How can I do this training?

H, I’m studying tensorflow. But for me to learn I need a practical example.

I wanted to do something really simple, but I don’t know where to start.

for example I have a sequence of numerical data, ranging from 1 to 25.

so I separated by columns basically like this:

id | Ball1 | Ball2 | Ball3 | Ball4 | Ball… | ball 25

1 | 0 | 0 | 1 |1 |0 |1

2 |1 | 1 | 1 |1 |0 |0


Being that in this data in each line it can have only 15 numbers “1” and 10 numbers “0”.

The idea is simple.

input data = dataframe (does not contain last data id)

output data = dataframe (does not contain the first id of the data)

And the training would take the input data and show him that according to that data the output would be the output data

That is, based on this training, I would put the input data, and it would give me the various output possibilities according to the training.

Please how can I do this? really it would unlock my brain to learn tensorflow

How can we do? anyone.

If you’re trying to learn from a practical example, it might help to try something simpler.

Try collecting some temperatures in Celsius and their equivalent measurements in Fahrenheit. You can create a single neuron “network” that takes one input and predicts one output.

During training, you’d feed in one Celsius temperature and try to predict the equivalent measurement. Tensorflow would compare the predicted measurement to the actual measurement (this is called calculating a loss value) and adjust the internal parameters of your network so that your next prediction will be closer to the actual value. By going through this loop over and over, your network’s loss value will get smaller, and its predictions will get closer to the actual Fahrenheit temperature.

After training is done, you can feed in Celsius temperatures that weren’t in your training data and your network will predict what the corresponding Fahrenheit temperature should be. Since this example uses a known formula (F = 1.8*C + 32), you can check for yourself how accurate the predicted value is. If you examine the neuron in your network, you will see that it has learned a weight value for the input that is close to 1.8, and a bias value close to 32, just from the training examples that you fed through it.

Hi, thanks for the tip @Jeff_Corpac,
And @Babak_Zahedi would you like to help too? how can you help me?

I managed to do the training and until I got a good result, not so good yet but it will improve. And as silly as it sounds, I can’t figure out how to put it into production.
for example,
I load the model. And I have the prediction:

input_data = tf.constant([[1,3,4,5,6,7,9,10,11,13,16,18,20,23,25]])

predictions = loaded_model.predict(input_data)
But I can’t get the expected result, I would like to get a list of result probabilities.

I only get this, when I use the test data. I think I’m mounting the input_data wrong
Any example links to mount the input data?

when I mount it like this it works:
predictions = loaded_model.predict(norm_test_X)

previously I had said that the input and output would be 0 and 1.
But I thought it best to use the real data.
which are in total 25 numbers and 15 numbers are drawn, so the numbers are from 1 to 25.
we can only mark 15 numbers, so I need to put the last number of the draw and the algorithm would have to return me a list of probabilities based on training and in each result it would return 15 numbers. that would be predicted.