Conversions represent the (idealized) computation of patterns in one representation from patterns in another representation. They are used to implement (arbitrary_conversion_connection) or set up (e.g., populate matrices) connections between layers.

some_conv description

The description of some_conv will contain additional elements Source and Target for the representations converted. There will also be a Weight element used to determined priorities (lower is more important) over other conversions. The cache of results is also described by Cache size, and Cache items contains a reference to the list_cache method. If a fallback conversion has been established, this will also be listed under Fallback.

some_conv add_fallback fb_conv

Register fb_conv as the fallback for some_conv. fb_conv is used if some_conv would otherwise return the empty (no rep) pattern.

some_conv set_source src_rep

Attempt to set the source representation of some_conv to src_rep. Some subtypes may reject the attempt.

some_conv set_target tgt_rep

Attempt to set the target representation of some_conv to tgt_rep. Some subtypes may reject the attempt.

some_conv get_source

Returns the source representation of some_conv.

some_conv get_target

Returns the target representation of some_conv.

some_conv list_cache

List the source-target pattern-pairs in the cache for some_conv.

some_conv set_weight weight

Set the weight of some_conv to the floating point weight.

some_conv set_auto_use 1
some_conv set_auto_use 0

Allow (1) or forbid (0) some_conv to be found in searches for conversions between the source and target representation of some_conv. This will affect the target representation's find_conversion, attempts to autoconnect layers, and other searches, such as in the automatic creation of conversion_chains.

some_conv apply in_pattern

Return a pattern that is the result of applying some_conv to in_pattern.