You can deploy this model and start earning money today!

studentscolab/
iris_keras

multiclass-classification

tabular-classification

mit

Views

No views yet

Iris MLP Classifier (Keras / TensorFlow)

This repository contains a simple multiclass classifier for the classic Iris dataset, implemented as a small MLP (Multi-Layer Perceptron) in TensorFlow / Keras.
The model predicts one of three classes based on four numerical features.

Task

Tabular multiclass classification
Given the 4 iris measurements, predict the class:

0 → setosa
1 → versicolor
2 → virginica

Dataset

Iris dataset (from sklearn.datasets.load_iris)

Input features (4)

The model expects 4 float features in this exact order:

sepal length (cm)
sepal width (cm)
petal length (cm)
petal width (cm)

Target (3 classes)

integer labels y ∈ {0,1,2}
no one-hot encoding required (training uses sparse_categorical_crossentropy)

Model architecture

A small feed-forward network with built-in feature normalization:

Why `Normalization` layer?

The tf.keras.layers.Normalization layer learns feature-wise mean and variance from the training set via adapt(...).
This makes inference easier and safer: the same scaling used during training is embedded inside the saved model.

Training configuration

Optimizer: Adam (learning_rate=1e-3)
Loss: sparse_categorical_crossentropy
Metric: accuracy
Train/test split: 80/20 (stratify=y, random_state=42)
Validation split (from train): 20% (validation_split=0.2)
Epochs: 100
Batch size: 16
Reproducibility:
- tf.random.set_seed(42)
- np.random.seed(42)

Note: Exact accuracy may vary slightly across environments due to numerical differences and nondeterminism in some TF ops.

Example: training script (reference)

The model was trained using the following core logic (simplified):

python

normalizer = tf.keras.layers.Normalization(axis=-1)
normalizer.adapt(X_train.to_numpy())

model = tf.keras.Sequential([
    tf.keras.Input(shape=(4,)),
    normalizer,
    tf.keras.layers.Dense(16, activation="relu"),
    tf.keras.layers.Dense(16, activation="relu"),
    tf.keras.layers.Dense(3, activation="softmax"),
])

model.compile(
    optimizer=tf.keras.optimizers.Adam(learning_rate=1e-3),
    loss="sparse_categorical_crossentropy",
    metrics=["accuracy"]
)

model.fit(
    X_train.to_numpy(), y_train.to_numpy(),
    validation_split=0.2,
    epochs=100,
    batch_size=16,
    verbose=0
)

Example