-2

I have colors in RGB form. There are 4 columns

  1. 'accent_color'-> (0.6901960784313725, 0.14901960784313725, 0.10588235294117647)
  2. 'dominant_colors' -> [(0.6470588235294118, 0.16470588235294117, 0.16470588235294117), (0.0, 0.0, 0.0), (1.0, 1.0, 1.0)]
  3. 'bg_color' > (0.6470588235294118, 0.16470588235294117, 0.16470588235294117)
  4. 'fore_color' -> (0.0, 0.0, 0.0)

I want to use these values as features in the textual model. These colors will be used to determine whether the post has more probability of getting likes wrt colors. Any tips will be appreciated.

Sandeep Kumar Kushwaha
  • 167
  • 7

1 Answers1

2

This question is very broad to answer from the info you provided but I think, first you need to conduct some explanatory analysis to understand if your dataset can represent your class well. Before this analysis you need to convert this data into a tabular form.

I would represent each color group as "one hot encoded like" form so one row from your dataset looks like:

accent_color_R : 0.6901960784313725
accent_color_G : 0.14901960784313725
accent_color_B : 0.10588235294117647

dominant_colors_first_group_R : 0.6470588235294118
dominant_colors_first_group_G : 0.16470588235294117
dominant_colors_first_group_B : 0.16470588235294117
dominant_colors_second_group_R : 0.0
dominant_colors_second_group_G : 0.0
dominant_colors_second_group_B : 0.0
dominant_colors_third_group_R : 1.0
dominant_colors_third_group_G : 1.0
dominant_colors_third_group_B : 1.0

bg_color_R : 0.6470588235294118
bg_color_G : 0.16470588235294117
bg_color_B : 0.16470588235294117

fore_color_R : 0.0
fore_color_G : 0.0
fore_color_B : 0.0

class_label : (0 or 1) OR a continuous value that represent probability

So it is going to be 18 columns(features) table. Depends on what kind of dataset you have, It can be a liked / not_liked binary classification problem or can be a regression problem, for the regression your label is going to be a continuous value and it represents the probability so your class label is going to be in

[0,1] = {0<= class_label <= 1}

You can use neural network to implement regression. You can find plenty of informations if you type "implementing regression with neural networks" to Google.

Inputvector
  • 1,061
  • 10
  • 22
  • I have used a similar technique, I wanted to know if there is another way of representing the color data, because it gives me lots of columns. With my previous data, I already have 80+ columns. Anyway, till now, this looks like a good fix to experiment with. I will update here if I find anything useful. – Sandeep Kumar Kushwaha May 08 '22 at 18:18
  • 1
    You can use some dimension reduction methods (PCA) or statistical tests if you are not happy with the result but 80 columns is pretty normal. I don't know how to represent this data in a different way. – Inputvector May 08 '22 at 20:38