TensorFlow Complete Cheat Sheet

1. Fundamental Concepts & Core Modules

Concept/Module Description Import Statement

Tensors Multi-dimensional arrays jo TF ka basic data structure hai. N/A (Core concept)

tf.keras High-level neural networks API, TF ka official high-level API. from tensorflow import keras

tf.data Data input pipelines build karne ke liye. from tensorflow import data

tf.keras.layers Neural network layers (Dense, Conv2D, LSTM, etc.). from tensorflow.keras import layers

tf.keras.models Model creation aur management. from tensorflow.keras import models

tf.keras.optimizers Optimization algorithms (Adam, SGD, etc.). from tensorflow.keras import optimizers

tf.keras.losses Loss functions. from tensorflow.keras import losses

tf.keras.metrics Evaluation metrics. from tensorflow.keras import metrics

2. Tensor Operations & Creation

Function Description Example

tf.constant() Constant tensor banata hai. tf.constant([[1, 2], [3, 4]])

tf.Variable() Mutable tensor banata hai. tf.Variable(tf.random.normal([2, 2]))

tf.convert_to_tensor() NumPy array ya list ko tensor mein convert karta hai. tf.convert_to_tensor(numpy_array)

tf.random.normal() Normal distribution se random tensor. tf.random.normal([3, 3])

tf.zeros() Sab elements 0 wala tensor. tf.zeros([2, 3])

tf.ones() Sab elements 1 wala tensor. tf.ones([4, 2])

tf.range() Sequence of numbers ka tensor. tf.range(0, 10, 2)

 Function Description Example

tf.shape() Tensor ka shape return karta hai. tf.shape(tensor)

tf.reshape() Tensor ki shape change karta hai. tf.reshape(tensor, [4, 1])

tf.cast() Tensor ka data type change karta hai. tf.cast(int_tensor, tf.float32)

3. Neural Network Layers (tf.keras.layers)

Layer Class Description Example

layers.Dense() Fully connected layer. layers.Dense(64, activation='relu')

layers.Conv2D() 2D Convolutional layer (Images ke liye). layers.Conv2D(32, (3, 3), activation='relu')

layers.MaxPooling2D() Max pooling layer. layers.MaxPooling2D((2, 2))

layers.Flatten() Multi-dimensional tensor ko 1D mein flatten karta hai. layers.Flatten()

layers.Dropout() Overfitting reduce karne ke liye dropout layer. layers.Dropout(0.5)

layers.LSTM() LSTM layer (Sequences ke liye). layers.LSTM(50, return_sequences=True)

layers.Embedding() Word embeddings ke liye layer. layers.Embedding(10000, 128)

layers.BatchNormalization() Batch normalization layer. layers.BatchNormalization()

layers.Input() Model ka input layer define karta hai. layers.Input(shape=(28, 28))

4. Model Building & Training

Function/Method Description Example

keras.Sequential() Sequential model banata hai. keras.Sequential([...])

model.compile() Training se pehle configure karta hai. model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'

model.fit() Model ko train karta hai. model.fit(X_train, y_train, epochs=10, validation_data=(X_val, y_val))
 Function/Method Description Example

model.evaluate() Trained model ko evaluate karta hai. model.evaluate(X_test, y_test)

model.predict() New data par predictions. model.predict(new_data)

model.save() Trained model save karta hai. model.save('my_model.h5')

keras.models.load_model() Saved model load karta hai. keras.models.load_model('my_model.h5')

5. Optimizers, Losses & Metrics

Category Options Example

Optimizers 'adam', 'sgd', 'rmsprop' optimizers.Adam(learning_rate=0.001)

Loss Functions 'binary_crossentropy', 'categorical_crossentropy', 'mse', 'mae' losses.SparseCategoricalCrossentropy()

Metrics 'accuracy', 'precision', 'recall', 'auc' ['accuracy', metrics.Precision()]

6. Callbacks
Callback Description Example

keras.callbacks.EarlyStopping() Training early stop kar deta hai. EarlyStopping(patience=3, restore_best_weights=True)

keras.callbacks.ModelCheckpoint()
Best model save karta hai. ModelCheckpoint('best_model.h5', save_best_only=True)

keras.callbacks.ReduceLROnPlateau()
Learning rate reduce karta hai. ReduceLROnPlateau(factor=0.5, patience=2)

keras.callbacks.TensorBoard() TensorBoard visualization ke liye. TensorBoard(log_dir='./logs')

7. Advanced Features
Feature Description Usage

Custom Training Loops Manual training with tf.GradientTape(). Research aur advanced models ke liye

Transfer Learning Pre-trained models use karna. keras.applications (VGG16, ResNet)

Data Augmentation Training data artificially badhane ke liye. keras.preprocessing.image.ImageDataGenerator

TensorBoard Training visualize karne ke liye. tensorboard --logdir=./logs

TPU/GPU Support Hardware acceleration automatic. tf.config.list_physical_devices('GPU')