import tensorflow as tf

a = tf.constant(5.0)
b = tf.constant([1.0, 2.0, 3.0])
c = tf.constant([[1, 2], [3, 4]], dtype=tf.float32)

print(a.shape, b.shape, c.shape)   # () (3,) (2, 2)
print(c.dtype)                       # <dtype: 'float32'>
print(tf.reduce_sum(c).numpy())      # 10.0
print(tf.matmul(c, c).numpy())       # [[ 7. 10.] [15. 22.]]

import numpy as np
import tensorflow as tf

arr = np.arange(12, dtype=np.float32)
t = tf.convert_to_tensor(arr)
print(tf.reshape(t, (3, 4)).shape)   # (3, 4)
print(tf.reshape(t, (-1, 2)).shape)  # (6, 2)  -1 means 'work it out'
print(t.numpy()[:3])                 # back to NumPy: [0. 1. 2.]

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers

model = keras.Sequential([
    keras.Input(shape=(4,)),
    layers.Dense(16, activation="relu"),
    layers.Dense(8, activation="relu"),
    layers.Dense(3, activation="softmax"),   # 3 classes
])
model.compile(optimizer="adam",
              loss="sparse_categorical_crossentropy",
              metrics=["accuracy"])
model.summary()

Problem type              Output layer                 Loss
Yes/No (binary)           Dense(1, 'sigmoid')          binary_crossentropy
One of many classes       Dense(n, 'softmax')          sparse_categorical_crossentropy (integer labels)
Predict a number          Dense(1) (no activation)     mse or mae

Getting this pairing right fixes many 'model won't learn' problems.

# pip install tensorflow scikit-learn
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler

X, y = load_iris(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42, stratify=y)
scaler = StandardScaler().fit(X_train)
X_train, X_test = scaler.transform(X_train), scaler.transform(X_test)

history = model.fit(X_train, y_train, epochs=50, batch_size=16,
                    validation_split=0.2, verbose=0)
loss, acc = model.evaluate(X_test, y_test, verbose=0)
print(f"test accuracy: {acc:.2f}")

import numpy as np

probs = model.predict(X_test[:3], verbose=0)
print(probs.round(2))             # probability for each class
print(np.argmax(probs, axis=1))   # predicted class ids
print(y_test[:3])                 # true class ids

model.save("iris_model.keras")
# reload later:
# from tensorflow import keras
# model = keras.models.load_model("iris_model.keras")

First TensorFlow Project Plan
Dataset: Iris or another small public dataset
Problem type: binary / multi-class / regression
Input shape: ______  Output: ______
Baseline: simple model from scikit-learn (e.g. logistic regression)
Keras model: 1-2 hidden Dense layers
Metric: accuracy / MAE
Goal: match or beat the baseline on test data
Write-up: what worked, what did not

Experiment Log
Run  Layers      Units   Epochs  Batch  Val acc  Test acc  Notes
1    2 Dense     16,8    50      16     ____     ____      baseline
2    2 Dense     32,16   50      16     ____     ____
3    + Dropout   32,16   80      16     ____     ____
Change only one thing per run.

TensorFlow Basics Checklist
☐ Create tensors and read shape/dtype
☐ Convert between NumPy and tensors
☐ Build a Sequential model
☐ Pick output layer and loss correctly
☐ compile, fit, evaluate, predict
☐ Read a training history plot
☐ Spot overfitting
☐ Save and load a model