I've implemented validation_epoch_end to produce and log metrics, and when I run trainer.validate, the metrics appear in my notebook.
However, when I run trainer.fit, only the training metrics appear; not the validation ones.
The validation step is still being run (because the validation code calls a print statement, which does appear), but the validation metrics don't appear, even though they're logged. Or, if they do appear, the next epoch immediately erases them, so that I can't see them.
(Likewise, tensorboard sees the validation metrics)
How can I see the validation epoch end metrics in a notebook, as each epoch occurs?
You could do the following. Let's say you have the following LightningModule:
class MNISTModel(LightningModule):
def __init__(self):
super().__init__()
self.l1 = torch.nn.Linear(28 * 28, 10)
def forward(self, x):
return torch.relu(self.l1(x.view(x.size(0), -1)))
def training_step(self, batch, batch_nb):
x, y = batch
loss = F.cross_entropy(self(x), y)
# prog_bar=True will display the value on the progress bar statically for the last complete train epoch
self.log("train_loss", loss, on_step=False, on_epoch=True, prog_bar=True)
return loss
def validation_step(self, batch, batch_nb):
x, y = batch
loss = F.cross_entropy(self(x), y)
# prog_bar=True will display the value on the progress bar statically for the last complete validation epoch
self.log("val_loss", loss, on_step=False, on_epoch=True, prog_bar=True)
return loss
def configure_optimizers(self):
return torch.optim.Adam(self.parameters(), lr=0.02)
The trick is to use prog_bar=True in combination with on_step and on_epoch depending on when you want the update on the progress bar. So, in this case, when training:
# Train the model ⚡
trainer.fit(mnist_model, MNIST_dm)
you will see:
Epoch 4: 100% -------------------------- 939/939 [00:09<00:00, 94.51it/s, loss=0.636, v_num=4, val_loss=0.743, train_loss=0.726]
Where loss will be updating each batch as it is the step loss. However, val_loss and train_loss will be static values that will only change after each validation or train epoch respectively.
Related
I´m facing a strange behaviour which I can´t figure out why it is happening. I´m getting a really high loss(BinaryCrossentropy) on my validation batch around 20 or even higher while training. But after the training I do a prediction on the tet set and I get a loss which is lower than 1. Why is that? I went through my code over and over and can´t find the problem.
I´m doing a binary image classification for brian tumors on a dataset provided via kaggle(Link.
And you can find my notebook here: Google-Colab Notebook
My data is loaded this way:
batch_size = 20
train_ds = tf.keras.utils.image_dataset_from_directory(
train_data_path,
subset='training',
seed=42,
color_mode='grayscale',
batch_size=batch_size,
validation_split=0.30
)
valid_ds = tf.keras.utils.image_dataset_from_directory(
train_data_path,
subset='validation',
seed=42,
batch_size=batch_size,
color_mode='grayscale',
validation_split=0.30
)
test_ds = tf.keras.utils.image_dataset_from_directory(
test_data_path,
color_mode='grayscale',
batch_size=batch_size,
shuffle=False
)
This is my modle strcuture
input_shape = image_batch[0].shape
# set up the model structure
model = tf.keras.Sequential([
layers.Conv2D(32, (3, 3), activation='relu', input_shape=input_shape),
layers.MaxPooling2D((2,2)),
layers.Conv2D(64, (3, 3), activation='relu'),
layers.MaxPooling2D((2, 2)),
layers.Dropout(0.3),
layers.Conv2D(64, (3, 3), activation='relu'),
layers.MaxPooling2D((2, 2)),
layers.Dropout(0.3),
layers.Flatten(),
tf.keras.layers.Dense(32, activation="relu"),
layers.Dropout(0.3),
layers.Dense(1, activation="sigmoid")
])
model.summary()
This is my callback function which returns the plots during training:
class PlotLearning(tf.keras.callbacks.Callback):
"""
Callback to plot the learning curves of the model during training.
