Tensorflow Utils

This is tensorflow utils

from preeminence_utils import tf_utils
model = tf_utils.Model()

Helper functions for using while making a neural network using tensorflow

class tf_utils.Model
get_latest_checkpoint(checkpoint_path='./model_weights/')

Get the name of the latest checkpoint in the checkpoints directory in order to load the latest weights to continue training for that point.

Parameters:checkpoint_path – Custom directory where checkpoints are saved
Returns:
graph_info()

Get ops in the graph

Returns:Graph ops
init()

Initialise a new model and return its graph. This function will spawn a new graph and return it. You’ll have to set it to default graph in order to add ops to it. Sample: model = tf_utils.Model() model_graph = model.init().as_default()

Returns:Returns a new graph.
next_batch(data, batch_start, batch_size)

Get next batch from the training data This should be a generator function :/

Parameters:
  • data – Training data
  • batch_start – batch start index
  • batch_size – size of the batch
Returns:

restore_weights(checkpoint_path='./model_weights/')

Restore weights from the checkpoint path to the latest checkpoint to resume training from that point.

Parameters:checkpoint_path – Custom directory where checkpoints are saved
Returns:
save_weights(checkpoint_path=None, checkpoint_number=None)

Save the current weights of the model to disk at the checkpoint path. :param checkpoint_path: Custom directory where checkpoints are saved :param checkpoint_number: Custom number to append at the end of checkpoint :return:

session()

Create and return a new session for training.

Returns:New session object
train(ops, x, y, x_data, y_data, num_epochs=1, batch_size=1)

Training function. Executes the graph on a given dataset.

Parameters:
  • ops – Graph ops to be calculated and returned. Must be [optimiser_op,loss_op]
  • x – placeholder tensor for x
  • y – placeholder tensor for y
  • x_data – Training data to be fed into x
  • y_data – Training data to be fed into y
  • num_epochs – Number of epochs to be executed
  • batch_size – Size of a batch to be fed at a time
Returns:

Nothing

visualise(logdir='./logs')

Save graph summary in the logidr to be visualised by tensorboard. Summaries for individual ops to be added.

Parameters:logdir – Destination for storing graph logs. ./logs by default
Returns:Nothing