GitLab

This reverts merge request !29

Set default optimizer, updated cnntrainer to replace unsupported optimizer by default, added JUnit test for unsupported train parameters

Added time measurements for training and testing, restricted use of accuracy for losses different than euclidean loss since it is not supported by the euclidean loss in Caffe2

Corrected dropout and output layer (include sigmoid). Added learning_rate_decay again since it is indeed supported in Caffe2

Set default (directories) lmdb datasets to 'train_lmdb' and 'test_lmdb'. Added check for lmdb datasets.

Removed template and generation for CNNBufferFile since it is not needed to read the Caffe2 protobuf files.

Modified CNNTrainer.ftl: improved way to enter the optimizer parameters in template since the support checker for training parameters is already implemented.
Corrected  order of training parameters in target files based on the modification.

CNNCreator.ftl: Modified path constants to make their value dynamic instead of hard coded during code generation. Modified methods save_net and load_net.

Added cnntrainer which calls the "train" function from cnncreator with the specified training parameters.

Added indented block in all layer templates due to changes made in CNNCreator.ftl in order to avoid a python error

Bugfix: ${input} in convolution template (case: input same as data input) should not be a reference anymore due to the new version of the add_input function.
Test must be made with Blob "data" always due to the new version of the add_input function.

-Deleted dummy data and label
-Added correct function add_input so lmdb data sets can be read as input for the networks.
-Save the deployment model instead of the test model to avoid including the lmdb reader.
-Adapted functions so they can take the data and label values from the function add_input.
-Included training parameters EPOCHS and BATCH_SIZE; they do not apply for testing.
-Included a proper image test.
-Improved print messages.