Tensorflow Basic CSLT NLP Tensorflow Basic Google had

Tensorflow Basic CSLT NLP

Tensorflow Basic • Google had published tensorflow, their deep learning structure. • For our team as we had already learned theano, for easily starting and seek for fast speed, we want to change our mainly deep learning toolkit from theano to tensorflow.

Tensorflow Basic • Start:

Tensorflow Basic • Result:

Tensorflow Basic • Reduce Mean

Tensorflow Basic • tf. train has 9 child classes. Ø Optimizers ØOptimizer ØGradient. Descent. Optimizer ØAdagrad. Optimizer ØMomentum. Optimizer ØAdam. Optimizer ØFtrl. Optimizer ØRMSProp. Optimizer

Tensorflow Basic • tf. train has 9 child classes. Ø Gradient Computation Øgradients ØAggregation. Method Østop_gradient

Tensorflow Basic • tf. train has 9 child classes. Ø Gradient Clipping Øclip_by_value Øclip_by_norm Øclip_by_average_norm Øclip_by_global_norm Øglobal_norm

Tensorflow Basic • tf. train has 9 child classes. Ø Decaying the learning rate Øexponential_decay

Tensorflow Basic • tf. train has 9 child classes. Ø Moving Averages ØExponential. Moving. Average

Tensorflow Basic • tf. train has 9 child classes. Ø Coordinator and Queue. Runner (sequence run) ØCoordinator ØQueue. Runner Øadd_queue_runner Østart_queue_runners

Tensorflow Basic • tf. train has 9 child classes. Ø Summary Operations (to show your model) Øscalar_summary Øimage_summary Øhistogram_summary Øzero_fraction Ømerge_summary Ømerge_all_summaries

Tensorflow Basic • tf. train has 9 child classes. Ø Adding Summaries to Event Files (Also to show your model) ØSummary. Writer Øsummary_iterator

Tensorflow Basic • tf. train has 9 child classes. Ø Training utilities (training trick) Øglobal_step Øwrite_graph

Tensorflow Basic • tf. train. Gradient. Descent. Optimizer – apply_gradients – compute_gradients – get_slot (adam, Momentum need additional variables) – get_slot_names (adam, Momentum need additional variables) – minimize (equal to apply_gradients + compute_gradients)

Home Work • Write a dnn code in tensorflow.