How to associate array of class probabilities with corresponding image in keras?

Question

I have a very basic convolutional neural net built in Keras with a TensorFlow backend. The model is based of this Kaggle kernel full model. The input training data are 256x256 images with a corresponding set of class labels in a csv file. For example, here are the class labels in the csv file:

image_name  |  tags
----------------------------------------
train_0     |  class1 class2 class3
train_1     |  class2
train_2     |  class3
and so on ...

The following code shows how I am building, training, and predicting results.

# Build model
model = Sequential()
model.add(Conv2D(32, kernel_size=(3, 3),
                 activation='relu',
                 input_shape=(32, 32, 3))) # Originally (32, 32, 3)

model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(3, activation='sigmoid')) # Originally 17

model.compile(loss='binary_crossentropy', # We NEED binary here, since categorical_crossentropy l1 norms the output before calculating loss.
              optimizer='adam',
              metrics=['accuracy'])

# Train the model              
model.fit(x_train, y_train,
          batch_size=128,
          epochs=4,
          verbose=1,
          validation_data=(x_valid, y_valid))

# Predict
p_valid = model.predict(x_valid, batch_size=128)

You can see the predicted results are an array of class probabilities:

p_valid
Out[29]: 
array([[2.4571007e-01, 1.4929530e-02, 9.3575776e-01],
       [2.6206359e-01, 1.2429485e-02, 9.5172155e-01],
       [3.3679003e-01, 2.8344743e-02, 8.5209453e-01],
       ...,
       [8.2605546e-03, 2.0513092e-07, 9.9999821e-01],
       [8.2605546e-03, 2.0513092e-07, 9.9999821e-01],
       [8.2605667e-03, 2.0513131e-07, 9.9999821e-01]], dtype=float32)

How do I know which probability array is corresponding to which image?

Answer 1

Keras Model.predict() method doesn't shuffle the data, so each row in p_valid corresponds to the row in x_valid. For instance, p_valid[2] holds the probabilities for the image x_valid[2].

By the way, note that the values in each row don't necessarily sum up to 1, because they correspond to the probability of class i being present, thus are not exclusive.