How do I split Tensorflow datasets? - tensorflow

I have a tensorflow dataset based on one .tfrecord file. How do I split the dataset into test and train datasets? E.g. 70% Train and 30% test?
Edit:
My Tensorflow Version: 1.8
I've checked, there is no "split_v" function as mentioned in the possible duplicate. Also I am working with a tfrecord file.

You may use Dataset.take() and Dataset.skip():
train_size = int(0.7 * DATASET_SIZE)
val_size = int(0.15 * DATASET_SIZE)
test_size = int(0.15 * DATASET_SIZE)
full_dataset = tf.data.TFRecordDataset(FLAGS.input_file)
full_dataset = full_dataset.shuffle()
train_dataset = full_dataset.take(train_size)
test_dataset = full_dataset.skip(train_size)
val_dataset = test_dataset.skip(test_size)
test_dataset = test_dataset.take(test_size)
For more generality, I gave an example using a 70/15/15 train/val/test split but if you don't need a test or a val set, just ignore the last 2 lines.
Take:
Creates a Dataset with at most count elements from this dataset.
Skip:
Creates a Dataset that skips count elements from this dataset.
You may also want to look into Dataset.shard():
Creates a Dataset that includes only 1/num_shards of this dataset.

This question is similar to this one and this one, and I am afraid we have not had a satisfactory answer yet.
Using take() and skip() requires knowing the dataset size. What if I don't know that, or don't want to find out?
Using shard() only gives 1 / num_shards of dataset. What if I want the rest?
I try to present a better solution below, tested on TensorFlow 2 only. Assuming you already have a shuffled dataset, you can then use filter() to split it into two:
import tensorflow as tf
all = tf.data.Dataset.from_tensor_slices(list(range(1, 21))) \
.shuffle(10, reshuffle_each_iteration=False)
test_dataset = all.enumerate() \
.filter(lambda x,y: x % 4 == 0) \
.map(lambda x,y: y)
train_dataset = all.enumerate() \
.filter(lambda x,y: x % 4 != 0) \
.map(lambda x,y: y)
for i in test_dataset:
print(i)
print()
for i in train_dataset:
print(i)
The parameter reshuffle_each_iteration=False is important. It makes sure the original dataset is shuffled once and no more. Otherwise, the two resulting sets may have some overlaps.
Use enumerate() to add an index.
Use filter(lambda x,y: x % 4 == 0) to take 1 sample out of 4. Likewise, x % 4 != 0 takes 3 out of 4.
Use map(lambda x,y: y) to strip the index and recover the original sample.
This example achieves a 75/25 split.
x % 5 == 0 and x % 5 != 0 gives a 80/20 split.
If you really want a 70/30 split, x % 10 < 3 and x % 10 >= 3 should do.
UPDATE:
As of TensorFlow 2.0.0, above code may result in some warnings due to AutoGraph's limitations. To eliminate those warnings, declare all lambda functions separately:
def is_test(x, y):
return x % 4 == 0
def is_train(x, y):
return not is_test(x, y)
recover = lambda x,y: y
test_dataset = all.enumerate() \
.filter(is_test) \
.map(recover)
train_dataset = all.enumerate() \
.filter(is_train) \
.map(recover)
This gives no warning on my machine. And making is_train() to be not is_test() is definitely a good practice.

Related

Discrepancy between results reported by TensorFlow model.evaluate and model.predict

