Numbers not showing in the tensorboard image
in
Convolutional Neural Networks with TensorFlow in Python
/
Confusion matrix and visualizing it with Tensorboard
Dear 365 Data Science Team,
When I use tensorboard for some reason I dont see the CM numbers on the image "Check below picture" before and after I tried to decrease my accuracy.
However, it is still showing the error even if I entered the commands in CMD but its printing.
Also, I have more than 3 check boxes in addition to train, val, and CM. What is the issue here
import io
import itertools
import sklearn
from sklearn import metrics
import tensorflow as tf
import tensorflow_datasets as tfds
import matplotlib.pyplot as plt
import numpy as np
import datetime
Downloading the data and preprocessing the data
In [448]:
BUFFER_SIZE = 100
BATCH_SIZE = 128
NUM_EPOCHS = 20
In [449]:
mnist_dataset, mnist_info = tfds.load(name='mnist', with_info = True, as_supervised=True)
In [450]:
mnist_train, mnist_test = mnist_dataset['train'], mnist_dataset['test']
In [451]:
def scale(image, label):
image=tf.cast(image, tf.float32)
image /= 255.
return image,label
In [452]:
train_and_validation_data = mnist_train.map(scale)
test_data = mnist_test.map(scale)
In [453]:
num_validation_samples = 0.1 * mnist_info.splits['train'].num_examples
num_validation_samples = tf.cast(num_validation_samples, tf.int64)
In [454]:
num_test_samples = mnist_info.splits['test'].num_examples
num_test_samples = tf.cast(num_test_samples, tf.int64)
In [455]:
train_and_validation_data = train_and_validation_data.shuffle(BUFFER_SIZE)
In [456]:
train_data = train_and_validation_data.skip(num_validation_samples)
validation_data = train_and_validation_data.take(num_validation_samples)
In [457]:
train_data = train_data.batch(BATCH_SIZE)
validation_data = validation_data.batch(num_validation_samples)
test_data = test_data.batch(num_test_samples)
In [458]:
for images, labels in validation_data:
images_val = images.numpy()
labels_val = labels.numpy()
Creating the model and training it
In [459]:
model = tf.keras.Sequential([
tf.keras.layers.Conv2D(50, 5, activation='relu', input_shape=(28,28,1)),
tf.keras.layers.MaxPooling2D(pool_size=(2,2)),
tf.keras.layers.Conv2D(50, 3, activation='relu'),
tf.keras.layers.MaxPooling2D(pool_size=(2,2)),
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(10)
##Why no Softmax function? because in some cases we might have mess up our calculations
#So its better to incorporate the softmax into the loss function itself
])
In [460]:
model.summary(line_length=75)
Model: "sequential_18"
___________________________________________________________________________
Layer (type) Output Shape Param #
===========================================================================
conv2d_35 (Conv2D) (None, 24, 24, 50) 1300
max_pooling2d_33 (MaxPooling2D) (None, 12, 12, 50) 0
conv2d_36 (Conv2D) (None, 10, 10, 50) 22550
max_pooling2d_34 (MaxPooling2D) (None, 5, 5, 50) 0
flatten_17 (Flatten) (None, 1250) 0
dense_18 (Dense) (None, 10) 12510
===========================================================================
Total params: 36,360
Trainable params: 36,360
Non-trainable params: 0
___________________________________________________________________________
In [461]:
loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)
In [462]:
model.compile(optimizer='Adadelta', loss=loss_fn, metrics=['accuracy'])
In [463]:
log_dir = 'logs\\fit\\'+ "run-1"
In [464]:
def plot_confusion_matrix(cm, class_names):
figure = plt.figure(figsize=(12,12))
plt.imshow(cm, interpolation='nearest', cmap=plt.cm.Blues)
plt.title("Confusion matrix")
plt.colorbar()
tick_marks = np.arange(len(class_names))
plt.xticks(tick_marks, class_names, rotation=45)
plt.yticks(tick_marks, class_names)
#Normalize the confusion matrix
cm=np.around(cm.astype('float') / cm.sum(axis=1)[:, np.newaxis], decimals=2)
#Use white text if squares are dark; otherwise black
threshold = cm.max() / 2.