"""
def on_train_begin(self, logs={}):
self.metrics = {}
for metric in logs:
self.metrics[metric] = []
def on_epoch_end(self, epoch, logs={}):
# Storing metrics
print(logs)
for metric in logs:
if metric in self.metrics:
self.metrics[metric].append(logs.get(metric))
else:
self.metrics[metric] = [logs.get(metric)]
# Plotting
metrics = [x for x in logs if 'val' not in x]
f, axs = plt.subplots(1, len(metrics), figsize=(15,5))
clear_output(wait=True)
for i, metric in enumerate(metrics):
axs[i].plot(range(1, epoch + 2),
self.metrics[metric],
label=metric)
if logs['val_' + metric]:
axs[i].plot(range(1, epoch + 2),
self.metrics['val_' + metric],
label='val_' + metric)
axs[i].legend()
axs[i].grid()
plt.tight_layout()
plt.show()
callbacks_list = [PlotLearning()]
and this is the part where I start the training
# compile model
optimizer = tf.keras.optimizers.Adam(learning_rate=0.0001)
model.compile(optimizer=optimizer,
loss=tf.keras.losses.BinaryCrossentropy(from_logits=False),
metrics=['accuracy']
)
# fit model
history = model.fit(prep_train_ds,
epochs=30,
validation_data=valid_ds,
callbacks=callbacks_list)
This is the output of the callback function after the last epoch run through:
As you can see the loss is really high and oscillating around 20, so I guess it is overfitting.
But as mentiod above, here is what I get when I make a prediction on the test set and calculate the binary crossentropy. The loss is again less than 1 and at least in the range of the training loss
I tried so many things like, chaning batch size, bcs. not enough samples of one class might be in one batch. Then I wanted to see if it is overfitting and changed the number of filters, applyed droput etc. But I couldn´t get the loss function down on the validation set. I´m quite new in the field of image classification and maybe I´m oversseing a thing.
I am training a VAE (using federated learning, but that is not so important) and wanted to keep the loss and train functions simple to exchange. The initial approach was to have a tf.function as loss function and a tf.function as train function as follows:
#tf.function
def kl_reconstruction_loss(model, model_input, beta):
x, y = model_input
mean, logvar = model.encode(x, y)
z = model.reparameterize(mean, logvar)
x_logit = model.decode(z, y)
cross_ent = tf.nn.sigmoid_cross_entropy_with_logits(logits=x_logit, labels=x)
reconstruction_loss = tf.reduce_mean(tf.reduce_sum(cross_ent, axis=[1, 2, 3]), axis=0)
kl_loss = tf.reduce_mean(0.5 * tf.reduce_sum(tf.exp(logvar) + tf.square(mean) - 1. - logvar, axis=-1), axis=0)
loss = reconstruction_loss + beta * kl_loss
return loss, kl_loss, reconstruction_loss
#tf.function
def train_fn(model: tf.keras.Model, batch, optimizer, kl_beta):
"""Trains the model on a single batch.
Args:
model: The VAE model.
batch: A batch of inputs [images, labels] for the vae.
optimizer: The optimizer to train the model.
beta: Weighting of KL loss
Returns:
The loss.
"""
def vae_loss():
"""Does the forward pass and computes losses for the generator."""
# N.B. The complete pass must be inside loss() for gradient tracing.
return kl_reconstruction_loss(model, batch, kl_beta)
with tf.GradientTape() as tape:
loss, kl_loss, rc_loss = vae_loss()
grads = tape.gradient(loss, model.trainable_variables)
grads_and_vars = zip(grads, model.trainable_variables)
optimizer.apply_gradients(grads_and_vars)
return loss
For my dataset this results in an epoch duration of approx. 25 seconds. However, since I have to call those functions directly in my code, I would have to enter different ones if I would want to try out different loss/train functions.
So, alternatively, I followed https://github.com/google-research/federated/tree/master/gans and wrapped the loss function in a class and the train function in another function. Now I have:
class VaeKlReconstructionLossFns(AbstractVaeLossFns):
#tf.function
def vae_loss(self, model, model_input, labels, global_round):
# KL Reconstruction loss
mean, logvar = model.encode(model_input, labels)
z = model.reparameterize(mean, logvar)
x_logit = model.decode(z, labels)
cross_ent = tf.nn.sigmoid_cross_entropy_with_logits(logits=x_logit, labels=model_input)
reconstruction_loss = tf.reduce_mean(tf.reduce_sum(cross_ent, axis=[1, 2, 3]), axis=0)
kl_loss = tf.reduce_mean(0.5 * tf.reduce_sum(tf.exp(logvar) + tf.square(mean) - 1. - logvar, axis=-1), axis=0)
loss = reconstruction_loss + self._get_beta(global_round) * kl_loss
if model.losses:
loss += tf.add_n(model.losses)
return loss, kl_loss, reconstruction_loss
def create_train_vae_fn(
vae_loss_fns: vae_losses.AbstractVaeLossFns,
vae_optimizer: tf.keras.optimizers.Optimizer):
"""Create a function that trains VAE, binding loss and optimizer.
Args:
vae_loss_fns: Instance of gan_losses.AbstractVAELossFns interface,
specifying the VAE training loss.
vae_optimizer: Optimizer for training the VAE.
Returns:
Function that executes one step of VAE training.
