import processor
from model import Model
# Command-line parameters
tf.flags.DEFINE_boolean("eval_all", False, "Evaluate on all data (default: false)")
tf.flags.DEFINE_string("dir", "", "Directory to read network from")
tf.flags.DEFINE_string("data_file", "data/features.csv",
"File to read data from (default: 'data/features.csv')")
F = tf.flags.FLAGS
F._parse_flags()
# Get data
print("Loading data...")
x, y = processor.load_data(F.data_file)
print("Data loaded.")
print("Preparing data...")
if F.eval_all:
x = np.array(x)
y = np.array(y)
else:
amount = int(0.1 * len(x))
x = np.array(x)[-amount:]
y = np.array(y)[-amount:]
print("Data prepared.")
print("Initializing model...")
checkpoint_file = tf.train.latest_checkpoint(os.path.join(F.dir, "checkpoints"))
print("Reading checkpoint from {}".format(checkpoint_file))
tf.flags.DEFINE_integer("max_data", -1, "Maximum number of data points to use")
tf.flags.DEFINE_integer("batch_size", 10, "Batch Size (default: 10)")
tf.flags.DEFINE_integer("num_epochs", 20, "Number of training epochs (default: 100)")
tf.flags.DEFINE_integer("evaluate_every", 50, "Evaluate model on dev set after this many steps (default: 5)")
tf.flags.DEFINE_integer("checkpoint_every", 100, "Save model after this many steps (default: 20)")
# Misc Parameters
tf.flags.DEFINE_boolean("allow_soft_placement", True, "Allow device soft device placement")
tf.flags.DEFINE_boolean("log_device_placement", False, "Log placement of ops on devices")
F = tf.flags.FLAGS
F._parse_flags()
# Data
print("Loading data...")
input_text, input_extra, y = processor.load_data(F.max_data)
print("Data loaded.")
# Build vocabulary
print("Building vocabulary...")
max_document_length = max([len(t.split(" ")) for t in input_text])
vocab_processor = learn.preprocessing.VocabularyProcessor(max_document_length)
text = np.array(list(vocab_processor.fit_transform(input_text)))
print("Vocabulary built.")
# shuffle data
print("Preparing data...")
shuffle_indices = np.random.permutation(np.arange(len(y)))
text_shuffled = text[shuffle_indices]
extra_shuffled = np.array(extra)[shuffle_indices]
y_shuffled = np.array(y)[shuffle_indices]
tf.flags.DEFINE_string("hidden_layer_sizes", "9", "Comma-separated list of hidden layer sizes (default: '9')")
tf.flags.DEFINE_float("learning_rate", 0.0001, "Learning rate (default: 0.0001)")
tf.flags.DEFINE_float("beta", 0.001, "Beta value for L2 regularization (default: 0.001)")
tf.flags.DEFINE_integer("max_data", -1, "Maximum number of data points to use")
tf.flags.DEFINE_integer("batch_size", 40, "Batch size (default: 40)")
tf.flags.DEFINE_integer("num_epochs", 80, "Number of training epochs (default: 80)")
tf.flags.DEFINE_integer("evaluate_every", 50, "Evaluate model on dev set after this many steps (default: 50)")
tf.flags.DEFINE_string("data_file", "data/features.csv",
"File to read data from (default: 'data/features.csv')")
F = tf.flags.FLAGS
F._parse_flags()
# Get data
print("Loading data...")
x, y = processor.load_data(F.data_file, F.max_data)
print("Data loaded.")
print("Preparing data...")
x = np.array(x)
y = np.array(y)
# split train vs test
amount = int(0.1 * len(x))
x_train, x_eval = x[:-amount], x[-amount:]
y_train, y_eval = y[:-amount], y[-amount:]
print("Data prepared.")
with tf.Session() as session:
print("Initializing model...")
network = Model(