Source code for holoviews.core.data.grid

from __future__ import absolute_import

from collections import OrderedDict, defaultdict, Iterable

try:
    import itertools.izip as zip
except ImportError:
    pass


import numpy as np
array_types = (np.ndarray,)

try:
    import dask.array as da
    array_types += (da.Array,)
except ImportError:
    da = None


from .dictionary import DictInterface
from .interface import Interface, DataError
from ..dimension import Dimension
from ..element import Element
from ..dimension import OrderedDict as cyODict
from ..ndmapping import NdMapping, item_check
from .. import util



[docs]class GridInterface(DictInterface): """ Interface for simple dictionary-based dataset format using a compressed representation that uses the cartesian product between key dimensions. As with DictInterface, the dictionary keys correspond to the column (i.e dimension) names and the values are NumPy arrays representing the values in that column. To use this compressed format, the key dimensions must be orthogonal to one another with each key dimension specifying an axis of the multidimensional space occupied by the value dimension data. For instance, given an temperature recordings sampled regularly across the earth surface, a list of N unique latitudes and M unique longitudes can specify the position of NxM temperature samples. """ types = (dict, OrderedDict, cyODict) datatype = 'grid' gridded = True @classmethod def init(cls, eltype, data, kdims, vdims): if kdims is None: kdims = eltype.kdims if vdims is None: vdims = eltype.vdims if not vdims: raise ValueError('GridInterface interface requires at least ' 'one value dimension.') ndims = len(kdims) dimensions = [d.name if isinstance(d, Dimension) else d for d in kdims + vdims] if isinstance(data, tuple): data = {d: v for d, v in zip(dimensions, data)} elif isinstance(data, list) and data == []: data = OrderedDict([(d, []) for d in dimensions]) elif not any(isinstance(data, tuple(t for t in interface.types if t is not None)) for interface in cls.interfaces.values()): data = {k: v for k, v in zip(dimensions, zip(*data))} elif isinstance(data, np.ndarray): if data.ndim == 1: if eltype._auto_indexable_1d and len(kdims)+len(vdims)>1: data = np.column_stack([np.arange(len(data)), data]) else: data = np.atleast_2d(data).T data = {k: data[:,i] for i,k in enumerate(dimensions)} elif isinstance(data, list) and data == []: data = {d: np.array([]) for d in dimensions[:ndims]} data.update({d: np.empty((0,) * ndims) for d in dimensions[ndims:]}) elif not isinstance(data, dict): raise TypeError('GridInterface must be instantiated as a ' 'dictionary or tuple') for dim in kdims+vdims: name = dim.name if isinstance(dim, Dimension) else dim if name not in data: raise ValueError("Values for dimension %s not found" % dim) if not isinstance(data[name], array_types): data[name] = np.array(data[name]) kdim_names = [d.name if isinstance(d, Dimension) else d for d in kdims] vdim_names = [d.name if isinstance(d, Dimension) else d for d in vdims] expected = tuple([len(data[kd]) for kd in kdim_names]) irregular_shape = data[kdim_names[0]].shape if kdim_names else () valid_shape = irregular_shape if len(irregular_shape) > 1 else expected[::-1] shapes = tuple([data[kd].shape for kd in kdim_names]) for vdim in vdim_names: shape = data[vdim].shape error = DataError if len(shape) > 1 else ValueError if (not expected and shape == (1,)) or (len(set((shape,)+shapes)) == 1 and len(shape) > 1): # If empty or an irregular mesh pass elif len(shape) != len(expected): raise error('The shape of the %s value array does not ' 'match the expected dimensionality indicated ' 'by the key dimensions. Expected %d-D array, ' 'found %d-D array.' % (vdim, len(expected), len(shape))) elif any((s!=e and (s+1)!