"""
# We check that the optimizer has not been used previously, which ensures
# that when it is bound the train fn isn't holding onto a different copy of
# the optimizer variables then the copy that is being exchanged b/w server and
# clients.
if vae_optimizer.variables():
raise ValueError(
'Expected vae_optimizer to not have been used previously, but '
'variables were already initialized.')
#tf.function
def train_vae_fn(model: tf.keras.Model,
model_inputs,
labels,
global_round,
new_optimizer_state=None):
"""Trains the model on a single batch.
Args:
model: The VAE model.
model_inputs: A batch of inputs (usually images) for the VAE.
labels: A batch of labels corresponding to the inputs.
global_round: The current glob al FL round for beta calculation
new_optimizer_state: A possible optimizer state to overwrite the current one with.
Returns:
The number of examples trained on.
The loss.
The updated optimizer state.
"""
def vae_loss():
"""Does the forward pass and computes losses for the generator."""
# N.B. The complete pass must be inside loss() for gradient tracing.
return vae_loss_fns.vae_loss(model, model_inputs, labels, global_round)
# Set optimizer vars
optimizer_state = get_optimizer_state(vae_optimizer)
if new_optimizer_state is not None:
# if optimizer is uninitialised, initialise vars
try:
tf.nest.assert_same_structure(optimizer_state, new_optimizer_state)
except ValueError:
initialize_optimizer_vars(vae_optimizer, model)
optimizer_state = get_optimizer_state(vae_optimizer)
tf.nest.assert_same_structure(optimizer_state, new_optimizer_state)
tf.nest.map_structure(lambda a, b: a.assign(b), optimizer_state, new_optimizer_state)
with tf.GradientTape() as tape:
loss, kl_loss, rc_loss = vae_loss()
grads = tape.gradient(loss, model.trainable_variables)
grads_and_vars = zip(grads, model.trainable_variables)
vae_optimizer.apply_gradients(grads_and_vars)
return tf.shape(model_inputs)[0], loss, optimizer_state
return train_vae_fn
This new formulation takes about 86 seconds per epoch.
I am struggling to understand why the second version performs so much worse than the first one. Does anyone have a good explanation for this?
Thanks in advance!
EDIT: My Tensorflow version is 2.5.0
I am trying to replicate the resnet18 paper. Before running this on the full Image Net dataset on disk, I'm doing some evaluation runs with the publicly available imagenette/320px dataset from TFDS (much much smaller subset of imagenet with 10 classes, already in .tfrecord format._
Note: the full notebook to do training and tracing is available here: resnet18_baseline.ipynb Just switch to a GPU runtime and run all the cells. It's already set-up with tensorboard profiling on the second batch. (You can use TPU as well, but some keras.layers.experimental.preprocessing layers do not support TPU ops yet and you have to enable soft device placement. Please use a GPU).
Input Operations
read images from the input dataset. These images usually have got different dimensions and we need some crop function because input tensors can not have different dimensions for batching. Therefore, for training I use random crop and for testing/validation datasets a center crop.
random_crop_layer = keras.layers.experimental.preprocessing.RandomCrop(224, 224)
center_crop_layer = keras.layers.experimental.preprocessing.CenterCrop(224, 224)
#tf.function(experimental_relax_shapes=True) # avoid retracing
def train_crop_fn(x, y):
return random_crop_layer(x), y
#tf.function(experimental_relax_shapes=True)
def eval_crop_fn(x, y):
return center_crop_layer(x), y
Perform some simple preprocessing/augmentations to the input data. These include rescaling to 0-1 and also scaling based on mean and stdev of the rgb colours on imagenet. Also, random
rescaling_layer = keras.layers.experimental.preprocessing.Rescaling(1./255)
train_preproc = keras.Sequential([
rescaling_layer
])
# from https://github.com/tensorflow/models/blob/master/official/vision/image_classification/preprocessing.py
# Calculated from the ImageNet training set
MEAN_RGB = (0.485 , 0.456, 0.406)
STDDEV_RGB = (0.229, 0.224, 0.225)
#tf.function
def z_score_scale(x):
return (x - MEAN_RGB) / STDDEV_RGB
#tf.function
def train_preproc_fn(x, y):
return z_score_scale(train_preproc(x)), y
#tf.function
def eval_preproc_fn(x, y):
return z_score_scale(eval_preproc(x)), y
Input Pipeline
def get_input_pipeline(input_ds, bs, crop_fn, augmentation_fn):
ret_ds = (
input_ds
.batch(1) # pre-crop are different dimensions and can't be batched
.map(crop_fn,
num_parallel_calls=tf.data.experimental.AUTOTUNE)
.unbatch()
.batch(bs)
.map(augmentation_fn, # augmentations can be batched though.
