index.md

Logistic Regression (quick intro) back

Logistic Regression - one of the ways to build a linear model for the available training set data (X_i,Y_i): X=>Y.

On the other hand, this method (Linear Regression) can be regarded as a degenerate case of MultiLayer Fully Connectad Netwok.

As parameters LogisticRegression model has a matrix of weights with size NxK

and biases vector with size Nx1, where N - Size of feature vector (if feature

vector is gray-scale image with size LxL, then N=L*L), and K - is number of

classes.

Non-linearity in the model is selected as SoftMax: Y=SoftMax(W*X + B).

CrossEntropy is used as a loss function for multi categorical classification.

MNIST image database used as training set.

---

Next: 2.1 Logistic Regression with Theano

---

• 2.1 Logistic Regression with Theano
  • 2.1.1 Theano: Blocks Framework
  • 2.1.2 Theano: Lasagne Framework
  • 2.1.3 Theano: Keras Framework
• 2.2 Logistic Regression with Torch
• 2.3 Logistic Regression with Caffe
• 2.4 Logisitc Regression with Tensorflow
• 2.5 Logistic Regression with Deeplearning4J