=e) for s, e in zip(shape, valid_shape)): raise error('Key dimension values and value array %s ' 'shapes do not match. Expected shape %s, ' 'actual shape: %s' % (vdim, valid_shape, shape), cls) return data, {'kdims':kdims, 'vdims':vdims}, {} @classmethod def irregular(cls, dataset, dim): return dataset.data[dim.name if isinstance(dim, Dimension) else dim].ndim > 1 @classmethod def isscalar(cls, dataset, dim): return np.unique(cls.values(dataset, dim, expanded=False)) == 1 @classmethod def validate(cls, dataset, vdims=True): Interface.validate(dataset, vdims) @classmethod def dimension_type(cls, dataset, dim): if dim in dataset.dimensions(): arr = cls.values(dataset, dim, False, False) else: return None return arr.dtype.type @classmethod def shape(cls, dataset, gridded=False): shape = dataset.data[dataset.vdims[0].name].shape if gridded: return shape else: return (np.product(shape), len(dataset.dimensions())) @classmethod def length(cls, dataset): return cls.shape(dataset)[0] @classmethod def _infer_interval_breaks(cls, coord, axis=0): """ >>> GridInterface._infer_interval_breaks(np.arange(5)) array([-0.5, 0.5, 1.5, 2.5, 3.5, 4.5]) >>> GridInterface._infer_interval_breaks([[0, 1], [3, 4]], axis=1) array([[-0.5, 0.5, 1.5], [ 2.5, 3.5, 4.5]]) """ coord = np.asarray(coord) deltas = 0.5 * np.diff(coord, axis=axis) first = np.take(coord, [0], axis=axis) - np.take(deltas, [0], axis=axis) last = np.take(coord, [-1], axis=axis) + np.take(deltas, [-1], axis=axis) trim_last = tuple(slice(None, -1) if n == axis else slice(None) for n in range(coord.ndim)) return np.concatenate([first, coord[trim_last] + deltas, last], axis=axis)
[docs] @classmethod def coords(cls, dataset, dim, ordered=False, expanded=False, edges=False): """ Returns the coordinates along a dimension. Ordered ensures coordinates are in ascending order and expanded creates ND-array matching the dimensionality of the dataset. """ dim = dataset.get_dimension(dim, strict=True) irregular = cls.irregular(dataset, dim) if irregular or expanded: if irregular: data = dataset.data[dim.name] else: data = util.expand_grid_coords(dataset, dim) if edges and data.shape == dataset.data[dataset.vdims[0].name].shape: data = cls._infer_interval_breaks(data, axis=1) data = cls._infer_interval_breaks(data, axis=0) return data data = dataset.data[dim.name] if ordered and np.all(data[1:] < data[:-1]): data = data[::-1] shape = cls.shape(dataset, True) if dim in dataset.kdims: idx = dataset.get_dimension_index(dim) isedges = (dim in dataset.kdims and len(shape) == dataset.ndims and len(data) == (shape[dataset.ndims-idx-1]+1)) else: isedges = False if edges and not isedges: data = cls._infer_interval_breaks(data) elif not edges and isedges: data = data[:-1] + np.diff(data)/2. return data
[docs] @classmethod def canonicalize(cls, dataset, data, data_coords=None, virtual_coords=[]): """ Canonicalize takes an array of values as input and reorients and transposes it to match the canonical format expected by plotting functions. In certain cases the dimensions defined via the kdims of an Element may not match the dimensions of the underlying data. A set of data_coords may be passed in to define the dimensionality of the data, which can then be used to np.squeeze the data to remove any constant dimensions. If the data is also irregular, i.e. contains multi-dimensional coordinates, a set of virtual_coords can be supplied, required by some interfaces (e.g. xarray) to index irregular datasets with a virtual integer index. This ensures these coordinates are not simply dropped. """ if data_coords is None: data_coords = dataset.dimensions('key', label='name')[::-1] # Reorient data invert = False slices = [] for d in data_coords: coords = cls.coords(dataset, d) if np.all(coords[1:] < coords[:-1]): slices.append(slice(None, None, -1)) invert = True else: slices.append(slice(None)) data = data[tuple(slices)] if invert else data # Transpose data dims = [name for name in data_coords if isinstance(cls.coords(dataset, name), array_types)] dropped = [dims.index(d) for d in dims if d not in dataset.kdims+virtual_coords] if dropped: data = np.squeeze(data, axis=tuple(dropped)) if not any(cls.irregular(dataset, d) for d in dataset.kdims): inds = [dims.index(kd.name) for kd in dataset.kdims] inds = [i - sum([1 for d in dropped if i>=d]) for i in inds] if inds: data = data.transpose(inds[::-1]) # Allow lower dimensional views into data if len(dataset.kdims) < 2: data = data.flatten() return data
