Tensorflow Cheat Sheet For Deep Learning Model Building
Uploaded by
sh.ashfaqueme49Tensorflow Cheat Sheet For Deep Learning Model Building
Uploaded by
sh.ashfaqueme49@ @ @ @ @
da da da da da
ta ta ta ta ta
sc sc sc sc sc
ie ie ie ie ie
nc nc nc nc nc
eb eb eb eb eb
ra ra ra ra ra
in in in in in
@ @ @ @ @
da da da da da
ta ta ta ta ta
sc sc sc sc sc
ie ie ie ie ie
nc nc nc nc nc
eb eb eb eb eb
Save for later reference
for Deep
ra ra ra ra ra
Learning
DATA SCIENCE BRAIN
in in in in in
@datasciencebrain
@ @ @ @ @
da da da da da
ta ta ta ta ta
sc sc sc sc sc
ie ie ie ie ie
nc nc nc nc nc
Model Building
CHEAT SHEET
TENSORFLOW
eb eb eb eb eb
ra ra ra ra ra
in in in in in
Our website: deepakjosecodes.com
01 FEEDFORWARD NEURAL NETWORK
in
in
in
model = models.Sequential()
ra
ra
ra
eb
eb
eb
model.add(layers.Flatten(input_shape=(input_size,))) #
nc
nc
nc
ie
ie
ie
Adjust input_size based on your data
sc
sc
sc
ta
ta
ta
da
da
da
in
in
in
@
@
ra
ra
ra
# Add hidden layers
eb
eb
eb
nc
nc
nc
model.add(layers.Dense(128, activation='relu'))
ie
ie
ie
sc
sc
sc
model.add(layers.Dropout(0.2)) # Optional: Add
ta
ta
ta
da
da
da
dropout for regularization
in
in
in
@
@
ra
ra
ra
eb
eb
eb
nc
nc
nc
# Add output layer
ie
ie
ie
sc
sc
sc
model.add(layers.Dense(output_size,
ta
ta
ta
da
da
da
in
in
in
activation='softmax')) # Adjust output_size based on
@
@
ra
ra
ra
eb
eb
eb
your problem
nc
nc
nc
ie
ie
ie
sc
sc
sc
ta
ta
ta
model.compile(optimizer='adam',
da
da
da
in
in
in
@
@
ra
ra
ra
loss='sparse_categorical_crossentropy', # Use
eb
eb
eb
nc
nc
nc
'categorical_crossentropy' for one-hot encoded labels
ie
ie
ie
sc
sc
sc
metrics=['accuracy'])
ta
ta
ta
da
da
da
@
DATA SCIENCE BRAIN
@datasciencebrain 2
Our website: deepakjosecodes.com
02 CONVOLUTIONAL NEURAL NETWORK
in
in
in
model = models.Sequential()
ra
ra
ra
eb
eb
eb
model.add(layers.Conv2D(32, (3, 3), activation='relu',
nc
nc
nc
ie
ie
ie
input_shape=(img_height, img_width, channels)))
sc
sc
sc
ta
ta
ta
model.add(layers.MaxPooling2D((2, 2)))
da
da
da
in
in
in
@
@
ra
ra
ra
eb
eb
eb
nc
nc
nc
# Add more convolutional and pooling layers as
ie
ie
ie
sc
sc
sc
needed
ta
ta
ta
da
da
da
in
in
in
@
@
ra
ra
ra
eb
eb
eb
model.add(layers.Flatten())
nc
nc
nc
ie
ie
ie
model.add(layers.Dense(128, activation='relu'))
sc
sc
sc
ta
ta
ta
model.add(layers.Dense(output_size,
da
da
da
in
in
in
@
@
ra
ra
ra
activation='softmax'))
eb
eb
eb
nc
nc
nc
ie
ie
ie
sc
sc
sc
model.compile(optimizer='adam',
ta
ta
ta
da
da
da
in
in
loss='sparse_categorical_crossentropy', in
@
@
ra
ra
ra
eb
eb
eb
metrics=['accuracy'])
nc
nc
nc
ie
ie
ie
sc
sc
sc
ta
ta
ta
da
da
da
@
DATA SCIENCE BRAIN
@datasciencebrain 3
Our website: deepakjosecodes.com
03 RECURRENT NEURAL NETWORK
in
in
in
model = models.Sequential()
ra
ra
ra
eb
eb
eb
model.add(layers.SimpleRNN(128,
nc
nc
nc
ie
ie
ie
sc
sc
sc
activation='relu', input_shape=(timesteps,
ta
ta
ta
da
da
da
in
in
in
features)))
@
@
ra
ra
ra
eb
eb
eb
nc
nc
nc
ie
ie
ie
sc
sc
sc
# Add more recurrent layers or use LSTM/GRU
ta
ta
ta
da
da
da
layers
in
in
in
@
@
ra
ra
ra
eb
eb
eb
nc
nc
nc
ie
ie
ie
model.add(layers.Dense(output_size,
