Computer Vision News - January 2019

23 Focus on Computer Vision News Now, once the code run ends, you can go to the website at any time and see the experiment outcomes report. The outcomes of each run are stored separately, and for each run the report is organized as a navigation bar like the one to the left, which shows all the parameters you can look at. Under Charts, you can find graphs for the accuracy and loss metrics, and any other parameter you choose to define for monitoring in the log_parameters function -- the graphs are very similar to those of TensorFlow’s TensorBoard. Additionally, comet.ml stores the code of a run alongside the report of that run (under Code); the output is stored (under Output) and so are all the packages that were installed for that run (under Installed packages). This systematic storage of all possible data, from installed packages, the code, the network graph, the precise parameters next to the output -- enables you to better analyze your data. It is not unusual in the course of development to keep only the run outcomes, without meticulous notation of every last detail of all parameter and network settings, not to mention the different versions of packages installed. The result is that it becomes difficult to impossible to reliably reproduce your results later. Hyperas Hyperas is a convenient and simple package that is built for training deep learning network using Keras. Hyperas allows you to quickly train networks in different configurations and with different hyperperameter for selecting the optimal configuration for your data and network. Hyperas allows you to use the power of hyperopt without having to learn its syntax. All you need to do is set up your Keras model in the usual way with only a simple template of scroll brackets (which we'll immediately demonstrate) to define options and ranges for evaluation. . First, you need to install the package locally, on your own computer, like this: Let’s say you have the following network set up using Keras that you want to train: pip install hyperas Tip - Train Your Network