Nan loss reasons deep learning

Requests to the Community Collaborations
1

What could cause a Convolutional Neural Network to fail to converge?
Specific-
I am using Tensorflow's iris_training model with some of my own data.

ERROR:tensorflow:Model diverged with loss = NaN.

Traceback...

tensorflow.contrib.learn.python.learn.monitors.NanLossDuringTrainingError: NaN loss during training.

Traceback originated with line-

tf.contrib.learn.DNNClassifier(feature_columns=feature_columns,
                                        hidden_units=[300, 300, 300],
                                        #optimizer=tf.train.ProximalAdagradOptimizer(learning_rate=0.001, l1_regularization_strength=0.00001),                                                          
                                        n_classes=11,
                                        model_dir="/tmp/iris_model")

Can anyone give me any advice on how to modify my network layers, data size, etc to improve my results, given that I have already adjusted the optimizer, put a zero for the learning rate, and opted not to use an optimizer?

2

I think you may be in the wrong place— you’re asking about TensorFlow in Python, right? If so, you should probably check out Stack Overflow. If you’re not talking about TensorFlow in Python, TensorFlow.js integration in Snap! is something that @toontalk made. This discussion about it includes loss becoming NaN.

3

okay.