I am not able to view metrics from tensorboard generated logs.
Train a tensorflow model with keras using the tesnorboad callback.
import keras
import numpy
inp = keras.layers.Input( (16, ) )
op = keras.layers.Dense(4)(inp)
mdl = keras.models.Model( inputs=[inp], outputs = [op] )
x = numpy.random.random((128, 16))
y = (numpy.random.random((128, 4))>0.75) * 1.0
mdl.compile()
metrics = [ keras.metrics.BinaryAccuracy(name = "ba")]
mdl.compile( optimizer = keras.optimizers.Adam(0.00001), loss="mean_squared_error", metrics = metrics)
callbacks = [ keras.callbacks.TensorBoard(log_dir='./junk-logs'), keras.callbacks.EarlyStopping(patience=5, monitor="loss", mode='min')]
mdl.fit( x, y, callbacks=callbacks, epochs=100)
Then I try to synch the logs with the neptune.
export NEPTUNE_PROJECT="project/name"
export NEPTUNE_API_TOKEN="xxx"
./neptune-env/bin/neptune tensorboard junk-logs
The result is a new entry in my project, but I cannot find any of the values that were tracked.
I can see graphs for the metrics in tensorboard, but I cannot see the graphs in neptune
./neptune-env/bin/python --version
> Python 3.10.13
./neptune-env/bin/pip list
This is a 64bit linux, I think centos with 4.18 kernel.