for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):
color = 'white' if cm[i,j] > threshold else "black"
plt.text(j, i, cm[i,j], horizontalalignment='center', color=color)
plt.tight_layout()
plt.ylabel('True label')
plt.xlabel('Predicted label')
return figure
In [465]:
def plot_to_image(figure):
"""Converts the matplotlib plot specified by 'figre' to a PNG image and returns it
The supplied figure is closed and inaccessible after this call"""
#Save the plot to a PNG in memory
buf = io.BytesIO()
plt.savefig(buf, format='png')
#Closing figure prevents it from being displayed directly inside the notebook
plt.close(figure)
buf.seek(0)
#Convert PNG buffer to TF image
image = tf.image.decode_png(buf.getvalue(), channels=4)
#Add the batch dimension
image= tf.expand_dims(image,0)
return image
In [466]:
#Define a file writer variable for logging purposes
file_writer_cm = tf.summary.create_file_writer(log_dir + '/cm')
def log_confusion_matrix(epoch, logs):
#Use the model to predcit the values from validation dataset
test_pred_raw = model.predict(images_val)
test_pred = np.argmax(test_pred_raw, axis=1)
#Calculate the confusion matrix
cm = sklearn.metrics.confusion_matrix(labels_val, test_pred)
#Log the confusion matrix as an image summary
figure = plot_confusion_matrix(cm, class_names=['0','1','2','3','4','5','6','7','8','9'])
cm_image = plot_to_image(figure)
#Log the confusion matrix as an image summary
with file_writer_cm.as_default():
tf.summary.image("Confusion Matrix", cm_image, step=epoch)
In [467]:
cm_callback = tf.keras.callbacks.LambdaCallback(on_epoch_end=log_confusion_matrix)
tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir=log_dir, histogram_freq=1, profile_batch=0)
In [468]:
early_stopping= tf.keras.callbacks.EarlyStopping(
monitor='val_loss',
mode = 'auto',
min_delta = 0,
patience = 2,
verbose = 0,
restore_best_weights = True
)
In [469]:
model.fit(
train_data,
epochs= NUM_EPOCHS,
callbacks = [tensorboard_callback, cm_callback, early_stopping],
validation_data = validation_data,
verbose = 2
)
Epoch 1/20
188/188 [==============================] - 1s 4ms/step
422/422 - 21s - loss: 2.3151 - accuracy: 0.0837 - val_loss: 2.3088 - val_accuracy: 0.0893 - 21s/epoch - 50ms/step
Epoch 2/20
188/188 [==============================] - 1s 4ms/step
422/422 - 20s - loss: 2.2996 - accuracy: 0.0991 - val_loss: 2.2932 - val_accuracy: 0.1018 - 20s/epoch - 47ms/step
Epoch 3/20
188/188 [==============================] - 1s 4ms/step
422/422 - 20s - loss: 2.2840 - accuracy: 0.1186 - val_loss: 2.2776 - val_accuracy: 0.1205 - 20s/epoch - 47ms/step
Epoch 4/20
188/188 [==============================] - 1s 4ms/step
422/422 - 20s - loss: 2.2681 - accuracy: 0.1490 - val_loss: 2.2613 - val_accuracy: 0.1617 - 20s/epoch - 48ms/step
Epoch 5/20
188/188 [==============================] - 1s 4ms/step
422/422 - 20s - loss: 2.2513 - accuracy: 0.1976 - val_loss: 2.2441 - val_accuracy: 0.2220 - 20s/epoch - 47ms/step