I've been back and forth with this for ages, but without being able to find a solution so far anywhere. So, I have a HuggingFace model ('bert-base-cased') that I'm using with TensorFlow and a custom dataset. I've: (1) tokenized my data (2) split the data; (3) converted the data to TF dataset format; (4) instantiated, compiled and fit the model.
During training, it behaves as you'd expect: training and validation accuracy go up. But when I evaluate the model on the test dataset using TF's model.evaluate and model.predict, the results are very different. The accuracy as reported by model.evaluate is higher (and more or less in line with the validation accuracy); the accuracy as reported by model.predict is about 10% lower. (Maybe it's just a coincidence, but it's similar to the reported training accuracy after the single epoch of fine-tuning.)
Can anyone figure out what's causing this? I include snippets of my code below.
# tokenize the dataset
tokenizer = AutoTokenizer.from_pretrained(pretrained_model_name_or_path="bert-base-cased",use_fast=False)
def tokenize_function(examples):
return tokenizer(examples['text'], padding = "max_length", truncation=True)
tokenized_datasets = dataset.map(tokenize_function, batched=True)
# splitting dataset
trainSize = 0.7
valTestSize = 1 - trainSize
train_testvalid = tokenized_datasets.train_test_split(test_size=valTestSize,stratify_by_column='class')
valid_test = train_testvalid['test'].train_test_split(test_size=0.5,stratify_by_column='class')
# renaming each of the datasets for convenience
train_set = train_testvalid['train']
val_set = valid_test['train']
test_set = valid_test['test']
# converting the tokenized datasets to TensorFlow datasets
data_collator = DefaultDataCollator(return_tensors="tf")
tf_train_dataset = train_set.to_tf_dataset(
columns=["attention_mask", "input_ids", "token_type_ids"],
label_cols=['class'],
shuffle=True,
collate_fn=data_collator,
batch_size=8)
tf_validation_dataset = val_set.to_tf_dataset(
columns=["attention_mask", "input_ids", "token_type_ids"],
label_cols=['class'],
shuffle=False,
collate_fn=data_collator,
batch_size=8)
tf_test_dataset = test_set.to_tf_dataset(
columns=["attention_mask", "input_ids", "token_type_ids"],
label_cols=['class'],
shuffle=False,
collate_fn=data_collator,
batch_size=8)
# loading tensorflow model
model = TFAutoModelForSequenceClassification.from_pretrained("bert-base-cased", num_labels=1)
# compiling the model
model.compile(
optimizer=tf.keras.optimizers.Adam(learning_rate=5e-6),
loss=tf.keras.losses.BinaryCrossentropy(),
metrics=tf.metrics.BinaryAccuracy())
# fitting model
history = model.fit(tf_train_dataset,
validation_data=tf_validation_dataset,
epochs=1)
# Evaluating the model on the test data using `evaluate`
results = model.evaluate(x=tf_test_dataset,verbose=2) # reports binary_accuracy: 0.9152
# first attempt at using model.predict method
hits = 0
misses = 0
for x, y in tf_test_dataset:
logits = tf.keras.backend.get_value(model(x, training=False).logits)
labels = tf.keras.backend.get_value(y)
for i in range(len(logits)):
if logits[i][0] < 0:
z = 0
else:
z = 1
if z == labels[i]:
hits += 1
else:
misses += 1
print(hits/(hits+misses)) # reports binary_accuracy: 0.8187
# second attempt at using model.predict method
modelPredictions = model.predict(tf_test_dataset).logits
testDataLabels = np.concatenate([y for x, y in tf_test_dataset], axis=0)
hits = 0
misses = 0
for i in range(len(modelPredictions)):
if modelPredictions[i][0] >= 0:
z = 1
else:
z = 0
if z == testDataLabels[i]:
hits += 1
else:
misses += 1
print(hits/(hits+misses)) # reports binary_accuracy: 0.8187
Things I've tried include:
different loss functions (it's a binary classification problem with the label column of the dataset filled with either a zero or a one for each row);
different ways of unpacking the test dataset and feeding it to model.predict;
altering the 'num_labels' parameter between 1 and 2.
I fixed the problem by changing the num_labels parameter to two and the loss function to sparse categorical cross entropy. (I then had to change my model.predict loop by taking the argmax of the two logits produced by the model.)

