import os
import tensorflow as tf
from tensorflow.contrib.tensorboard.plugins import projector
from tensorflow.python.training.summary_io import SummaryWriterCache
def accuracy_of_minibatch(labels, predictions, name='accuracy'):
''' both inputs are `tf.int64` '''
with tf.name_scope('MinibatchAccuracy'):
accuracy = tf.reduce_mean(
tf.cast(
tf.equal(labels, predictions),
tf.float32
)
)
tf.summary.scalar(name, accuracy)
return accuracy
def visualize_embeddings(logdir, var_list, tsv_list):
assert len(var_list) == len(tsv_list), 'Inconsistent length of lists'
config = projector.ProjectorConfig()
for v, f in zip(var_list, tsv_list):
embedding = config.embeddings.add()
embedding.tensor_name = v.name
if f is not None:
_, filename = os.path.split(f)
meta_tsv = os.path.join(logdir, filename)
tf.gfile.Copy(f, meta_tsv)
embedding.metadata_path = filename # save relative path
writer = SummaryWriterCache.get(logdir)
projector.visualize_embeddings(writer, config)
