¿how to set a random seed?

[IgnacioRodriguez98]

Description

Hi there.

I´m came from python and there is necessary to set a random seed to ensure that results are reproducible, how can I set that seed in tensorflow.net?

here are examples to set it in python:

np.random.seed(7) tf.set_random_seed(7)

[Wanglongzhi2001]

In TensorFlow.NET, you can use tf.set_random_seed. Here is an example:

using static Tensorflow.Binding;

tf.set_random_seed(10);
var randNum = tf.random.uniform(1);
Console.WriteLine($"randnum: {randNum}"); // randnum: tf.Tensor: shape=(1,), dtype=float32, numpy=array([0.6364622])

[IgnacioRodriguez98]

In TensorFlow.NET, you can use tf.set_random_seed. Here is an example:

using static Tensorflow.Binding;

tf.set_random_seed(10);
var randNum = tf.random.uniform(1);
Console.WriteLine($"randnum: {randNum}"); // randnum: tf.Tensor: shape=(1,), dtype=float32, numpy=array([0.6364622])

Excuse my ignorance, but if I put my code like this (in python if it works for me like this) I would already have my seed and could I have my reproducible experiment?

var LSTM = new Sequential();
      LSTM.Add(new Bidirectional(new LSTM(50, activation: "relu", input_shape: input_shape)));
      LSTM.Add(new Dense(1));
      LSTM.Compile(optimizer: "adam", loss: "mean_squared_error");
tf.set_random_seed(7);
LSTM.fit()

[Wanglongzhi2001]

Sorry, even though TensorFlow and TensorFlow.NET use the same random seed and the same way random numbers are generated, they can't currently produce the same random number. BTW, tf.set_random_seed is used in tf1, you are better use tf.random.set_seed in tf2, and there is no alignment tf.random.set_seed yet in TensorFlow.NET, you can use tf.set_random_seed in TensorFlow.NET instead.