Keras data generator predict same number of values

I have implemented a CNN-based regression model that uses a data generator to use the huge amount of data I have. Training and evaluation work well, but there's an issue with the prediction. If for example I want to predict values from a test dataset of 50 samples, I use model.predict with a batch size of 5. The problem is that model.predict returns 5 values repeated 10 times, instead of 50 different values . The same thing happens if I change to batch size to 1, it will return one value 50 times.
To solve this issue, I used a full batch size (50 in my example), and it worked. But I can't I use this method on my whole test data because it's too huge.
Do you have any other solution, or what is the problem in my approach?
My data generator code:
import numpy as np
import keras
class DataGenerator(keras.utils.Sequence):
'Generates data for Keras'
def __init__(self, list_IDs, data_X, data_Z, target_y batch_size=32, dim1=(120,120),
dim2 = 80, n_channels=1, shuffle=True):
'Initialization'
self.dim1 = dim1
self.dim2 = dim2
self.batch_size = batch_size
self.data_X = data_X
self.data_Z = data_Z
self.target_y = target_y
self.list_IDs = list_IDs
self.n_channels = n_channels
self.shuffle = shuffle
self.on_epoch_end()
def __len__(self):
'Denotes the number of batches per epoch'
return int(np.floor(len(self.list_IDs) / self.batch_size))
def __getitem__(self, index):
'Generate one batch of data'
# Generate indexes of the batch
indexes = self.indexes[index*self.batch_size:(index+1)*self.batch_size]
# Find list of IDs
list_IDs_temp = [self.list_IDs[k] for k in range(len(indexes))]
# Generate data
([X, Z], y) = self.__data_generation(list_IDs_temp)
return ([X, Z], y)
def on_epoch_end(self):
'Updates indexes after each epoch'
self.indexes = np.arange(len(self.list_IDs))
if self.shuffle == True:
np.random.shuffle(self.indexes)
def __data_generation(self, list_IDs_temp):
'Generates data containing batch_size samples' # X : (n_samples, *dim, n_channels)
# Initialization
X = np.empty((self.batch_size, *self.dim1, self.n_channels))
Z = np.empty((self.batch_size, self.dim2))
y = np.empty((self.batch_size))
# Generate data
for i, ID in enumerate(list_IDs_temp):
# Store sample
X[i,] = np.load('data/' + data_X + ID + '.npy')
Z[i,] = np.load('data/' + data_Z + ID + '.npy')
# Store target
y[i] = np.load('data/' + target_y + ID + '.npy')
How I call model.predict()
predict_params = {'list_IDs': 'indexes',
'data_X': 'images',
'data_Z': 'factors',
'target_y': 'True_values'
'batch_size': 5,
'dim1': (120,120),
'dim2': 80,
'n_channels': 1,
'shuffle'=False}
# Prediction generator
prediction_generator = DataGenerator(test_index, **predict_params)
predition_results = model.predict(prediction_generator, steps = 1, verbose=1)
If we look at your __getitem__ function, we can see this code:
list_IDs_temp = [self.list_IDs[k] for k in range(len(indexes))]
This code will always return the same numbers IDs, because the length len of the indexes is always the same (at least as long as all batches have an equal amount of samples) and we just loop over the first couple of indexes every time.
You are already extracting the indexes of the current batch beforehand, so the line with the error is not needed at all. The following code should work:
def __getitem__(self, index):
'Generate one batch of data'
# Generate indexes of the batch
list_IDs_temp = self.indexes[index*self.batch_size:(index+1)*self.batch_size]
# Generate data
([X, Z], y) = self.__data_generation(list_IDs_temp)
return ([X, Z], y)
See if this code works and you get different results. You should now get bad predictions, because during training, your model would also only have trained on the same few data points as of now.
When you use a generator you specify a batch size. model.predict will produce batch size number of output predictions. If you set steps=1 that is all the predictions you will get. To set the steps you should take the number of samples you have and divide it by the batch size. For example if you have 50 images with a batch size of 5 then you should set steps equal to 10. Ideally you want to go through your test set exactly once. The code below will determine the batch size and steps to do that. In the code b_max is a value you select that limits the maximum batch size . You should set this based on your memory size to avoid a OOM (out of memory) error. In the code below parameter length is equal to the number of test samples you have.
length=500
b_max=80
batch_size=sorted([int(length/n) for n in range(1,length+1) if length % n ==0 and length/n<=b_max],reverse=True)[0]
steps=int(length/batch_size)
The result will be batch_size= 50 steps=10. Note if length is a prime number the result will be batch_size=1 and steps=length
According to this solution, you need to change your steps to the total number of images you want to test on. Try:
# Assuming test_index is a list
predition_results = model.predict(prediction_generator, steps = len(test_index), verbose=1)

