TODOs.md

TODOs

To Do

Bulk

• fix encoding issue when using a Windows system and a language which needs Unicode support (like most languages), as was highlighted with this commit

New syntax

• add a way to specify a rule on several consecutive lines

• add a way to give several different names to units (unit name aliases)

• add probabilities weighting of generation for rules in defintions (cf. rodrigopivi/Chatito#48 and new implementation)

• add custom annotations as Chatito does

• add an annotation to consider each rule of the unit as a raw string

• add a way to make some generation mandatory in the training set, test set or both (cf. rodrigopivi/Chatito#51)

• add an annotation to allow a unit to shadow another one

• add percentages of all possible examples for training and test (rather than simple numbers)

• support several arguments in one rule

• support rules as arguments

• add support for Chatito's augmentations (cf. rodrigopivi/Chatito#48)

• add an annotation for generating typos (linked to Chatito's augmentations)

• reverse regex

• accept # as intent symbol (as well as % currently) to get closer to IBM Watson's syntax?

• add several files import in one line?

• add some kind of optional version number within template files

New generation features

• accept anything inside an arg, especially unit references

• add a way to force a rule at least once in the training/testing set

• add regex to rasa JSON file

• add default aliases and slots (problem: not language agnostic)

• add examples with multiple intents (see here)

Other features

• add a new part to the library that allows to build template files programmatically
• find a way to make template files edition on phone less cumbersome?
• add a way to pass some arguments from the command to the generator (cf. rodrigopivi/Chatito#84)

Adapters

• Rasa markdown adapter add a small comment at the top of the output file that states it was made with Chatette
• add an adapter to output raw lists of questions (rather than a JSON file) (without entities?)
• add an adapter for Rasa answer templates (the bot's utterances)
• add an adapter for Snips (cf. Chatito)
• add an adapter for Google DialogFlow
• add an adapter for Microsoft LUIS as described here (cf. Chatito)
• add an adapter for IBM Watson
• add an adapter for Flair

Warnings

• warn if there are slots within slots
• warn if the number of examples asked is larger than the maximum number of combinations generatable
• detect and warn about circular references
• warn if a unit reference is used within its own declaration
• warn if the limit of examples generated was reached(?)
• warn if an argument has no value
• check that intent definitions don't overlap

User interface

• improve the error message and the display of the error when specifying an unexisting file as a master template
• add colors to the terminal
• make console help (--help) automatically reflect the possible values of the arguments (e.g. rasamd for the --adapter)

Command line options

• add bulk generation
• add "auto-aliases" feature (cf. Chatito v2.3.4) (or/and as an annotation?)
• add a flag to enable/disable the slot = slot synonym behavior (cf. rodrigopivi/Chatito#50) (or a new syntax => annotation?)
• add a command line option to specifiy the max number of examples to generate
• add program options to change the names of the output files
• add a command line option to run in case insensitive
• add a development command line option that runs the debug in verbose and prevents .pyc and __pycache__/ from being created
• add a flag to make the generators take into account the variability of each rule (cf. rodrigopivi/Chatito#72)

Interactive mode

• fix inconsistent spacing between results and next prompt
• add support for variations
• Interactive mode add support for argument values in relevant commands
• Interactive mode use tabulate to make tables and make command output more readable

Improve code

• fix issues with random generation modifier (see PR #49)
• design patterns
• add more unit tests
• improve logging (remove prints and use a logging library)
• prepare code to be "threadable" by separating generating items in an immutable generator and a mutable builder

• replace print by six.print_?
• refactor facade (and its interactions with parser, generator and adapter)

Improve performance

• use re-exports to simplify the import statements
• improve lexing error messages
• improve the systems of singletons to reduce code duplication
• fix the code duplication between GeneratingItem and ModifiableItem
• use multithreading or multiprocessing to optimize the execution time (+ program option to set that on/off)
• use decorators to make caches and improve code readability

Documentation

• Docs make docs available from help() function

• Docs explain that each rule has the same probability of being chosen, whatever the number of things it can generate

• Add a adapters page which describes all the current adapters

• Docs add a lexicon page in the wiki

• Docs clearly state the objective (scope) of the program

• Docs add a representation of the architecture of the project

• Docs specify which version of Rasa NLU chatette can work with

• Add a FAQ page

• Docs document the differences between Chatito and Chatette

• Docs add a "quickstart" section for the people who already know Chatito

• Docs make a "get started" section that is a simple hands-on example (kind of like the Angular quickstart tutorial)

• Docs add a "contributors" part

• Docs multilingual

• Update documentation by more separating different topics

• Examples remove all deprecated syntax

• Contribute add a part stating how big a contribution needs to be in order for a person to be cited as a contributor

Installation

• make a dockerfile for easy installation and put it on docker hub
• make an installer to use the script directly from the command line (at least in *nix systems)

Branding

• Add image for social networks on GitHub (how the logo to have a brand image, and an image that illustrates what the program is for)

