CODEWITHSUNDEEP
tensorflow
Chris
Chris

Reputation: 3715

How to define the data to fit the classifier

I am new to tensorflow. I created a 204x4 matrix where the first 3 colums are feature and the last colum is the target. How do I need to convert the array so that tensorflow can train the data?

TRAINING_SET = np.asarray(seq[:llength])
VALIDATION_SET= np.asarray(seq[llength:llength+tlength])
TEST_SET = np.asarray(seq[llength+tlength:])
num_epochs=100
batch_size = 32
featureColumns = np.shape(TRAINING_SET)[1]

# define a function to get data as batch, you can use this function for test and validation also by simply changing shuffle=False and replacing tf.train.shuffle_batch as tf.train.batch
def data_input_fn(trainset, batch_size, num_epochs, toShuffle):
    data_f = trainset[:, :(featureColumns-1)]
    data_l = trainset[:, (featureColumns-1)]
    data_f_single, data_l_single = tf.train.slice_input_producer([data_f, data_l], num_epochs=num_epochs, shuffle=toShuffle)

    if toShuffle is True:
        data_f_batch, data_l_batch = tf.train.shuffle_batch([data_f_single, data_l_single], batch_size=batch_size, capacity=400, min_after_dequeue=2*batch_size)
    else:
        data_f_batch, data_l_batch = tf.train.batch([data_f_single, data_l_single], batch_size=batch_size, capacity=400, min_after_dequeue=2*batch_size)

    return data_f_batch, data_l_batch

def main():

  # Specify that all features have real-value data
  feature_columns = [tf.contrib.layers.real_valued_column("", dimension=3)]

  # Build 3 layer DNN with 10, 20, 10 units respectively.
  classifier = tf.contrib.learn.DNNClassifier(feature_columns=feature_columns,
                                              hidden_units=[10, 20, 10],
                                              n_classes=10,
                                              model_dir="/tmp/iris_model")

  # Fit model.
  classifier.fit(input_fn=lambda: data_input_fn(TRAINING_SET, batch_size, num_epochs, True), steps=4000)

  # Evaluate accuracy.
  accuracy_test_score = classifier.evaluate(input_fn=lambda: data_input_fn(VALIDATION_SET, batch_size, num_epochs, False),
                                       steps=1)["accuracy"]

  accuracy_validation_score = classifier.evaluate(input_fn=lambda: data_input_fn(TEST_SET, batch_size, num_epochs, False),
                                       steps=1)["accuracy"]

  print ("\nValidation Accuracy: {0:0.2f}\nTest Accuracy: {1:0.2f}\n".format(accuracy_validation_score,accuracy_test_score))

  # Classify two new flower samples.
  def new_samples():
    return np.array(
      [[327,8,3],
       [47,8,0]], dtype=np.float32)

  predictions = list(classifier.predict_classes(input_fn=new_samples))

gives

TypeError: 'Tensor' object is not callable

Upvotes: 1

Views: 528

Answers (1)

Ishant Mrinal
Ishant Mrinal

Reputation: 4918

You need use a function for the input_fn not just a tensor

TRAINING_SET = np.asarray(seq[:llength])
VALIDATION_SET= np.asarray(seq[llength:llength+tlength])
TEST_SET = np.asarray(seq[llength+tlength:])   
num_epochs=100
batch_size = 32
# define a function to get data as batch, you can use this function for test and validation also by simply changing shuffle=False and replacing tf.train.shuffle_batch as tf.train.batch
def data_input_fn(trainset, batch_size, num_epochs):
     data_f = trainset[:, :3]
     data_l = trainset[:, 3]
     data_f_single, data_l_single = tf.train.slice_input_producer([data_f, data_l], num_epochs=num_epochs, shuffle=True)
     data_f_batch, data_l_batch = tf.train.shuffle_batch([data_f_single, data_l_single], batch_size=batch_size, capacity=400, min_after_dequeue=2*batch_size)
     return data_f_batch, data_l_batch

# use this function as input_fn to fit
classifier.fit(input_fn=lambda: data_input_fn(TRAINING_SET, batch_size, num_epochs), steps=4000)

Upvotes: 1

Related Questions