@classmethod def invert_index(cls, index, length): if np.isscalar(index): return length - index elif isinstance(index, slice): start, stop = index.start, index.stop new_start, new_stop = None, None if start is not None: new_stop = length - start if stop is not None: new_start = length - stop return slice(new_start-1, new_stop-1) elif isinstance(index, Iterable): new_index = [] for ind in index: new_index.append(length-ind) return new_index @classmethod def ndloc(cls, dataset, indices): selected = {} adjusted_inds = [] all_scalar = True for i, (kd, ind) in enumerate(zip(dataset.kdims[::-1], indices)): coords = cls.coords(dataset, kd.name, True) if np.isscalar(ind): ind = [ind] else: all_scalar = False selected[kd.name] = coords[ind] adjusted_inds.append(ind) for kd in dataset.kdims: if kd.name not in selected: coords = cls.coords(dataset, kd.name) selected[kd.name] = coords all_scalar = False for d in dataset.dimensions(): if d in dataset.kdims and not cls.irregular(dataset, d): continue arr = cls.values(dataset, d, flat=False, compute=False) if all_scalar and len(dataset.vdims) == 1: return arr[tuple(ind[0] for ind in adjusted_inds)] selected[d.name] = arr[tuple(adjusted_inds)] return tuple(selected[d.name] for d in dataset.dimensions()) @classmethod def values(cls, dataset, dim, expanded=True, flat=True, compute=True): dim = dataset.get_dimension(dim, strict=True) if dim in dataset.vdims or dataset.data[dim.name].ndim > 1: data = dataset.data[dim.name] data = cls.canonicalize(dataset, data) if compute and da and isinstance(data, da.Array): data = data.compute() return data.T.flatten() if flat else data elif expanded: data = cls.coords(dataset, dim.name, expanded=True) return data.T.flatten() if flat else data else: return cls.coords(dataset, dim.name, ordered=True) @classmethod def groupby(cls, dataset, dim_names, container_type, group_type, **kwargs): # Get dimensions information dimensions = [dataset.get_dimension(d, strict=True) for d in dim_names] if 'kdims' in kwargs: kdims = kwargs['kdims'] else: kdims = [kdim for kdim in dataset.kdims if kdim not in dimensions] kwargs['kdims'] = kdims invalid = [d for d in dimensions if dataset.data[d.name].ndim > 1] if invalid: if len(invalid) == 1: invalid = "'%s'" % invalid[0] raise ValueError("Cannot groupby irregularly sampled dimension(s) %s." % invalid) # Update the kwargs appropriately for Element group types group_kwargs = {} group_type = dict if group_type == 'raw' else group_type if issubclass(group_type, Element): group_kwargs.update(util.get_param_values(dataset)) else: kwargs.pop('kdims') group_kwargs.update(kwargs) drop_dim = any(d not in group_kwargs['kdims'] for d in kdims) # Find all the keys along supplied dimensions keys = [cls.coords(dataset, d.name) for d in dimensions] transpose = [dataset.ndims-dataset.kdims.index(kd)-1 for kd in kdims] transpose += [i for i in range(dataset.ndims) if i not in transpose] # Iterate over the unique entries applying selection masks grouped_data = [] for unique_key in zip(*util.cartesian_product(keys)): select = dict(zip(dim_names, unique_key)) if drop_dim: group_data = dataset.select(**select) group_data = group_data if np.isscalar(group_data) else group_data.columns() else: group_data = cls.select(dataset, **select) if np.isscalar(group_data) or (isinstance(group_data, array_types) and group_data.shape == ()): group_data = {dataset.vdims[0].name: np.atleast_1d(group_data)} for dim, v in zip(dim_names, unique_key): group_data[dim] = np.atleast_1d(v) elif not drop_dim: if isinstance(group_data, array_types): group_data = {dataset.vdims[0].name: group_data} for vdim in dataset.vdims: data = group_data[vdim.name] data = data.transpose(transpose[::-1]) group_data[vdim.name] = np.squeeze(data) group_data = group_type(group_data, **group_kwargs) grouped_data.append((tuple(unique_key), group_data)) if issubclass(container_type, NdMapping): with item_check(False): return container_type(grouped_data, kdims=dimensions) else: return container_type(grouped_data) @classmethod def key_select_mask(cls, dataset, values, ind): if isinstance(ind, tuple): ind = slice(*ind) if isinstance(ind, array_types): mask = ind elif isinstance(ind, slice): mask = True if ind.start is not None: mask &= ind.start <= values if ind.stop is not None: mask &= values < ind.stop # Expand empty mask if mask is True: mask = np.ones(values.shape, dtype=np.bool) elif isinstance(ind, (set, list)): iter_slcs = [] for ik in ind: iter_slcs.append(values == ik) mask = np.logical_or.reduce(iter_slcs) elif callable(ind): mask = ind(values) elif ind is None: mask = None else: index_mask = values == ind if (dataset.ndims == 1 or dataset._binned) and np.sum(index_mask) == 0: data_index = np.argmin(np.abs(values - ind)) mask = np.zeros(len(values), dtype=np.bool) mask[data_index] = True else: mask = index_mask if mask is None: mask = np.ones(values.shape, dtype=bool) return mask @classmethod def select(cls, dataset, selection_mask=None, **selection): dimensions = dataset.kdims val_dims = [vdim for vdim in dataset.vdims if vdim in selection] if val_dims: raise IndexError('Cannot slice value dimensions in compressed format, ' 'convert to expanded format before slicing.') indexed = cls.indexed(dataset, selection) full_selection = [(d, selection.get(d.name, selection.get(d.label))) for d in dimensions] data = {} value_select = [] for i, (dim, ind) in enumerate(full_selection): irregular = cls.irregular(dataset, dim) values = cls.coords(dataset, dim, irregular) mask = cls.key_select_mask(dataset, values, ind) if irregular: if np.isscalar(ind) or isinstance(ind, (set, list)): raise IndexError("Indexing not supported for irregularly " "sampled data. %s value along %s dimension." "must be a slice or 2D boolean mask." % (ind, dim)) mask = mask.max(axis=i) elif dataset._binned: edges = cls.coords(dataset, dim, False, edges=True) inds = np.argwhere(mask) if np.isscalar(ind): emin, emax = edges.min(), edges.max() if ind < emin: raise IndexError("Index %s less than lower bound " "of %s for %s dimension." % (ind, emin, dim)) elif ind >= emax: raise IndexError("Index %s more than or equal to upper bound " "of %s for %s dimension." % (ind, emax, dim)) idx = max([np.digitize([ind], edges)[0]-1, 0]) mask = np.zeros(len(values), dtype=np.bool) mask[idx] = True values = edges[idx:idx+2] elif len(inds): values = edges[inds.min(): inds.max()+2] else: values = edges[0:0] else: values = values[mask] values, mask = np.asarray(values), np.asarray(mask) value_select.append(mask) data[dim.name] = np.array([values]) if np.isscalar(values) else values int_inds = [np.argwhere(v) for v in value_select][::-1] index = np.ix_(*[np.atleast_1d(np.squeeze(ind)) if ind.ndim > 1 else np.atleast_1d(ind) for ind in int_inds]) for kdim in dataset.kdims: if cls.irregular(dataset, dim): if da and isinstance(dataset.data[kdim.name], da.Array): data[kdim.name] = dataset.data[kdim.name].vindex[index] else: data[kdim.name] = np.asarray(data[kdim.name])[index] for vdim in dataset.vdims: if da and isinstance(dataset.data[vdim.name], da.Array): data[vdim.name] = dataset.data[vdim.name].vindex[index] else: data[vdim.name] = np.asarray(dataset.data[vdim.name])[index] if indexed: if len(dataset.vdims) == 1: arr = np.squeeze(data[dataset.vdims[0].name]) if da and isinstance(arr, da.Array): arr = arr.compute() return arr if np.isscalar(arr) else arr[()] else: return np.array([np.squeeze(data[vd.name]) for vd in dataset.vdims]) return data