Done

• fix line number count for different files
• add unicode support for all files (io.open autodetects encoding i think)
• add argument support
• generate all possible sentences when asked
• add synonyms into the synonym object of rasa JSON file
• fix overriding of rules
• add entities when generating slot in OOP rewriting
• add casegen inside declarations
• add choice support in OOP rewriting
• slot value should be the whole text and not only the first word if no value is given
• remove escapement in slot values
• add a choose function utility
• give a default nb of indent for printDBG methods
• add support for synonyms in OOP rewriting
• check that slot value in slots content has any use (removed)
• redo as OOP
• support generation without a max number given
• do something with arg when generating all the possibilities
• add a changelog
• change comments symbols from ; to // to more closely resemble Chatito v2.1.x
• parser support Chatito v2.1.x's syntax for asking training and testing generations
• as Chatito v2.1.x does, generate a testing dataset if asked
• don't generate twice the same sentence
• check that casegen is applicable and useful before setting it for a rule content
• add the new training and testing syntax into the syntax specifications
• Docs deprecate semi-colong syntax
• Docs add warning about circular references
• tie the "all possibilities" generation to a maximum
• training and testing datasets should never overlap
• support arg inside synonym lists
• generate_random and generate_all don't have coherent arguments
• allow for (most) special characters without escapement outside of units
• replace modifier regex by several ones to allow for escaping special characters everywhere
• use symbols from parser_utils everywhere needed
• improve the overall command line experience (with argparse)
• change main.py to a more user-friendly name
• accept train AND training for training set number of intents
• add a seed for random number generation
• add a program argument for setting the seed
• set output file path with respect to current working directory rather than input file directory
• add a --version program argument
• use python's built-in DeprecationWarning rather (print a warning for deprecations)
• Docs make a wiki rather than a markdown file
• add interactive mode (generate what the user asks through a CLI)
• Docs explain '/' syntax for alternative slot value
• wipe the output directory before writing new files
• Interactive mode: add support for variations in relevant commands
• Interactive mode: show list of variation names in command show
• check for circular includes
• use more list/dict comprehensions (faster than using append)
• replace getcwd by six.moves.getcwd to be sure to have python 2 and 3 compliant code
• replace range by six.moves.range
• complete refactor of the code: the code is almost unmaintainable
• refactor units to remove duplicated code: make modifiers act after the string has been generated
• Requirements add six
• add an adapter for Rasa markdown
• make the division between processing and lookup more important in parser
• print the seed when starting the program, even if no seed was provided
• add sections in TODO list to make it more readable
• parse in a better way asked number of generation of intents
• add support for any sub-rule modifier for choices
• make double quotes an ignored character in annotations
• add opposite randgen names
• cache the possible number of generatable examples for each unit
• cache the examples when they are all generated and they are not too numerous
• Docs explain that redefining a unit silently appends the rules to the already declared unit (with the same modifiers as the first time) Not applicable anymore

Bugs

To confirm

• BUG: a leading space is generated even though a unit has a random gen modifier and the unit wasn't generated
• BUG: arguments are not given down when an argument is transmitted as the argument of a token

Fixed bugs

• fixed: no case changing when asked with uppercase feeding
• fixed: escapment not currently working
• fixed: slots starting with a word crash the script
• fixed: can't parse when a content line is commented out
• fixed: sometimes the adapter can't find the text entity inside the example
• fixed: choice flattens its rules
• fixed: double spaces sometimes
• fixed: comment lines inside definitions aren't truly ignored
• fixed: fix bug with ']' generated in word groups within choices ({[test ~[this]]/[and ~[this]]})
• fixed: some slot values are not present in the synonym list even if they have synonyms
• fixed: when changing case, some entities cannot be found any longer
• fixed: when no DummySlotValRuleContent, synonyms keep entities
• fixed: duplicates in synonym list
• fixed: in synonyms, too many synonyms are added because variations are not taken into account at generation
• fixed: entities are not generated with generate_all
• fixed: arguments are not correctly managed within entities list
• fixed: potential ImportErrors when running from run.py from the command line
• fixed: somewhere in the synonyms: Rasa NLU can't generate it without crashing (unhashable type 'dict' when looking for "value")
• fixed: parser exists as a basic Python module and is sometimes imported in place of the parser (renamed parsing)
• fixed: empty examples were duplicated on generation when trying to change their leading letter's case
• fixed: if a slot generated a certain string and this string could already be found somewhere before in the example, the first string was wrongly selected as the entity with the Rasa adapter
• fixed: indentation error raised for lines with only spaces
• fixed: when an empty alias definition is used, nothing is generated
• fixed: when using save command, variations print several times the same rule
• fixed: when using save command, \$ are saved instead of $
• fixed: when using save command, choice are appended ?True
• fixed: choices get a randgen when there is 1 ? somewhere in their content
• fixed: several ? in a choice's content crash the program (because of randgen)
• fixed: synonyms are synonyms of themselves (look at simple airport example)
• fixed: restaurant example doesn't seem to work anymore
• fixed: possible to have several times the same example generated
• fixed: encoding errors under Windows
• fixed: the interactive console seems to crash at input()?
• fixed: space at the end of lines in json outputs (fixed in newest versions of json: https://bugs.python.org/issue16333)
• fixed: {my [own?]/a} ~[religion] can generate a double space
• fixed: random generation modifiers' names are not taken into account when generating all examples
• fixed: issues with random generation modifier (see PR #49)

Ideas

Rejected

• rejected: use overrides pip package (no, doesn't work in python 2)
• rejected: make arguments several layers deep? no, it is not useful as you can pass arguments down the references
• rejected: maybe add more modifiers to choices? no, rather merge it into word groups