num_parallel_calls=tf.data.experimental.AUTOTUNE)
)
return ret_ds
# dataset loading
def load_imagenette():
train_ds, ds_info = tfds.load('imagenette/320px', split='train', as_supervised=True, with_info=True)
valid_ds = tfds.load('imagenette/320px', split='validation', as_supervised=True)
return train_ds, valid_ds, valid_ds, ds_info.features['label'].num_classes
# pipeline construction
train_ds, valid_ds, test_ds, num_classes = load_imagenette()
# datasets used for training (notice that I use prefetch here)
train_samples = get_input_pipeline(train_ds, BS, train_crop_fn, train_preproc_fn).prefetch(tf.data.experimental.AUTOTUNE)
valid_samples = get_input_pipeline(valid_ds, BS, eval_crop_fn, eval_preproc_fn).prefetch(tf.data.experimental.AUTOTUNE)
test_samples = get_input_pipeline(test_ds, BS, eval_crop_fn, eval_preproc_fn).prefetch(tf.data.experimental.AUTOTUNE)
Training and Profiling
I use tensorboard profiler to check the second batch size and I get a warning that this is highly input bound, with about 40% of processing wasted on inputs.
For a classic resnet18 model, you can drive the batch size up to 768 without getting a OOM error, which is what I use. A single step with bs 256 takes about 2-3 seconds.
I also get a warning that on_train_batch_size_end is slow, at around ~1.5 seconds, compared to the 1s batch time.
The model training code is very simple keras:
model.fit(
train_samples,
validation_data=valid_samples,
epochs=100,
batch_size=BS,
use_multiprocessing=True
callbacks=[tensorboard_callback, model_checkpoint_callback, early_stop_callback, reduce_lr_callback]
)
and the callbacks are specified as:
log_dir = os.path.join(os.getcwd(), 'logs')
tensorboard_callback = TensorBoard(log_dir=log_dir, update_freq="epoch", profile_batch=2)
reduce_lr_callback = tf.keras.callbacks.ReduceLROnPlateau(monitor='val_loss', factor=0.1, patience=5, min_lr=0.001, verbose=1)
model_checkpoint_callback = tf.keras.callbacks.ModelCheckpoint(filepath='model.{epoch:02d}-{val_loss:.4f}.h5',
monitor='val_loss',
verbose=1,
save_best_only=True)
early_stop_callback = keras.callbacks.EarlyStopping(monitor='val_loss', patience=15)
Lastly, here are some sample tensorboard profiling screenshots. I can't figure out how to make this run faster:
I am training a model with keras using the model.fit() method.
I would like to use multiple validation sets that should be validated on separately after each training epoch so that i get one loss value for each validation set. If possible they should be both displayed during training and as well be returned by the keras.callbacks.History() callback.
I am thinking of something like this:
history = model.fit(train_data, train_targets,
epochs=epochs,
batch_size=batch_size,
validation_data=[
(validation_data1, validation_targets1),
(validation_data2, validation_targets2)],
shuffle=True)
I currently have no idea how to implement this. Is it possible to achieve this by writing my own Callback? Or how else would you approach this problem?
I ended up writing my own Callback based on the History callback to solve the problem. I'm not sure if this is the best approach but the following Callback records losses and metrics for the training and validation set like the History callback as well as losses and metrics for additional validation sets passed to the constructor.