[docs] @classmethod def sample(cls, dataset, samples=[]): """ Samples the gridded data into dataset of samples. """ ndims = dataset.ndims dimensions = dataset.dimensions(label='name') arrays = [dataset.data[vdim.name] for vdim in dataset.vdims] data = defaultdict(list) for sample in samples: if np.isscalar(sample): sample = [sample] if len(sample) != ndims: sample = [sample[i] if i < len(sample) else None for i in range(ndims)] sampled, int_inds = [], [] for d, ind in zip(dimensions, sample): cdata = dataset.data[d] mask = cls.key_select_mask(dataset, cdata, ind) inds = np.arange(len(cdata)) if mask is None else np.argwhere(mask) int_inds.append(inds) sampled.append(cdata[mask]) for d, arr in zip(dimensions, np.meshgrid(*sampled)): data[d].append(arr) for vdim, array in zip(dataset.vdims, arrays): flat_index = np.ravel_multi_index(tuple(int_inds)[::-1], array.shape) if da and isinstance(array, da.Array): data[vdim.name].append(array.flatten().vindex[tuple(flat_index)]) else: data[vdim.name].append(array.flat[flat_index]) concatenated = {d: np.concatenate(arrays).flatten() for d, arrays in data.items()} return concatenated
@classmethod def aggregate(cls, dataset, kdims, function, **kwargs): kdims = [kd.name if isinstance(kd, Dimension) else kd for kd in kdims] data = {kdim: dataset.data[kdim] for kdim in kdims} axes = tuple(dataset.ndims-dataset.get_dimension_index(kdim)-1 for kdim in dataset.kdims if kdim not in kdims) for vdim in dataset.vdims: data[vdim.name] = np.atleast_1d(function(dataset.data[vdim.name], axis=axes, **kwargs)) return data @classmethod def reindex(cls, dataset, kdims, vdims): dropped_kdims = [kd for kd in dataset.kdims if kd not in kdims] dropped_vdims = ([vdim for vdim in dataset.vdims if vdim not in vdims] if vdims else []) constant = {} for kd in dropped_kdims: vals = cls.values(dataset, kd.name, expanded=False) if len(vals) == 1: constant[kd.name] = vals[0] data = {k: values for k, values in dataset.data.items() if k not in dropped_kdims+dropped_vdims} if len(constant) == len(dropped_kdims): joined_dims = kdims+dropped_kdims axes = tuple(dataset.ndims-dataset.kdims.index(d)-1 for d in joined_dims) dropped_axes = tuple(dataset.ndims-joined_dims.index(d)-1 for d in dropped_kdims) for vdim in vdims: vdata = data[vdim.name] if len(axes) > 1: vdata = vdata.transpose(axes[::-1]) if dropped_axes: vdata = np.squeeze(vdata, axis=dropped_axes) data[vdim.name] = vdata return data elif dropped_kdims: return tuple(dataset.columns(kdims+vdims).values()) return data @classmethod def add_dimension(cls, dataset, dimension, dim_pos, values, vdim): if not vdim: raise Exception("Cannot add key dimension to a dense representation.") dim = dimension.name if isinstance(dimension, Dimension) else dimension return dict(dataset.data, **{dim: values}) @classmethod def sort(cls, dataset, by=[], reverse=False): if not by or by in [dataset.kdims, dataset.dimensions()]: return dataset.data else: raise Exception('Compressed format cannot be sorted, either instantiate ' 'in the desired order or use the expanded format.') @classmethod def iloc(cls, dataset, index): rows, cols = index scalar = False if np.isscalar(cols): scalar = np.isscalar(rows) cols = [dataset.get_dimension(cols, strict=True)] elif isinstance(cols, slice): cols = dataset.dimensions()[cols] else: cols = [dataset.get_dimension(d, strict=True) for d in cols] if np.isscalar(rows): rows = [rows] new_data = [] for d in cols: new_data.append(cls.values(dataset, d, compute=False)[rows]) if scalar: if new_data and isinstance(new_data[0], da.Array): return new_data[0].compute()[0] return new_data[0][0] return tuple(new_data) @classmethod def range(cls, dataset, dimension): if dataset._binned and dimension in dataset.kdims: expanded = cls.irregular(dataset, dimension) column = cls.coords(dataset, dimension, expanded=expanded, edges=True) else: column = cls.values(dataset, dimension, flat=False) if column.dtype.kind == 'M': dmin, dmax = column.min(), column.max() if da and isinstance(column, da.Array): return da.compute(dmin, dmax) return dmin, dmax elif len(column) == 0: return np.NaN, np.NaN else: try: dmin, dmax = (np.nanmin(column), np.nanmax(column)) if da and isinstance(column, da.Array): return da.compute(dmin, dmax) return dmin, dmax except TypeError: column.sort() return column[0], column[-1]
Interface.register(GridInterface)