+ 
+ [projector_config.pbtxt](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/tensorboard/plugins/projector/projector_config.proto)
+in the same directory as your checkpoint file.
+
+### Setup
+
+For in depth information on how to run TensorBoard and make sure you are
+logging all the necessary information, see
+[TensorBoard: Visualizing Learning](../get_started/summaries_and_tensorboard.md).
+
+To visualize your embeddings, there are 3 things you need to do:
+
+1) Setup a 2D tensor that holds your embedding(s).
+
+```python
+embedding_var = tf.get_variable(....)
+```
+
+2) Periodically save your model variables in a checkpoint in
+LOG_DIR.
+
+```python
+saver = tf.train.Saver()
+saver.save(session, os.path.join(LOG_DIR, "model.ckpt"), step)
+```
+
+3) (Optional) Associate metadata with your embedding.
+
+If you have any metadata (labels, images) associated with your embedding, you
+can tell TensorBoard about it either by directly storing a
+[projector_config.pbtxt](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/tensorboard/plugins/projector/projector_config.proto)
+in the LOG_DIR, or use our python API.
+
+For instance, the following projector_config.ptxt associates the
+word_embedding tensor with metadata stored in $LOG_DIR/metadata.tsv:
+
+```
+embeddings {
+ tensor_name: 'word_embedding'
+ metadata_path: '$LOG_DIR/metadata.tsv'
+}
+```
+
+The same config can be produced programmatically using the following code snippet:
+
+```python
+from tensorflow.contrib.tensorboard.plugins import projector
+
+# Create randomly initialized embedding weights which will be trained.
+vocabulary_size = 10000
+embedding_size = 200
+embedding_var = tf.get_variable('word_embedding', [vocabulary_size, embedding_size])
+
+# Format: tensorflow/tensorboard/plugins/projector/projector_config.proto
+config = projector.ProjectorConfig()
+
+# You can add multiple embeddings. Here we add only one.
+embedding = config.embeddings.add()
+embedding.tensor_name = embedding_var.name
+# Link this tensor to its metadata file (e.g. labels).
+embedding.metadata_path = os.path.join(LOG_DIR, 'metadata.tsv')
+
+# Use the same LOG_DIR where you stored your checkpoint.
+summary_writer = tf.summary.FileWriter(LOG_DIR)
+
+# The next line writes a projector_config.pbtxt in the LOG_DIR. TensorBoard will
+# read this file during startup.
+projector.visualize_embeddings(summary_writer, config)
+```
+
+After running your model and training your embeddings, run TensorBoard and point
+it to the LOG_DIR of the job.
+
+```python
+tensorboard --logdir=LOG_DIR
+```
+
+Then click on the *Embeddings* tab on the top pane
+and select the appropriate run (if there are more than one run).
+
+
+### Metadata
+Usually embeddings have metadata associated with it (e.g. labels, images). The
+metadata should be stored in a separate file outside of the model checkpoint
+since the metadata is not a trainable parameter of the model. The format should
+be a [TSV file](https://en.wikipedia.org/wiki/Tab-separated_values)
+(tab characters shown in red) with the first line containing column headers
+(shown in bold) and subsequent lines contain the metadata values:
+
+
+Word\tFrequency
+ Airplane\t345
+ Car\t241
+ ...
+
+
+There is no explicit key shared with the main data file; instead, the order in
+the metadata file is assumed to match the order in the embedding tensor. In
+other words, the first line is the header information and the (i+1)-th line in
+the metadata file corresponds to the i-th row of the embedding tensor stored in
+the checkpoint.
+
+Note: If the TSV metadata file has only a single column, then we don’t expect a
+header row, and assume each row is the label of the embedding. We include this
+exception because it matches the commonly-used "vocab file" format.
+
+### Images
+If you have images associated with your embeddings, you will need to
+produce a single image consisting of small thumbnails of each data point.
+This is known as the
+[sprite image](https://www.google.com/webhp#q=what+is+a+sprite+image).
+The sprite should have the same number of rows and columns with thumbnails
+stored in row-first order: the first data point placed in the top left and the
+last data point in the bottom right:
+
+| 0 | +1 | +2 | +
| 3 | +4 | +5 | +
| 6 | +7 | ++ |
|
+
+ |
+
+
+ |
+
| + Custom projection controls. + | ++ Custom projection of neighbors of "politics" onto "best" - "worst" vector. + | +
+
+
+## Mini-FAQ
+
+**Is "embedding" an action or a thing?**
+Both. People talk about embedding words in a vector space (action) and about
+producing word embeddings (things). Common to both is the notion of embedding
+as a mapping from discrete objects to vectors. Creating or applying that
+mapping is an action, but the mapping itself is a thing.
+
+**Are embeddings high-dimensional or low-dimensional?**
+It depends. A 300-dimensional vector space of words and phrases, for instance,
+is often called low-dimensional (and dense) when compared to the millions of
+words and phrases it can contain. But mathematically it is high-dimensional,
+displaying many properties that are dramatically different from what our human
+intuition has learned about 2- and 3-dimensional spaces.
+
+**Is an embedding the same as an embedding layer?**
+No; an embedding layer is a part of neural network, but an embedding is a more
+general concept.