Automatic tag identification for color themes

1. A method comprising:

receiving, by a processing device, an indication of a color theme, the indication of the color theme comprising an indication of multiple colors in a color space;

generating, by the processing device, a color theme embedding for the color theme by providing the indication of the multiple colors to a convolutional neural network that generates the color theme embedding including a vector encoding of possible tags for the color theme from the indication of the multiple colors;

generating, by the processing device, an initial tag based on a user setting;

determining, by the processing device, multiple tags for the color theme by providing the color theme embedding to a Long Short Term Memory (LSTM) recurrent neural network that generates multiple tags for the color theme from the color theme embedding and the initial tag, each of the multiple tags comprising at least one word that describes the color theme, the LSTM recurrent neural network generating multiple tags in multiple iterations until a product of probabilities for the multiple tags in one of the multiple iterations is less than a threshold amount, the multiple tags for the color theme including one tag generated in each of the multiple iterations, the LSTM recurrent neural network including an embedding map that maps tags to an M-dimensional space based on which tags appear together in color themes, each tag of the multiple tags being one of a set of tags that includes a threshold number of known tags for corresponding color themes;

determining, by the processing device, a sequence for the multiple tags by biasing the LSTM recurrent neural network;

automatically associating, by the processing device, the multiple tags for the color theme with the color theme; and

displaying, by the processing device, the multiple tags for the color theme as corresponding to the color theme.

2. The method as described in claim 1 , the convolutional neural network having a convolutional layer that includes multiple convolutional filters, the generating the color theme embedding including using a different one of the multiple convolutional filters to analyze each different pair of colors in the multiple colors.

3. The method as described in claim 1 , the receiving comprising receiving the multiple colors via user selection of the multiple colors on a color wheel display representing the color space.

4. The method as described in claim 1 , the LSTM recurrent neural network having as initial inputs a pre-configured initial tag and the color theme embedding.

5. The method as described in claim 1 , further comprising training the LSTM recurrent neural network using known color themes and corresponding user-selected tags.

6. A non-transitory computer-readable storage medium storing executable instructions, which when executed by a processing device, cause the processing device to perform operations comprising:

receiving an indication of a color theme, the indication of the color theme comprising an indication of multiple colors in a color space;

generating, using a convolutional neural network, a color theme embedding including a vector encoding of possible tags for the color theme from the indication of the multiple colors;

generating an initial tag based on a user setting;

using the color theme embedding to determine multiple tags for the color theme and using a Long Short Term Memory (LSTM) recurrent neural network to generate multiple tags in multiple iterations from the color theme embedding and the initial tag until a product of probabilities for the multiple tags in one of the multiple iterations is less than a threshold amount, the multiple tags for the color theme including one tag generated in each of the multiple iterations, and the multiple tags for the color theme being automatically associated with the color theme, the LSTM recurrent neural network including an embedding map that maps tags to an M-dimensional space based on which tags appear together in color themes, each tag of the multiple tags being one of a set of tags that includes a threshold number of known tags for corresponding color themes;

determining a sequence for the multiple tags by biasing the LSTM recurrent neural network; and

displaying the multiple tags in the sequence for the color theme as corresponding to the color theme.

7. The non-transitory computer-readable storage medium of claim 6 , wherein generating the color theme embedding further comprises using a different one of multiple convolutional filters of a convolutional layer of the convolutional neural network to analyze each different pair of colors in the multiple colors to generate the color theme embedding.

8. The non-transitory computer-readable storage medium of claim 6 , wherein the LSTM recurrent neural network has initial inputs that include a pre-configured initial tag and the color theme embedding.

9. In a digital medium environment to generate color theme tags, a computing device comprising:

a processor; and

computer-readable storage media having stored multiple instructions that, responsive to execution by the processor, cause the processor to perform operations comprising:

receiving an indication of a color theme, the indication of the color theme comprising an indication of multiple colors in a color space;

generating a color theme embedding for the color theme by providing the indication of the multiple colors to a convolutional neural network that generates the color theme embedding including a vector encoding of possible tags for the color theme from the indication of the multiple colors;

generating an initial tag based on a user setting;

determining one or more tags for the color theme by providing the color theme embedding to a Long Short Term Memory (LSTM) recurrent neural network that generates multiple tags for the color theme from the color theme embedding and the initial tag, each of the multiple tags comprising at least one word that describes the color theme, the LSTM recurrent neural network generating multiple tags in multiple iterations until a product of probabilities for the multiple tags in one of the multiple iterations is less than a threshold amount, the multiple tags for the color theme including one tag generated in each of the multiple iterations, the LSTM recurrent neural network including an embedding map that maps tags to an M-dimensional space based on which tags appear together in color themes, each tag of the multiple tags being one of a set of tags that includes a threshold number of known tags for corresponding color themes;

determining a sequence for the multiple tags by biasing the LSTM recurrent neural network; and

automatically associating the multiple tags for the color theme with the color theme.

10. The computing device as described in claim 9 , convolutional neural network having a convolutional layer that includes multiple convolutional filters, the generating the color theme embedding including using a different one of the multiple convolutional filters to analyze each different pair of colors in the multiple colors.

11. The computing device as described in claim 9 , the operations further comprising:

receiving user selection of at least one tag of the one or more tags; and

saving an indication of the at least one tag as corresponding to the color theme.

12. The computing device as described in claim 9 , wherein the LSTM recurrent neural network has a pre-configured initial tag and the color theme embedding as initial inputs.

13. The computing device as described in claim 9 , the operations further comprising training the LSTM recurrent neural network using known color themes and corresponding user-selected tags.

14. The method as described in claim 1 , wherein the set of tags include most commonly appearing tags in a set of training data used to train the LSTM recurrent neural network.

15. The method as described in claim 1 , wherein the initial tag is a most commonly appearing tag in a set of training data used to train the LSTM recurrent neural network.

16. The computing device as described in claim 9 , the operations further comprising:

storing a record of the color theme as being associated with the multiple tags for the color theme.

17. The method as described in claim 1 , further comprising passing the multiple tags through a fully connected layer to obtain a vector in a vocabulary space of possible tags and passing the multiple tags through a softmax activation to generate probabilities for each tag as an M-dimensional vector output.

18. The method as described in claim 17 , wherein the each of the multiple tags from the LSTM recurrent neural network corresponds to a component of the M-dimensional vector output having a value greater than 0.

19. The method as described in claim 18 , wherein the each of the multiple tags corresponds to a component of the M-dimensional vector output by the softmax activation having a largest value.

20. The method as described in claim 1 , wherein the multiple colors include five colors, each of the five colors represented as a set of three values including a red value, a green value, and a blue value.