Weird behavior of zipped tensorflow dataset with random tensors

In the example below (Tensorflow 2.0), we have a dummy tensorflow dataset with three elements. We map a function on it (replace_with_float) that returns a randomly generated value in two copies. As we expect, when we take elements from the dataset, the first and second coordinates have the same value.
Now, we create two "slice" datasets from the first coordinates and the second coordinates, respectively and we zip the two datasets back together. The slicing and the zipping operations seems inverses of each other, so I would expect the resulting dataset to be equivalent to the previous one. However, as we see, now the first and second coordinates are different randomly generated values.
Maybe even more interestingly, if we zip the "same" dataset with itself by
df = tf.data.Dataset.zip((df.map(lambda x, y: x), df.map(lambda x, y: x))), the two coordinates will also have different values.
How can this behavior be explained? Perhaps two different graphs are constructed for the two datasets to be zipped and they are run independently?
import tensorflow as tf
def replace_with_float(element):
rand = tf.random.uniform([])
return (rand, rand)
df = tf.data.Dataset.from_tensor_slices([0, 0, 0])
df = df.map(replace_with_float)
print('Before zipping: ')
for x in df:
print(x[0].numpy(), x[1].numpy())
df = tf.data.Dataset.zip((df.map(lambda x, y: x), df.map(lambda x, y: y)))
print('After zipping: ')
for x in df:
print(x[0].numpy(), x[1].numpy())
Sample output:
Before zipping:
0.08801079 0.08801079
0.638958 0.638958
0.800568 0.800568
After zipping:
0.9676769 0.23045003
0.91056764 0.6551999
0.4647777 0.6758332
The short answer is that datasets don't cache intermediate values between full iterations, unless you explicitly request that using df.cache(), and they don't deduplicate common inputs either.
So in the second loop, the entire pipeline runs again.
Similarly, in the second instance, the two df.map calls cause df to run twice.
Adding a tf.print helps explain what happens:
def replace_with_float(element):
rand = tf.random.uniform([])
tf.print('replacing', element, 'with', rand)
return (rand, rand)
I've also pulled the lambdas on separate lines to avoid the autograph warning:
first = lambda x, y: x
second = lambda x, y: y
df = tf.data.Dataset.zip((df.map(first), df.map(second)))
Before zipping:
replacing 0 with 0.624579549
0.62457955 0.62457955
replacing 0 with 0.471772075
0.47177207 0.47177207
replacing 0 with 0.394005418
0.39400542 0.39400542
After zipping:
replacing 0 with 0.537954807
replacing 0 with 0.558757305
0.5379548 0.5587573
replacing 0 with 0.839109302
replacing 0 with 0.878996611
0.8391093 0.8789966
replacing 0 with 0.0165234804
replacing 0 with 0.534951568
0.01652348 0.53495157
To avoid the duplicate input problem, you can use use a single map call:
swap = lambda x, y: (y, x)
df = df.map(swap)
Or you can use df = df.cache() to avoid both effects:
df = df.map(replace_with_float)
df = df.cache()
Before zipping:
replacing 0 with 0.728474379
0.7284744 0.7284744
replacing 0 with 0.419658661
0.41965866 0.41965866
replacing 0 with 0.911524653
0.91152465 0.91152465
After zipping:
0.7284744 0.7284744
0.41965866 0.41965866
0.91152465 0.91152465

predicting using pre-trained model becomes slower and slower

I'm using a very naive way to make predictions based on pre-trained model in keras. But it becomes much slower later. Anyone knows why? I'm very very very new to tensorflow.
count = 0
first = True
for nm in image_names:
img = image.load_img(TEST_PATH + nm, target_size=(299, 299))
img = image.img_to_array(img)
image_batch = np.expand_dims(img, axis=0)
processed_image = inception_v3.preprocess_input(image_batch.copy())
prob = inception_model.predict(processed_image)
df1 = pd.DataFrame({'photo_id': [nm]})
df2 = pd.DataFrame(prob, columns=['feat' + str(j + 1) for j in range(prob.shape[1])])
df = pd.concat([df1, df2], axis=1)
header = first
mode = 'w' if first else 'a'
df.to_csv(outfile, index=False, header=header, mode=mode)
first = False
count += 1
if count % 100 == 0:
print('%d processed' % count)
I doubt the TF is slowing down. However there is another stack overflow question showing that to_csv slows down on append.
Performance: Python pandas DataFrame.to_csv append becomes gradually slower
If the images come batched you may also benefit from making larger batches rather than predicting one image at a time.
You can also explore tf.data for better data pipelining.

What does batch, repeat, and shuffle do with TensorFlow Dataset?

