TensorFlow library is imported with tf qualified name usually. Qualifying TensorFlow with the tf is a great idea to stay consistent with open-source TensorFLow Projects and other developers. After knowing about importing TensorFlow into Python source file, it is time for you to know about the method of using it efficiently. Feature vectors can be highly effective for machine learning. Each data item is loaded with feature vector. A list of vectors in known as matrix. A wide range of functions regarding statistical distributions are provided by TensorFlow located in tf.contrib.distributions,includes but certainly not limited to distributions such as Uniform, Gamma, Dirichlet, Chi2, Beta, Bernoulli etc. they are extremely important building blocks when it is the matter of building machine learning algorithm. You will find layer operations producing functions and related weight and bias variables inside tf.contrib.layers,. They come in use for creating different types of deep learning constructions. There are different functions for dropout layer, convolution layer, and batch normalization etc.In tf.contrib.layers.optimizersyou will find different types of optimizers like Momentum, SGD, Adagrad etc. They come in use for solving optimization issues regarding numeric analysis.

In tf.contrib.layers.regularizers module you will find regularizers like L1 and L2. They come in use for reducing the overfitting risk by penalizing large volume of features utilized in the model. For machine learning blocks, it comes in use as building blocks for example Ridge and Lasso Regression.Deep learning algorithms need gradients’ calculation for model optimization. TensorFlow provides provided plenty of initializers like Xavierinitializer in tf.contrib.layers.initializers,used for the weights for keeping gradients’ scale similar in all layers. TensorFlow gives a wide range of loss functions to choose from tf.contrib.losses, like sum of pairwise squares, sum of squares, hinge loss, log-loss, SoftMax cross entry, and sigmoid etc. if you want more variety of metrics like MSE, auc, accuracy, recall, and precision etc. in tf.contrib.metrics