class AdditionalValidationSets(Callback):
def __init__(self, validation_sets, verbose=0, batch_size=None):
"""
:param validation_sets:
a list of 3-tuples (validation_data, validation_targets, validation_set_name)
or 4-tuples (validation_data, validation_targets, sample_weights, validation_set_name)
:param verbose:
verbosity mode, 1 or 0
:param batch_size:
batch size to be used when evaluating on the additional datasets
"""
super(AdditionalValidationSets, self).__init__()
self.validation_sets = validation_sets
for validation_set in self.validation_sets:
if len(validation_set) not in [3, 4]:
raise ValueError()
self.epoch = []
self.history = {}
self.verbose = verbose
self.batch_size = batch_size
def on_train_begin(self, logs=None):
self.epoch = []
self.history = {}
def on_epoch_end(self, epoch, logs=None):
logs = logs or {}
self.epoch.append(epoch)
# record the same values as History() as well
for k, v in logs.items():
self.history.setdefault(k, []).append(v)
# evaluate on the additional validation sets
for validation_set in self.validation_sets:
if len(validation_set) == 3:
validation_data, validation_targets, validation_set_name = validation_set
sample_weights = None
elif len(validation_set) == 4:
validation_data, validation_targets, sample_weights, validation_set_name = validation_set
else:
raise ValueError()
results = self.model.evaluate(x=validation_data,
y=validation_targets,
verbose=self.verbose,
sample_weight=sample_weights,
batch_size=self.batch_size)
for metric, result in zip(self.model.metrics_names,results):
valuename = validation_set_name + '_' + metric
self.history.setdefault(valuename, []).append(result)
which i am then using like this:
history = AdditionalValidationSets([(validation_data2, validation_targets2, 'val2')])
model.fit(train_data, train_targets,
epochs=epochs,
batch_size=batch_size,
validation_data=(validation_data1, validation_targets1),
callbacks=[history]
shuffle=True)
I tested this on TensorFlow 2 and it worked. You can evaluate on as many validation sets as you want at the end of each epoch:
class MyCustomCallback(tf.keras.callbacks.Callback):
def on_epoch_end(self, epoch, logs=None):
res_eval_1 = self.model.evaluate(X_test_1, y_test_1, verbose = 0)
res_eval_2 = self.model.evaluate(X_test_2, y_test_2, verbose = 0)
print(res_eval_1)
print(res_eval_2)
And later:
my_val_callback = MyCustomCallback()
# Your model creation code
model.fit(..., callbacks=[my_val_callback])
Considering the current keras docs, you can pass callbacks to evaluate and evaluate_generator. So you can call evaluate multiple times with different datasets.
I have not tested it, so I am happy if you comment your experiences with it below.
I'm trying to train a dataset with 357 features using Isolation Forest sklearn implementation. I can successfully train and get results when the max features variable is set to 1.0 (the default value).
However when max features is set to 2, it gives the following error:
ValueError: Number of features of the model must match the input.
Model n_features is 2 and input n_features is 357
It also gives the same error when the feature count is 1 (int) and not 1.0 (float).
How I understood was that when the feature count is 2 (int), two features should be considered in creating each tree. Is this wrong? How can I change the max features parameter?
The code is as follows:
from sklearn.ensemble.iforest import IsolationForest
def isolation_forest_imp(dataset):
estimators = 10
samples = 100
features = 2
contamination = 0.1
bootstrap = False
random_state = None
verbosity = 0
estimator = IsolationForest(n_estimators=estimators, max_samples=samples, contamination=contamination,
max_features=features,
bootstrap=boostrap, random_state=random_state, verbose=verbosity)
model = estimator.fit(dataset)
In the documentation it states:
max_features : int or float, optional (default=1.0)
The number of features to draw from X to train each base estimator.
- If int, then draw `max_features` features.
- If float, then draw `max_features * X.shape[1]` features.
So, 2 should mean take two features and 1.0 should mean take all of the features, 0.5 take half and so on, from what I understand.
I think this could be a bug, since, taking a look in IsolationForest's fit:
# Isolation Forest inherits from BaseBagging
# and when _fit is called, BaseBagging takes care of the features correctly
super(IsolationForest, self)._fit(X, y, max_samples,
max_depth=max_depth,
sample_weight=sample_weight)
# however, when after _fit the decision_function is called using X - the whole sample - not taking into account the max_features
self.threshold_ = -sp.stats.scoreatpercentile(
-self.decision_function(X), 100. * (1. - self.contamination))
then:
# when the decision function _validate_X_predict is called, with X unmodified,
# it calls the base estimator's (dt) _validate_X_predict with the whole X
X = self.estimators_[0]._validate_X_predict(X, check_input=True)
...
# from tree.py:
def _validate_X_predict(self, X, check_input):
"""Validate X whenever one tries to predict, apply, predict_proba"""
if self.tree_ is None:
raise NotFittedError("Estimator not fitted, "
"call `fit` before exploiting the model.")
if check_input:
X = check_array(X, dtype=DTYPE, accept_sparse="csr")
if issparse(X) and (X.indices.dtype != np.intc or
X.indptr.dtype != np.intc):
raise ValueError("No support for np.int64 index based "
"sparse matrices")
# so, this check fails because X is the original X, not with the max_features applied
n_features = X.shape[1]
if self.n_features_ != n_features:
raise ValueError("Number of features of the model must "
"match the input. Model n_features is %s and "
"input n_features is %s "
% (self.n_features_, n_features))
return X
So, I am not sure on how you can handle this. Maybe figure out the percentage that leads to just the two features you need - even though I am not sure it'll work as expected.
Note: I am using scikit-learn v.0.18
Edit: as #Vivek Kumar commented this is an issue and upgrading to 0.20 should do the trick.