ribs.archives.ArrayStore¶

class ribs.archives.ArrayStore(field_desc, capacity)[source]¶

Maintains a set of arrays that share a common dimension.

The ArrayStore consists of several fields of data that are manipulated simultaneously via batch operations. Each field is a NumPy array with a dimension of (capacity, ...) and can be of any type.

Since the arrays all share a common first dimension, they also share a common index. For instance, if we retrieve() the data at indices [0, 2, 1], we would get a dict that contains the objective and measures at indices 0, 2, and 1, e.g.:

{
    "objective": [-1, 3, -5],
    "measures": [[0, 0], [2, 1], [3, 5]],
}

The ArrayStore supports several further operations, in particular a flexible add() method that inserts data into the ArrayStore.

Parameters

field_desc (dict) – Description of fields in the array store. The description is a dict mapping from a str to a tuple of (shape, dtype). For instance, {"objective": ((), np.float32), "measures": ((10,), np.float32)} will create an “objective” field with shape (capacity,) and a “measures” field with shape (capacity, 10). Note that field names must be valid Python identifiers.
capacity (int) – Total possible entries in the store.

Variables

_props (dict) –
Properties that are common to every ArrayStore.
- ”capacity”: Maximum number of data entries in the store.
- ”occupied”: Boolean array of size (capacity,) indicating whether each index has data associated with it.
- ”n_occupied”: Number of data entries currently in the store.
- ”occupied_list”: Array of size (capacity,) listing all occupied indices in the store. Only the first n_occupied elements will be valid.
- ”updates”: Int array recording number of calls to functions that modified the store.
_fields (dict) – Holds all the arrays with their data.

Raises

ValueError – One of the fields in field_desc has a reserved name (currently, “index” is the only reserved name).
ValueError – One of the fields in field_desc has a name that is not a valid Python identifier.

Methods

`__iter__`()	Iterates over entries in the store.
`__len__`()	Number of occupied indices in the store, i.e., number of indices that have a corresponding data entry.
`add`(indices, new_data, extra_args, transforms)	Adds new data to the store at the given indices.
`as_raw_dict`()	Returns the raw data in the ArrayStore as a one-level dictionary.
`clear`()	Removes all entries from the store.
`data`([fields, return_type])	Retrieves data for all entries in the store.
`from_raw_dict`(d)	Loads an ArrayStore from a dict of raw info.
`resize`(capacity)	Resizes the store to the given capacity.
`retrieve`(indices[, fields, return_type])	Collects data at the given indices.

Attributes

`capacity`	Maximum number of data entries in the store.
`field_desc`	Description of fields in the store.
`field_list`	List of fields in the store.
`occupied`	Boolean array of size `(capacity,)` indicating whether each index has a data entry.
`occupied_list`	int32 array listing all occupied indices in the store.

__iter__()[source]¶

Iterates over entries in the store.

When iterated over, this iterator yields dicts mapping from the fields to the individual entries. For instance, if we had an “objective” field, one entry might look like {"index": 1, "objective": 6.0} (similar to retrieve(), the index is included in the output).

Example

for entry in store:
    entry["index"]
    entry["objective"]
    ...

__len__()[source]¶: Number of occupied indices in the store, i.e., number of indices that have a corresponding data entry.

add(indices, new_data, extra_args, transforms)[source]¶

Adds new data to the store at the given indices.

The indices, new_data, and add_info are passed through transforms before adding to the store. The general idea is that these transforms will gradually modify the indices, new_data, and add_info. For instance, they can add new fields to new_data (new_data may not initially have all the same fields as the store). Alternatively, they can filter out duplicate indices, eg if multiple entries are being inserted at the same index we can choose one with the best objective. As another example, the transforms can add stats to the add_info or delete fields from the add_info.

The signature of a transform is as follows:

def transform(indices, new_data, add_info, extra_args,
              occupied, cur_data) ->
    (indices, new_data, add_info):

Transform parameters:

indices (array-like): Array of indices at which new_data should be inserted.
new_data (dict): New data for the given indices. Maps from field name to the array of new data for that field.
add_info (dict): Information to return to the user about the addition process. Example info includes whether each entry was ultimately inserted into the store, as well as general statistics. For the first transform, this will be an empty dict.
extra_args (dict): Additional arguments for the transform.
occupied (array-like): Whether the given indices are currently occupied. Same as that given by retrieve().
cur_data (dict): Data at the current indices in the store. Same as that given by retrieve().

Transform outputs:

indices (array-like): Modified indices. We do NOT assume that the final indices will be unique.
new_data (dict): Modified new_data. At the end of the transforms, it should have the same keys as the store. If indices is empty, new_data will be ignored.
add_info (dict): Modified add_info.

Parameters

indices (array-like) – Initial list of indices for addition.
new_data (dict) – Initial data for addition.
extra_args (dict) – Dict containing additional arguments to pass to the transforms. The dict is passed directly (i.e., no unpacking like with kwargs).
transforms (list) – List of transforms on the data to be added.

