            y_test = tf.keras.utils.to_categorical(y_test, num_classes=n_classes)
            model = tf.keras.models.Sequential()

            # define the neural network architecture
            model.add(
                tf.keras.layers.Dense(50, input_dim=hidden_dim, activation="relu")
            )
            model.add(tf.keras.layers.Dense(n_classes))

            model.compile(
                optimizer=tf.keras.optimizers.Adam(learning_rate=lr),
                loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),
                metrics=["mse", "accuracy"],
            )

            history = model.fit(
                x_train,
                y_train,
                epochs=epochs,
                batch_size=batch_size,
                validation_data=(x_test, y_test),
                verbose=0,
            )

and I put in some test data to get a prediction.

    for _trait, _model in dict_models.items():
        script = tf.convert_to_tensor(inputs, dtype=float) 
        _results = _model(script, training=False)
        #_results = _model.predict_classes(script)
        _results_numeric = _results.numpy()
        pred = np.argmax(_results, axis=1)
        authors = author_ids[0]
        print(f"results for {dict_trait[_trait]}")
        print(f"                {authors[0]}: {pred[0]} : {_results_numeric[0, :]}")
        print(f"                {authors[1]}: {pred[1]} : {_results_numeric[1, :]}")

What I would like from the results is a probability that the class prediction is correct. The results scores I am getting are outside of the 0 - 1 range and so are not probabilities. So how do I fix this?

2

Hi @Geoffrey_Payne,

Can you please incorporate the below code changes into your code and check:

probabilities = tf.nn.softmax(_results).numpy()
pred = np.argmax(probabilities, axis=1)

Your model predictions are not in the range of 0 to 1, which means that your model is not producing class probabilities directly. In such cases, you can use the softmax function to convert the raw scores or logits into probabilities.

I hope this helps!

Thanks

1 Like
4

This is the correct answer. I should have known. Thank you very much!