Name Scope Summary
Cloud ML export Code convention Export models using the convention supported by Google Cloud ML.
Evaluate function Function For functions that support evaluation using test data, define a function named evaluate.
Feed dict as positional arg Code convention Specify a feed dict as the second positional argument session.run, rather than as a keyword argument.
Flags Code convention Use a global FLAGS variable to let users list and redefine script parameters.
Function defined globals Code convention Use functions to define global variables used by a script.
Init functions Code convention Use functions named init_xxx to perform state or subsystem initialization.
Log model status function Function Define a function named log_model_status to log summaries and print updates to the user.
Loss operation Tensor/Operation Use a TensorFlow operation loss to represent the loss being minimized in a training operation.
PEP-8 style guide Code convention Follow the PEP 8 Python style guide unless otherwise specified.
Prepare, train, evaluate Workflow Build scripts that can be used in discrete prepare, train, and evaluate operations.
RUNDIR Workflow Use a new directory for each run to store training and other artifacts.
Save model function Function Use a function named save_model to create a checkpoint of a model.
Summaries operation Tensor/Operation Use TensorFlow event writers and summary operations to log training progress and other details useful to understanding the training operation.
Train function Function For training scripts, define a function named train that is responsible for the training loop.
Train operation Tensor/Operation Use a TensorFlow operation named train_op that implements the optimization used to train model variables.
Use default graph Code convention Rely on TensorFlow's default graph rather than use new graphs as a context manager.