Prepare your dataset for machine learning (Coding TensorFlow)
ฝัง
- เผยแพร่เมื่อ 18 ต.ค. 2024
- Interested in learning how to use JavaScript in the browser? In the last episode of Coding TensorFlow, we showed you a very basic ML scenario in the browser that predicted future values. Shaping your data and getting it ready for training is an important step in data science. Watch as Laurence shows how you can prepare your own raw data and get it ready for a machine learning model. In the next episode, we’ll train a neural network with this data, and go into designing the network, and doing classification given the trained model. Subscribe to the channel for more!
TensorFlow.js → bit.ly/2HN1wj5
Try TensorFlow.js in your browser (pt.1) → bit.ly/2uV88aA
Coding TensorFlow playlist → bit.ly/2zoZfvt
Subscribe to the TensorFlow channel → bit.ly/TensorFlow1
The precise definition of a MVP, Sir, many thanks for the clarrity of your explanations and details to deep dive into this awesome library!
I think this should also cover how to deal with messy data, like missing values and how the deal with categorical input data.
For those who are interested, "the coding train" made an extensive tutorial on how to train a color classifier with tensorflow js
Nice suggestion, thanks!
I hope I understand this correctly: you first split the data per flower type (classification) then split those into 80%-20% training sets/test sets and then 'glue' them back together, 'unsplitting' the classifications. Is this to make sure you split the data proportionally for each flower type? That is, to avoid that the training data contains all the '0' and '1' flowers and the test data, the '2' flowers?
And I suppose the HotOne conversion, is to have a separate output neuron for each flower type? (rather than have 1 output neuron where the value of 0 is for flower '0', value of .5 is for flower '1' and value 1 is for flower '2')? I guess, only with the split neurons that get 'hot' per flower type, would you be able to tell that a flower might match 2 (or more) flower types at once.
As usual, Great work. Thank you :)
Laurence would you save us time and share the code ?
konrad jj definitely! Will blog it next week
I found the data here: github.com/tensorflow/tfjs-examples/blob/master/iris/data.js
More examples in the repository besides IRIS.
If this example is used, remember it's for node.js, so remove the 'import' line, and remove the 'export' prefixes to functions and variables. It then runs fine in the browser.
Great video and very helpful.. Thank you!
Thank you for this awesome video and series, I am waiting for next video
Coming soon! :)
Hi,
I have a problem statement, in which the input data is an image + few sensor data associated with each image.
i have few thousands of data collected now, but i'm not able to figure out how to make the input data set for training.
the output should be 0,1,2 for each pixel in an image.
i will annotate the data telling how each pixel in output of training set will look like( i.e 0,1,2 ).
Can you help me with the input data shape that i have to create?
Tensorflow epochs pauses working while fit process when minimize browser window or switch to another browser tab. I waited for 2-3 hours but epoch not logs to console on onEpochEnd callback. When i switch tab, eochs continues and logs again. How may i set for fit continues in the background?
At Line: 97, shouldn't it be `const yTest = ys.slice([numTrainExamples, 0], [numTestExamples, IRIS_NUM_CLASSES]);`?
I think you're right. At least, that's the only way I could get it to work.
If you have a difficult dataset that is small and very diverse can you do the training/test split with random samples several times over that same data to build more than one model and then average those neural networks somehow to increase the quality of the output? Has anyone ever tried that and if so does it improve the final network accuracy?
it's called cross-validation and is implemented in many machine learning libraries and software. It often works really well for the situation you have outlined
Thanks Heinz. that was the term I needed, th-cam.com/video/SIyMm5DFwQ8/w-d-xo.html
Looking at the title of the video I thought you were going to talk about TFRecords or something like that. Maybe a more accurate title would be "Prepare your dataset for machine learning with JavaScript"? Anyway, great videos, keep up the good work!
That's a really good point! Sorry about that. I'll have to work on a series for that!
how can prepare my own dataset?
thats preparing dataset? using the iris?
First of all, what editor are you using, second of all where do I get the script from.
I use sypder linked with tensor flow and the problem is I am unable to the the output for Mnist dataset
In this video I'm using geany on Linux. I love it! :)
sorry, i'm viewing the videos but the order of reproduction list are in disorder. great work! i'm enjoing this material!! thank you! grettings from chile!
Nice Videos but where is the source code?
Tensorflow Python is great.
Is there any proper documentation how do I build a Tensorflow c++ API on IOS or Android platform? Example: I have a code written in C++ and I want to create a shared library of it and use it in Android or IOS. #AskTensorFlow
How can i Create Dataset and Model For Resume Parser Can you Share Code
Geaannny! Love it!
Well spotted! :)
Can you add more visuals please?
Trying to show more code in these videos. Will work on visuals too, thanks!
Please expand the topic and use other language please. Thank you.
This is a Tensorflow.js tutorial.
@@МартинТемелакиев it's in the thumbnail and in the description. noone reads those nowadays 🤣
Yay
:)