MinecraftToAIConverter2 is the successor of my old project MinecraftToAIConverter. Since I didn't know much about artificial intelligence and Python back then, it was more of a "project to learn". Therefore, after almost 2 years, I finally decided to revise the project. Well, actually I rather rewrote it.
The old version had many problems. A lot, in fact. I have improved that:
- Structured Code: The old version was a bunch of files and scripts thrown together. I didn't even really use classes. In the new version, everything is in a functional structure.
- Translation to TensorFlow: I don't want to start a dispute or a discussion, so let's see this neither as a contra nor as a pro. I have translated the model from PyTorch to TensorFlow.
- Word2Vec: The old function worked using the well-known Bag of Words method. What we are all thinking now: Shit. But don't worry, this has of course been revised. The new version uses a simple and easily customizable Word2Vec class.
- Multiple datasets: Yes, you heard right. The Train function automatically merges your datasets as long as they all follow the structure described below. So you can swap out all your data and don't have to keep a 1GB .json file.
- Usable response: The old Python server actually only had a random response that belonged to the tag. So the plugin behind it couldn't do anything with it except send it directly to the player. The new code responds in the JSON response, which makes it possible to have more control in the plugin itself. For example, you can add custom quests and menus.
You can integrate the whole thing into your own plugins however you like. If you want the AI to answer something, simply open a TCP socket to the Python backend. But please do not forget that you have to adapt the data set beforehand and then re-execute the Python script. You must send the request in json format, and it must look like this:
{
"sentence": "<what-the-user-wrote>"
}The Python backend will then return two possible answers. If everything was successful:
{
"status": "ok",
"sentence": "<what-the-user-wrote>",
"tag": "<classified-tag>",
"responses": "<list-of-responses>",
"confidence": "<confidence>"
}- status: The status actually only has 2 options. 'error' or 'ok'. With an error, something has gone wrong, with an ok, everything is fine
- sentence: This is simply the sentence that was sent to the AI
- tag: The tag contains the classified tag. As you can see in the dataset structure below, all example records have a large tag. This allows you to tell what the user actually meant
- responses: Responses is a list of strings. To be precise, it contains the responses specified in the dataset
- confidence: Confidence tells you how safe the model is. How likely she thinks it is that the day is really the right one.
Or if an error has occurred:
{
"status": "error",
"error_msg": "<some-error-msg>"
}- status: Same as above
- error_msg: An error message that tells you exactly what was wrong
Sure, I have written an example plugin. Go to the following repo. What the plugin does in short:
You can talk to a villager close to you with the command /ask <msg>. Your sentence is sent to an AI, which
classifies it.
The Dataset is pretty basic. The code is easily extendable, so if you are missing something, have fun extending it. Anyway, here are all the parameters that needs to be given:
- tag: A unique tag/identifier for the intent
- patterns: A list of strings that contains all your example sentences
- responses: A list of strings containing all your answer sentences
NOTE: Your intents must be in an "intents" list, which is absolute. Example:
{
"intents": [
{
"tag": "greeting",
"patterns": [
"hello",
"hi",
"hey",
"howdy",
"good morning",
"good afternoon",
"greetings",
"whats up",
"hows it going"
],
"responses": [
"hello",
"hi there",
"hey",
"howdy",
"good day how can i assist you",
"greetings what brings you here",
"hello how can i help you"
]
},
{
"tag": "bye",
"patterns": [
"adios",
"bye",
"see you",
"see ya",
"farewell",
"goodbye",
"take care",
"until next time"
],
"responses": [
"goodbye",
"farewell",
"see you later",
"take care",
"until next time",
"adios",
"bye for now",
"see ya"
]
}
]
}