def define_columns(self):
""" See base class """
columns, excluded_cols = [], self.amenities_list + ['picture_url']
col_info = get_column_info(self.data,
excluded=excluded_cols)
for (name, ftype) in col_info:
mapped_type = map_feature_type(self.data[name].dtype)
if is_categorical(ftype):
# Categorical data
if name in self.hardcoded_buckets:
# If hardcoded, take value
keys = self.hardcoded_buckets[name]
else:
# Otherwise is safe to take it from the dataset itself
keys = self.categories[name]
logger.info('Creating column for feature "{}" with keys "{}"'
.format(name, keys))
columns.append(SparseColumn(name,
keys=keys,
type=mapped_type))
elif is_numeric(ftype) or is_bool(ftype):
logger.info('Creating column for numerical feature "{}"'
.format(name))
columns.append(NumericColumn(name, type=mapped_type))
else:
raise RuntimeError('Unknown column type "{}" for column "{}"'
.format(name, ftype))
# Add amenities column
am_keys = self.hardcoded_buckets['amenities'] \
if 'amenities' in self.hardcoded_buckets \
else self.amenities_list
logger.info('Creating column for "amenities" with keys {}'
.format(am_keys))
am_type = map_feature_type(np.dtype('object'))
columns.append(SparseColumn('amenities',
keys=am_keys,
type=am_type))
# Image-specific columns
columns += [
ImageColumn('image', format='JPEG'),
NumericColumn('height', type=map_feature_type(np.dtype('int'))),
NumericColumn('width', type=map_feature_type(np.dtype('int'))),
SparseColumn('path', map_feature_type(np.dtype('object'))),
SparseColumn('format', map_feature_type(np.dtype('object'))),
SparseColumn('colorspace', map_feature_type(np.dtype('object')))
]
for c in columns:
logger.info('Creating column for feature "{}"'.format(c.name))
return columns
def define_columns(self):
# Image columns
base_columns = [
NumericColumn('label', type=map_feature_type(np.dtype('int'))),
ImageColumn('image', format='JPEG'),
SparseColumn('colorspace', map_feature_type(np.dtype('object')))
]
# Categorical and numerical columns for each pixel position
pixel_columns, deep_columns = [], []
for i in range(self.height):
for j in range(self.width):
# Wide column
sparse_type = map_feature_type(np.dtype('object'))
sparse_col = SparseColumn(name=self._get_pixel_name(i, j),
type=sparse_type,
keys=256)
pixel_columns.append(sparse_col)
# Deep column
numeric_name = self._get_pixel_name(i, j) + '_num'
numeric_type = map_feature_type(np.dtype('float'))
numeric_col = NumericColumn(name=numeric_name,
type=numeric_type)
deep_columns.append(numeric_col)
return base_columns + pixel_columns + deep_columns
def define_columns(self):
cols = []
# Columns
for i in range(self.features.shape[1]):
current_col = NumericColumn(name=str(i),
type=map_feature_type(
np.dtype('float')))
cols.append(current_col)
# Label
cols.append(
NumericColumn(name='class',
type=map_feature_type(np.dtype('int'))))
return cols
def define_columns(self):
return [
NumericColumn('label',
type=map_feature_type(np.dtype('int'))),
ImageColumn('image', format='JPEG'),
SparseColumn('colorspace',
map_feature_type(np.dtype('object')))
]
def define_columns(self):
""" See base class """
columns = []
for i in range(self.features.shape[1]):
numeric_type = map_feature_type(self.features[:, i].dtype)
columns.append(NumericColumn(self.get_column_names()[i],
type=numeric_type))
# Add column for index
columns.append(NumericColumn('index',
type=map_feature_type(np.int)))
# Add column for label
columns.append(NumericColumn(self.get_target_name(),
type=map_feature_type(np.float)))
return columns
def define_columns(self):
return [
NumericColumn('label', type=map_feature_type(np.dtype('int'))),
ImageColumn('image', format='JPEG'),
]