sc
sc
sc
ta
ta
ta
da
da
activation='softmax')) da
in
in
in
@
@
ra
ra
ra
eb
eb
eb
nc
nc
nc
ie
ie
ie
model.compile(optimizer='adam',
sc
sc
sc
ta
ta
ta
da
da
da
loss='sparse_categorical_crossentropy',
in
in
in
@
@
ra
ra
ra
eb
eb
eb
metrics=['accuracy'])
nc
nc
nc
ie
ie
ie
sc
sc
sc
ta
ta
ta
da
da
da
@
DATA SCIENCE BRAIN
@datasciencebrain 4
Our website: deepakjosecodes.com
04 LONG SHORT-TERM MEMORY
in
in
in
model = models.Sequential()
ra
ra
ra
eb
eb
eb
model.add(layers.LSTM(128, activation='relu',
nc
nc
nc
ie
ie
ie
sc
sc
sc
input_shape=(timesteps, features)))
ta
ta
ta
da
da
da
in
in
in
@
@
ra
ra
ra
eb
eb
eb
# Add more LSTM layers if needed
nc
nc
nc
ie
ie
ie
sc
sc
sc
ta
ta
ta
da
da
da
model.add(layers.Dense(output_size,
in
in
in
@
@
ra
ra
ra
eb
eb
eb
activation='softmax'))
nc
nc
nc
ie
ie
ie
sc
sc
sc
ta
ta
ta
da
da
model.compile(optimizer='adam', da
in
in
in
@
@
ra
ra
ra
eb
eb
eb
loss='sparse_categorical_crossentropy',
nc
nc
nc
ie
ie
ie
metrics=['accuracy'])
sc
sc
sc
ta
ta
ta
da
da
da
in
in
in
@
@
ra
ra
ra
eb
eb
eb
nc
nc
nc
ie
ie
ie
sc
sc
sc
ta
ta
ta
da
da
da
@
DATA SCIENCE BRAIN
@datasciencebrain 5
Our website: deepakjosecodes.com
05 GATED RECURRENT UNIT
in
in
in
model = models.Sequential()
ra
ra
ra
eb
eb
eb
model.add(layers.GRU(128, activation='relu',
nc
nc
nc
ie
ie
ie
sc
sc
sc
input_shape=(timesteps, features)))
ta
ta
ta
da
da
da
in
in
in
@
@
ra
ra
ra
eb
eb
eb
# Add more GRU layers if needed
nc
nc
nc
ie
ie
ie
sc
sc
sc
ta
ta
ta
da
da
da
model.add(layers.Dense(output_size,
in
in
in
@
@
ra
ra
ra
eb
eb
eb
activation='softmax'))
nc
nc
nc
ie
ie
ie
sc
sc
sc
ta
ta
ta
da
da
model.compile(optimizer='adam', da
in
in
in
@
@
ra
ra
ra
eb
eb
eb
loss='sparse_categorical_crossentropy',
nc
nc
nc
ie
ie
ie
metrics=['accuracy'])
sc
sc
sc
ta
ta
ta
da
da
da
in
in
in
@
@
ra
ra
ra
eb
eb
eb
nc
nc
nc
ie
ie
ie
sc
sc
sc
ta
ta
ta
da
da
da
@
DATA SCIENCE BRAIN
@datasciencebrain 6
Our website: deepakjosecodes.com
06 TRANSFER LEARNING(E.G., VGG16)
from tensorflow.keras.applications import VGG16
in
in
in
ra
ra
ra
eb
eb
eb
nc
nc
nc
# Load pre-trained VGG16 model without the top layer
ie
ie
ie
sc
sc
sc
base_model = VGG16(weights='imagenet', include_top=False,
ta
ta
ta
da
da
da
input_shape=(img_height, img_width, channels))
in
in
in
@
@
ra
ra
ra
eb
eb
eb
nc
nc
nc
# Freeze convolutional layers
ie
ie
ie
sc
sc
sc
for layer in base_model.layers:
ta
ta
ta
layer.trainable = False
da
da
da
in
in
in
@
@
ra
ra
ra
eb
eb
eb
model = models.Sequential()
nc
nc
nc
ie
ie
ie
model.add(base_model)
sc
sc
sc
ta
ta
ta
da
da
da
in
in
in
@
@
# Add custom classification layers
ra
ra
ra
eb
eb
eb
model.add(layers.Flatten())
nc
nc
nc
ie
ie
model.add(layers.Dense(256, activation='relu'))
ie
sc
sc
sc
ta
ta
ta
model.add(layers.Dropout(0.5))
da
da
da
in
in
model.add(layers.Dense(output_size, activation='softmax')) in
@
@
ra
ra
ra
eb
eb
eb
nc
nc
nc
ie
ie
ie
model.compile(optimizer='adam',
sc
sc
sc
loss='sparse_categorical_crossentropy',
ta
ta
ta
da
da
da
metrics=['accuracy'])
@
DATA SCIENCE BRAIN
@datasciencebrain 7
Our website: deepakjosecodes.com
07 BATCH NORMALIZATION
