data science tutorials and snippets prepared by greysweater42
Just kidding with pytorch ;) but still you will have to choose between these two (or even more) frameworks at the very early stage of the analysis.
Keras is a high-level framework for working with neural networks, written in Python and capable of running on top of either TensorFlow, Microsoft Cognitive Toolkit (CNTK) or Theano.
Yes, it’s going to be MNIST ;)
Based on Deep learning with Python, chapter 2.
You can use Sequential API or Model API (there is also Functional API, but I will not cover it as it is almost exactly the same Sequential API), they both work exactly the same, but as you can see, Sequential API's syntax is much shorter and Model API's syntax looks almost exactly the same as pytorch.
from tensorflow.keras.datasets import mnist
from tensorflow.keras.utils import to_categorical
(X_train, y_train), (X_test, y_test) = mnist.load_data()
X_train = X_train.reshape((60000, 28 * 28)).astype('float32') / 255
X_test = X_test.reshape((10000, 28 * 28)).astype('float32') / 255
y_train = to_categorical(y_train)
y_test = to_categorical(y_test)
# you can use Sequential API
import keras
network = keras.models.Sequential()
network.add(keras.layers.Dense(512, activation='relu'))
network.add(keras.layers.Dense(10, activation='softmax'))
# or Model API - up to you!
from tensorflow import nn, keras
class Network(keras.Model):
def __init__(self):
super(Network, self).__init__()
self.d1 = keras.layers.Dense(512, activation=nn.relu)
self.d2 = keras.layers.Dense(10, activation=nn.softmax)
def call(self, inputs):
x = self.d1(inputs)
return self.d2(x)
network = Network()
####
network.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])
network.fit(X_train, y_train, epochs=5, batch_size=128)
test_loss, test_acc = network.evaluate(X_test, y_test)
print(test_acc)
y_hat = network.predict(X_test) # and this is how you make predictions
The easiest way to run keras on GPU is to pull a tensorflow-gpu image with:
docker pull tensorflow/tensorflow:latest-gpu
start it up with:
docker run --gpus all -it --rm tensorflow/tensorflow:latest-gpu bash
and run your code in it. But first you should configure GPU on the machine you’re working at. Just follow this procedure (basically you should get to the point, when you can run nvidia-smi from your terminal).
if you still prefer R over Python, take a look at this introductory article: keras: deep learning in R
a very good book not only about keras, but about deep learning in general: Deep learning with Python
is there an easy way to read images into the network? TODO yes, with ImageDataGenerator
saving and loading networks? TODO
transfer learning? TODO
data augmentation? TODO
YOLO or object detection, in general TODO