Returns

Final add_info from the transforms. new_data and indices are not returned; rather, the new_data is added into the store at indices.

Return type

dict

Raises

ValueError – The final version of new_data does not have the same keys as the fields of this store.
ValueError – The final version of new_data has fields that have a different length than indices.

as_raw_dict()[source]¶

Returns the raw data in the ArrayStore as a one-level dictionary.

To collapse the dict, we prefix each key with props. or fields., so the result looks as follows:

{
  "props.capacity": ...,
  "props.occupied": ...,
  ...
  "fields.objective": ...,
  ...
}

Returns: See description above.
Return type: dict

clear()[source]¶: Removes all entries from the store.

data(fields=None, return_type='dict')[source]¶

Retrieves data for all entries in the store.

Equivalent to calling retrieve() with occupied_list.

Parameters

fields (str or array-like of str) – See retrieve().
return_type (str) – See retrieve().

Returns

See data in retrieve(). occupied is not returned since all indices are known to be occupied in this method.

static from_raw_dict(d)[source]¶

Loads an ArrayStore from a dict of raw info.

Parameters: d (dict) – Dict returned by as_raw_dict().
Returns: The new ArrayStore created from d.
Return type: ArrayStore
Raises: ValueError – The loaded props dict has the wrong keys.

resize(capacity)[source]¶

Resizes the store to the given capacity.

Parameters: capacity (int) – New capacity.
Raises: ValueError – The new capacity is less than or equal to the current capacity.

retrieve(indices, fields=None, return_type='dict')[source]¶

Collects data at the given indices.

Parameters

indices (array-like) – List of indices at which to collect data.
fields (str or array-like of str) – List of fields to include. By default, all fields will be included, with an additional “index” as the last field (“index” can also be placed anywhere in this list). This can also be a single str indicating a field name.
return_type (str) – Type of data to return. See the data returned below. Ignored if fields is a str.

Returns

2-element tuple consisting of:

occupied: Array indicating which indices, among those passed in, have an associated data entry. For instance, if indices is [0, 1, 2] and only index 2 has data, then occupied will be [False, False, True].

Note that if a given index is not marked as occupied, it can have any data value associated with it. For instance, if index 1 was not occupied, then the 6.0 returned in the dict example below should be ignored.
data: The data at the given indices. If fields was a single str, this will just be an array holding data for the given field. Otherwise, this data can take the following forms, depending on the return_type argument:
- return_type="dict": Dict mapping from the field name to the field data at the given indices. For instance, if we have an objective field and request data at indices [4, 1, 0], we would get data that looks like {"objective": [1.5, 6.0, 2.3], "index": [4, 1, 0]}. Observe that we also return the indices as an index entry in the dict. The keys in this dict can be modified using the fields arg; duplicate keys will be ignored since the dict stores unique keys.
- return_type="tuple": Tuple of arrays matching the order given in fields. For instance, if fields was ["objective", "measures"], we would receive a tuple of (objective_arr, measures_arr). In this case, the results from retrieve could be unpacked as:
```
occupied, (objective, measures) = store.retrieve(
    ...,
    return_type="tuple",
)
```
  Unlike with the dict return type, duplicate fields will show up as duplicate entries in the tuple, e.g., fields=["objective", "objective"] will result in two objective arrays being returned.
  
  By default, (i.e., when fields=None), the fields in the tuple will be ordered according to the field_desc argument in the constructor, along with index as the last field.
- return_type="pandas": A pandas.DataFrame with the following columns (by default):
  - For fields that are scalars, a single column with the field name. For example, objective would have a single column called objective.
  - For fields that are 1D arrays, multiple columns with the name suffixed by its index. For instance, if we have a measures field of length 10, we create 10 columns with names measures_0, measures_1, …, measures_9. We do not currently support fields with >1D data.
  - 1 column of integers (np.int32) for the index, named index.
  In short, the dataframe might look like this:
  
  objective
  
  measures_0
  
  …
  
  index
  
  …
  
  Like the other return types, the columns can be adjusted with the fields parameter.

All data returned by this method will be a copy, i.e., the data will not update as the store changes.

Return type

tuple

Raises

ValueError – Invalid field name provided.
ValueError – Invalid return_type provided.

property capacity¶

Maximum number of data entries in the store.

Type: int

property field_desc¶

Description of fields in the store.

Example

store.field_desc == {
    "objective": ((), np.float32),
    "measures": ((10,), np.float32)
}

See the constructor field_desc parameter for more info. Unlike in the field_desc in the constructor, which accepts ints for 1D field shapes (e.g., 5), this field_desc shows 1D field shapes as tuples of 1 entry (e.g., (5,)). Since dicts in Python are ordered, note that this dict will have the same order as in the constructor.

Type: dict

property field_list¶

List of fields in the store.

Example

store.field_list == ["objective", "measures"]

Type: list

property occupied¶

Boolean array of size (capacity,) indicating whether each index has a data entry.

Type: numpy.ndarray

property occupied_list¶

int32 array listing all occupied indices in the store.

Type: numpy.ndarray