I'm currently learning TensorFlow but I came across a confusion in the below code snippet:
dataset = dataset.shuffle(buffer_size = 10 * batch_size)
dataset = dataset.repeat(num_epochs).batch(batch_size)
return dataset.make_one_shot_iterator().get_next()
I know that first the dataset will hold all the data but what shuffle(),repeat(), and batch() do to the dataset?
Please help me with an example and explanation.
Update: Here is a small collaboration notebook for demonstration of this answer.
Imagine, you have a dataset: [1, 2, 3, 4, 5, 6], then:
How ds.shuffle() works
dataset.shuffle(buffer_size=3) will allocate a buffer of size 3 for picking random entries. This buffer will be connected to the source dataset.
We could image it like this:
Random buffer
|
| Source dataset where all other elements live
| |
↓ ↓
[1,2,3] <= [4,5,6]
Let's assume that entry 2 was taken from the random buffer. Free space is filled by the next element from the source buffer, that is 4:
2 <= [1,3,4] <= [5,6]
We continue reading till nothing is left:
1 <= [3,4,5] <= [6]
5 <= [3,4,6] <= []
3 <= [4,6] <= []
6 <= [4] <= []
4 <= [] <= []
How ds.repeat() works
As soon as all the entries are read from the dataset and you try to read the next element, the dataset will throw an error.
That's where ds.repeat() comes into play. It will re-initialize the dataset, making it again like this:
[1,2,3] <= [4,5,6]
What will ds.batch() produce
The ds.batch() will take the first batch_size entries and make a batch out of them. So, a batch size of 3 for our example dataset will produce two batch records:
[2,1,5]
[3,6,4]
As we have a ds.repeat() before the batch, the generation of the data will continue. But the order of the elements will be different, due to the ds.random(). What should be taken into account is that 6 will never be present in the first batch, due to the size of the random buffer.
The following methods in tf.Dataset :
repeat( count=0 ) The method repeats the dataset count number of times.
shuffle( buffer_size, seed=None, reshuffle_each_iteration=None) The method shuffles the samples in the dataset. The buffer_size is the number of samples which are randomized and returned as tf.Dataset.
batch(batch_size,drop_remainder=False) Creates batches of the dataset with batch size given as batch_size which is also the length of the batches.
An example that shows looping over epochs. Upon running this script notice the difference in
dataset_gen1 - shuffle operation produces more random outputs (this may be more useful while running machine learning experiments)
dataset_gen2 - lack of shuffle operation produces elements in sequence
Other additions in this script
tf.data.experimental.sample_from_datasets - used to combine two datasets. Note that the shuffle operation in this case shall create a buffer that samples equally from both datasets.
import os
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3" # to avoid all those prints
os.environ["TF_GPU_THREAD_MODE"] = "gpu_private" # to avoid large "Kernel Launch Time"
import tensorflow as tf
if len(tf.config.list_physical_devices('GPU')):
tf.config.experimental.set_memory_growth(tf.config.list_physical_devices('GPU')[0], True)
class Augmentations:
def __init__(self):
pass
#tf.function
def filter_even(self, x):
if x % 2 == 0:
return False
else:
return True
class Dataset:
def __init__(self, aug, range_min=0, range_max=100):
self.range_min = range_min
self.range_max = range_max
self.aug = aug
def generator(self):
dataset = tf.data.Dataset.from_generator(self._generator
, output_types=(tf.float32), args=())
dataset = dataset.filter(self.aug.filter_even)
return dataset
def _generator(self):
for item in range(self.range_min, self.range_max):
yield(item)
# Can be used when you have multiple datasets that you wish to combine
class ZipDataset:
def __init__(self, datasets):
self.datasets = datasets
self.datasets_generators = []
def generator(self):
for dataset in self.datasets:
self.datasets_generators.append(dataset.generator())
return tf.data.experimental.sample_from_datasets(self.datasets_generators)
if __name__ == "__main__":
aug = Augmentations()
dataset1 = Dataset(aug, 0, 100)
dataset2 = Dataset(aug, 100, 200)
dataset = ZipDataset([dataset1, dataset2])
epochs = 2
shuffle_buffer = 10
batch_size = 4
prefetch_buffer = 5
dataset_gen1 = dataset.generator().shuffle(shuffle_buffer).batch(batch_size).prefetch(prefetch_buffer)
# dataset_gen2 = dataset.generator().batch(batch_size).prefetch(prefetch_buffer) # this will output odd elements in sequence
for epoch in range(epochs):
print ('\n ------------------ Epoch: {} ------------------'.format(epoch))
for X in dataset_gen1.repeat(1): # adding .repeat() in the loop allows you to easily control the end of the loop
print (X)
# Do some stuff at end of loop