Epoch 6/20
188/188 [==============================] - 1s 4ms/step
422/422 - 20s - loss: 2.2335 - accuracy: 0.2607 - val_loss: 2.2254 - val_accuracy: 0.2837 - 20s/epoch - 48ms/step
Epoch 7/20
188/188 [==============================] - 1s 4ms/step
422/422 - 20s - loss: 2.2140 - accuracy: 0.3336 - val_loss: 2.2049 - val_accuracy: 0.3633 - 20s/epoch - 47ms/step
Epoch 8/20
188/188 [==============================] - 1s 4ms/step
422/422 - 20s - loss: 2.1923 - accuracy: 0.4089 - val_loss: 2.1820 - val_accuracy: 0.4372 - 20s/epoch - 48ms/step
Epoch 9/20
188/188 [==============================] - 1s 4ms/step
422/422 - 20s - loss: 2.1680 - accuracy: 0.4753 - val_loss: 2.1563 - val_accuracy: 0.4988 - 20s/epoch - 47ms/step
Epoch 10/20
188/188 [==============================] - 1s 4ms/step
422/422 - 20s - loss: 2.1405 - accuracy: 0.5290 - val_loss: 2.1268 - val_accuracy: 0.5503 - 20s/epoch - 47ms/step
Epoch 11/20
188/188 [==============================] - 1s 4ms/step
422/422 - 20s - loss: 2.1091 - accuracy: 0.5700 - val_loss: 2.0931 - val_accuracy: 0.5852 - 20s/epoch - 48ms/step
Epoch 12/20
188/188 [==============================] - 1s 4ms/step
422/422 - 20s - loss: 2.0733 - accuracy: 0.6020 - val_loss: 2.0551 - val_accuracy: 0.6112 - 20s/epoch - 48ms/step
Epoch 13/20
188/188 [==============================] - 1s 4ms/step
422/422 - 20s - loss: 2.0330 - accuracy: 0.6267 - val_loss: 2.0124 - val_accuracy: 0.6345 - 20s/epoch - 48ms/step
Epoch 14/20
188/188 [==============================] - 1s 4ms/step
422/422 - 20s - loss: 1.9874 - accuracy: 0.6484 - val_loss: 1.9640 - val_accuracy: 0.6507 - 20s/epoch - 48ms/step
Epoch 15/20
188/188 [==============================] - 1s 4ms/step
422/422 - 20s - loss: 1.9365 - accuracy: 0.6654 - val_loss: 1.9099 - val_accuracy: 0.6657 - 20s/epoch - 47ms/step
Epoch 16/20
188/188 [==============================] - 1s 4ms/step
422/422 - 20s - loss: 1.8799 - accuracy: 0.6782 - val_loss: 1.8506 - val_accuracy: 0.6817 - 20s/epoch - 48ms/step
Epoch 17/20
188/188 [==============================] - 1s 4ms/step
422/422 - 20s - loss: 1.8178 - accuracy: 0.6891 - val_loss: 1.7855 - val_accuracy: 0.6940 - 20s/epoch - 48ms/step
Epoch 18/20
188/188 [==============================] - 1s 4ms/step
422/422 - 20s - loss: 1.7502 - accuracy: 0.6999 - val_loss: 1.7152 - val_accuracy: 0.7020 - 20s/epoch - 48ms/step
Epoch 19/20
188/188 [==============================] - 1s 4ms/step
422/422 - 21s - loss: 1.6783 - accuracy: 0.7096 - val_loss: 1.6412 - val_accuracy: 0.7085 - 21s/epoch - 49ms/step
Epoch 20/20
188/188 [==============================] - 1s 4ms/step
422/422 - 20s - loss: 1.6039 - accuracy: 0.7182 - val_loss: 1.5674 - val_accuracy: 0.7203 - 20s/epoch - 48ms/step
Out[469]:
<keras.callbacks.History at 0x23b39bb21f0>
Visualizing in Tensorboard
In [470]:
%load_ext tensorboard
%tensorboard --logdir "logs\fit"
0 answers ( 0 marked as helpful)