model.add(layers.BatchNormalization())
in
in
in
ra
ra
ra
eb
eb
eb
nc
nc
nc
08 DATA AUGMENTATION
ie
ie
ie
sc
sc
sc
ta
ta
ta
from tensorflow.keras.preprocessing.image import
da
da
da
in
in
in
@
@
ra
ra
ra
ImageDataGenerator
eb
eb
eb
nc
nc
nc
ie
ie
ie
datagen = ImageDataGenerator(
sc
sc
sc
ta
ta
ta
rotation_range=20,
da
da
da
in
in
in
@
@
width_shift_range=0.2,
ra
ra
ra
eb
eb
eb
height_shift_range=0.2,
nc
nc
nc
ie
ie
ie
horizontal_flip=True,
sc
sc
sc
ta
ta
ta
shear_range=0.2
da
da
da
in
in
in
)
@
@
ra
ra
ra
eb
eb
eb
nc
nc
datagen.fit(X_train) # X_train is your training data nc
ie
ie
ie
sc
sc
sc
ta
ta
ta
da
da
da
in
in
model.fit(datagen.flow(X_train, y_train, batch_size=batch_size), in
@
@
ra
ra
ra
eb
eb
eb
epochs=epochs)
nc
nc
nc
ie
ie
ie
sc
sc
sc
ta
ta
ta
da
da
da
@
DATA SCIENCE BRAIN
@datasciencebrain 8
Our website: deepakjosecodes.com
09 EARLY STOPPING
from tensorflow.keras.callbacks import EarlyStopping
in
in
in
ra
ra
ra
eb
eb
eb
nc
nc
nc
early_stopping = EarlyStopping(monitor='val_loss', patience=3,
ie
ie
ie
sc
sc
sc
restore_best_weights=True)
ta
ta
ta
da
da
da
in
in
in
@
@
ra
ra
ra
model.fit(X_train, y_train, epochs=epochs, validation_data=
eb
eb
eb
nc
nc
nc
(X_val, y_val), callbacks=[early_stopping])
ie
ie
ie
sc
sc
sc
ta
ta
ta
10
da
da
da
LEARNING RATE SCHEDULER
in
in
in
@
@
ra
ra
ra
eb
eb
eb
from tensorflow.keras.callbacks import LearningRateScheduler
nc
nc
nc
ie
ie
ie
sc
sc
sc
ta
ta
ta
def scheduler(epoch, lr):
da
da
da
in
in
in
if epoch % 10 == 0 and epoch != 0:
@
@
ra
ra
ra
eb
eb
eb
return lr * 0.9
nc
nc
else: nc
ie
ie
ie
sc
sc
sc
return lr
ta
ta
ta
da
da
da
in
in
in
@
@
ra
ra
ra
lr_scheduler = LearningRateScheduler(scheduler)
eb
eb
eb
nc
nc
nc
ie
ie
ie
sc
sc
sc
model.fit(X_train, y_train, epochs=epochs, validation_data=
ta
ta
ta
da
da
da
(X_val, y_val), callbacks=[lr_scheduler])
@
DATA SCIENCE BRAIN
@datasciencebrain 9
Our website: deepakjosecodes.com
11 EARLY STOPPING
from sklearn.model_selection import GridSearchCV
in
in
in
ra
ra
ra
from tensorflow.keras.wrappers.scikit_learn import KerasClassifier
eb
eb
eb
nc
nc
nc
# Define your model creation function
ie
ie
ie
def create_model(optimizer='adam', hidden_units=128, dropout_rate=0.2):
sc
sc
sc
model = models.Sequential()
ta
ta
ta
model.add(layers.Flatten(input_shape=(input_size,)))
da
da
da
in
in
in
@
@
# Add hidden layers
ra
ra
ra
eb
eb
eb
model.add(layers.Dense(hidden_units, activation='relu'))
nc
nc
nc
model.add(layers.Dropout(dropout_rate))
ie
ie
ie
sc
sc
sc
# Add output layer
ta
ta
ta
model.add(layers.Dense(output_size, activation='softmax'))
da
da
da
in
in
in
@
@
model.compile(optimizer=optimizer,
ra
ra
ra
eb
eb
eb
loss='sparse_categorical_crossentropy',
nc
nc
nc
metrics=['accuracy'])
return model
ie
ie
ie
sc
sc
sc
ta
ta
ta
# Create a KerasClassifier with your model creation function
da
da
da
model = KerasClassifier(build_fn=create_model, epochs=10, batch_size=32, verbose=0)
in
in
in
@
@
ra
ra
ra
eb
eb
eb
# Define the hyperparameters to search
param_grid = {
nc
nc
nc
'optimizer': ['adam', 'sgd', 'rmsprop'],
ie
ie
ie
sc
sc
sc
'hidden_units': [64, 128, 256],
ta
ta
ta
