Evaluating pairwise distances between the output of a tf.keras.model - python-3.x

Question

I am trying to create a custom loss function in tensorflow. I am using tensorflow v2.0.rc0 for running the code. Following is the code and the function min_dist_loss computes the pairwise loss between the output of the neural network. Here's the code
def min_dist_loss(_, y_pred):
distances = []
for i in range(0, 16):
for j in range(i + 1, 16):
distances.append(tf.linalg.norm(y_pred[i] - y_pred[j]))
return -tf.reduce_min(distances)
and the module is being initialized and compiled as follows
import tensorflow as tf
from tensorboard.plugins.hparams import api as hp
HP_NUM_UNITS = hp.HParam('num_units', hp.Discrete([6, 7]))
HP_OPTIMIZER = hp.HParam('optimizer', hp.Discrete(['adam', 'sgd']))
METRIC_ACCURACY = 'accuracy'
with tf.summary.create_file_writer('logs\hparam_tuning').as_default():
hp.hparams_config(
hparams=[HP_NUM_UNITS, HP_OPTIMIZER],
metrics=[hp.Metric(METRIC_ACCURACY, display_name='Accuracy')]
)
def train_test_model(logdir, hparams):
weight1 = np.random.normal(loc=0.0, scale=0.01, size=[4, hparams[HP_NUM_UNITS]])
init1 = tf.constant_initializer(weight1)
weight2 = np.random.normal(loc=0.0, scale=0.01, size=[hparams[HP_NUM_UNITS], 7])
init2 = tf.constant_initializer(weight2)
model = tf.keras.models.Sequential([
# tf.keras.layers.Flatten(),
tf.keras.layers.Dense(hparams[HP_NUM_UNITS], activation=tf.nn.sigmoid, kernel_initializer=init1),
tf.keras.layers.Dense(7, activation=tf.nn.sigmoid, kernel_initializer=init2) if hparams[HP_NUM_UNITS] == 6 else
None,
])
model.compile(
optimizer=hparams[HP_OPTIMIZER],
loss=min_dist_loss,
# metrics=['accuracy'],
)
x_train = [list(k) for k in itertools.product([0, 1], repeat=4)]
shuffle(x_train)
x_train = 2 * np.array(x_train) - 1
model.fit(
x_train, epochs=1, batch_size=16,
callbacks=[
tf.keras.callbacks.TensorBoard(logdir),
hp.KerasCallback(logdir, hparams)
],
)
Now since the tensor object y_pred in the min_dist_loss is an object of shape [?, 7], indexing with i is throwing the following error:
Traceback (most recent call last):
File "/home/pc/Documents/user/code/keras_tensorflow/src/try1.py", line 95, in <module>
run('logs\hparam_tuning' + run_name, hparams)
File "/home/pc/Documents/user/code/keras_tensorflow/src/try1.py", line 78, in run
accuracy = train_test_model(run_dir, hparams)
File "/home/pc/Documents/user/code/keras_tensorflow/src/try1.py", line 66, in train_test_model
hp.KerasCallback(logdir, hparams)
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training.py", line 734, in fit
use_multiprocessing=use_multiprocessing)
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training_v2.py", line 324, in fit
total_epochs=epochs)
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training_v2.py", line 123, in run_one_epoch
batch_outs = execution_function(iterator)
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training_v2_utils.py", line 86, in execution_function
distributed_function(input_fn))
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/eager/def_function.py", line 427, in __call__
self._initialize(args, kwds, add_initializers_to=initializer_map)
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/eager/def_function.py", line 370, in _initialize
*args, **kwds))
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/eager/function.py", line 1847, in _get_concrete_function_internal_garbage_collected
graph_function, _, _ = self._maybe_define_function(args, kwargs)
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/eager/function.py", line 2147, in _maybe_define_function
graph_function = self._create_graph_function(args, kwargs)
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/eager/function.py", line 2038, in _create_graph_function
capture_by_value=self._capture_by_value),
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/framework/func_graph.py", line 915, in func_graph_from_py_func
func_outputs = python_func(*func_args, **func_kwargs)
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/eager/def_function.py", line 320, in wrapped_fn
return weak_wrapped_fn().__wrapped__(*args, **kwds)
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training_v2_utils.py", line 73, in distributed_function
per_replica_function, args=(model, x, y, sample_weights))
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/distribute/distribute_lib.py", line 760, in experimental_run_v2
return self._extended.call_for_each_replica(fn, args=args, kwargs=kwargs)
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/distribute/distribute_lib.py", line 1787, in call_for_each_replica
return self._call_for_each_replica(fn, args, kwargs)
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/distribute/distribute_lib.py", line 2132, in _call_for_each_replica
return fn(*args, **kwargs)
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/autograph/impl/api.py", line 292, in wrapper
return func(*args, **kwargs)
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training_v2_utils.py", line 264, in train_on_batch
output_loss_metrics=model._output_loss_metrics)
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training_eager.py", line 311, in train_on_batch
output_loss_metrics=output_loss_metrics))
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training_eager.py", line 252, in _process_single_batch
training=training))
File "/home/pc/Documents/user/code/keras_tensorflow/venv/lib/python3.6/site-packages/tensorflow_core/python/keras/engine/training_eager.py", line 166, in _model_loss
per_sample_losses = loss_fn.call(targets[i], outs[i])
IndexError: list index out of range
How do I compute the minimum distance in this setting? Any help is appreciated. Also, if there are any errors in other parts of the code, please feel free to point it out. I am new to using keras on tensorflow.

Answer 1

Keras is expecting you to provide the true labels as well. Since you're defining your own loss function and you're not using the true labels, you can pass some garbage labels. Eg: np.arange(16).
Change your model.fit as below and it should work
model.fit(
x_train, np.arange(x_train.shape[0]), epochs=1, batch_size=16,
callbacks=[
tf.keras.callbacks.TensorBoard(logdir),
hp.KerasCallback(logdir, hparams)
],
)