zhuyitan / igtd Goto Github PK
View Code? Open in Web Editor NEWImage Generator for Tabular Data (IGTD): Converting Tabular Data to Images for Deep Learning Using Convolutional Neural Networks
License: MIT License
Image Generator for Tabular Data (IGTD): Converting Tabular Data to Images for Deep Learning Using Convolutional Neural Networks
License: MIT License
Hello,
I ran the Examples_Of_Table_To_Image_Conversion.py file and it produced the expected output images. I now am attempting to create my own images, but when I add my own test data, and follow the same steps as Examples_Of_Table_To_Image_Conversion.py, it produces empty images and data text files. The script runs otherwise as expected, and produces all other images and text files correctly with no error. Is there a way I can debug what is going on with the algo?
When using the image transformation, depending on the number of columns (for example ~20 thousand colons), I encounter quite long algorithm run times, for example about 30 hours, but as a result, I get a crash. I got around this by reducing the number of columns. What are your suggestions, what would you recommend?
Thank in advance
I have read the paper in nature and I am unsure if the image created needs to be symmetrical? If I have 200 features can I create an image of 20 x 10 or does it need to be symmetrical by adding padding features so it has an integer square root?
@sparcycram The converted images do not need to be symmetrical. A size of 20 x 10 will be perfectly fine.
@maxwell1386 To include new data in image conversion, you will need to combine new data with old data and use the max and min of combined data for normalization.
Originally posted by @zhuyitan in #1 (comment)
What is the final image for training a CNN and where can it be called from the Test_1 and Test_2 directories?
Not necessarily an issue and more of a question...
Would it be possible to convert from the image back to panel data?
The algorithm saves the image as 1_images. Is this the same order as it gets from Excel?
Results.pkl in Test_1 and Test_2 directories should store the original tabular data, the generated image data, and the names of samples. It does not display all these files when pickle.load() is executed.
The results.pkl file only seems to contain the original tabular data and nothing else but the IGTD seems to have worked as I have optimized_feature_ranking.png image along with some others. If I want to use the CNN regressor provided am I meant to use one of the png files generated or the contents of the Results.pkl file?
Hello,
I'm trying to use your code for prediction based on chemistry descriptors. How can I create such images for test set, when I've already obtained model based on training set presented with IGTD? In different words, after creation of images are there any settings which I should give to test set?
Hi !
I have a project about Binary Classification with Tabular data. Data contains 3 columns as input and 1 column as output. Also the number of rows is 6646.
I want bigger matrix. What I mean is because of the 3 inputs, generated images only becomes like 1-3 or 3-1. For example I want 4*4 images which generated from 3 inputs.
I want images like this :
Is there any way to do this ?
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.