'dropout_rate': [0.2, 0.5, 0.8]
da
da
da
}
in
in
in
@
@
ra
ra
ra
# Use GridSearchCV for hyperparameter search
eb
eb
eb
grid = GridSearchCV(estimator=model, param_grid=param_grid, cv=3)
nc
nc
nc
grid_result = grid.fit(X_train, y_train)
ie
ie
ie
sc
sc
sc
ta
ta
ta
# Print the best parameters and corresponding accuracy
da
da
da
print("Best Parameters: ", grid_result.best_params_)
@
print("Best Accuracy: ", grid_result.best_score_)
DATA SCIENCE BRAIN
@datasciencebrain 10
Never Miss a Post!
Turn on the Notifications
Was it
helpful?
Checkout our YouTube Channel
for Machine Learning Projects and Other
amazing Data Science Related Content
youtube.com/@dsbrain
DATA SCIENCE BRAIN
@datasciencebrain
LIKE TO SUPPORT
SAVE FOR LATER
COMMENT
SHARE
Checkout Our
Other Posts
Save for later reference
You might also like
- Introduction To Genetic Algorithm Neural NetworksNo ratings yetIntroduction To Genetic Algorithm Neural Networks44 pages
- Exp. No.: I. Aim: AIML634P Neural Network Lab 2262034No ratings yetExp. No.: I. Aim: AIML634P Neural Network Lab 22620346 pages
- Deep Learning Lab Manual - IGDTUW - Vinisky Kumar100% (1)Deep Learning Lab Manual - IGDTUW - Vinisky Kumar33 pages
- Experiment 3 (A, B, C) (RNN) (Recuurent) (IMDB) )No ratings yetExperiment 3 (A, B, C) (RNN) (Recuurent) (IMDB) )11 pages
- Deep Neural Network for Image ClassificationNo ratings yetDeep Neural Network for Image Classification17 pages
- Deep Learning Fundamentals and ArchitecturesDeep Learning Fundamentals and ArchitecturesNo ratings yetDeep Learning Fundamentals and ArchitecturesDeep Learning Fundamentals and Architectures9 pages
- CCS355 Neural Networks and Deep Learning LabNo ratings yetCCS355 Neural Networks and Deep Learning Lab43 pages
- PRO Level Data Visualization Cheat SheetNo ratings yetPRO Level Data Visualization Cheat Sheet15 pages
- Pandas For Python Pro Level Cheat SheetNo ratings yetPandas For Python Pro Level Cheat Sheet14 pages
- The Working Day British English Student BWNo ratings yetThe Working Day British English Student BW6 pages
- Predictive Analytics: A Survey, Trends, Applications, Oppurtunities & ChallengesNo ratings yetPredictive Analytics: A Survey, Trends, Applications, Oppurtunities & Challenges5 pages
- Module 5 Conflict Styles and Leadership Styles Self Assessments WorksheetNo ratings yetModule 5 Conflict Styles and Leadership Styles Self Assessments Worksheet3 pages
- 2019 (George, Walker, & Monster) Does Strategic Planning Improve Organizational Performance A Meta-AnalysisNo ratings yet2019 (George, Walker, & Monster) Does Strategic Planning Improve Organizational Performance A Meta-Analysis10 pages
- Understanding The Need For Value EducationNo ratings yetUnderstanding The Need For Value Education18 pages
- Rubric For Written Communication Skills2No ratings yetRubric For Written Communication Skills22 pages
- (Rogers Brubaker) Ethnicity Without Groups100% (7)(Rogers Brubaker) Ethnicity Without Groups148 pages
- REport in Test Construction CHAPTER 2 TEST DEVELOPMENTNo ratings yetREport in Test Construction CHAPTER 2 TEST DEVELOPMENT11 pages
