(THE FOLLOWING README WAS HANDED IN AS PART OF THE PROJECT)

RUN WITH PYTHON 3
Note: this was developed on a Linux machine, I have had memory issues (specifically, MemoryError, with a failure to allocate about 850 megs) running this on Windows which I am trying to figure out how to resolve.

PACKAGES TO INSTALL

textblob
fuzzywuzzy pandas sklearn numpy tqdm

you can easily do this by running my install packages script :)
probably needs root access, though. Maybe my calls to pip need root access as well, not just the script.

ALSO NEEDED (for textblob creation)

python -m textblob.download_corpora

RUNNING THE PROGRAM

My package scanner/installer takes in a list of file names as arguments. Easy enough.

My real project is easy enough as well: if no argument is provided, the program will create a dataframe from the massive amount of data it has in memory. This is time consuming (takes ~5 minutes on my laptop). It will save it as a csv for future use, however. I provided the csv, but if you have doubts or just wanna run it fresh yourself, go run and get something to eat in the meantime. Fair warning.

python3 final.py

HOW IT "WORKS"

• A dictionary of sentences/nlp strings is made, mapping a statement to a response
• A Markov chain is made in a multimap-esque structure, where the initial key is a real word and it is mapped onto a list of POS tags, each of which is a key to another map to a list of real words that follow the original key.
• A dataframe is created (if not user-supplied) where the stats of the parts of speech per statement are tallied, and associated with the nlp string associated with the statement's response (from the main dictionary)
• A decision tree is trained on this dataframe.
• The main loop starts up, prompting the user for input
• The user's input is tokenized and shoved into the model for prediction
• When the model finds a prediction, the Markov chain springs into action and populates the response with appropriate words that fit the current part-of-speech in the phrase that most closely follows the previously selected word.
• Repeat until user enters 'quit()' or hits CTRL+C

BUGS

• Sometimes it can't find a word in the markov chain, so it crashes. (Should be handled as of 05-05-18) This is probably due to the off-by-one model of the keys vs the values in the main dictionary: on the slim (but apparently existent) chance that a word exists in one but not the other, it'll crash.

• Wouldn't call it a bug, but it usually doesn't generate sentences that are gramatically correct.