Game Of Thrones Generative ChatBot(GOT-BOT)

Game Of Thrones ChatBot Trained using Seq2Seq Model on Game Of Thrones Subs and Book data

Dependencies

• Python 2.7
• Tensorflow > 1.1
• nltk
• flask

Pretrained Model

• Download Link: https://www.dropbox.com/s/o28ep1sachep2gc/checkpoints_got_bot.zip?dl=0

Model Information

• RNN Encoder-Decoder Model(Seq2Seq)
• MultiRNNCell GRU Cell
• Attention Decoder and Embedding Layer with Buckets
• Sampled SoftMax Loss and SGD Optimizer
• Greedy Response Construction

Dataset Preparation

The Data has been prepared from Game Of Thrones Subtitles(all episodes, https://www.opensubtitles.org) and Game Of Thrones Book(https://archive.org), All pre-processing and post-processing steps are available in Data_Prep file.

GOT SUBS: https://www.opensubtitles.org/en/ssearch/sublanguageid-all/idmovie-63130
GOT-BOOK: https://archive.org/details/A_Game_of_Thrones_Series

It can be used to train on any set of subtitles within the same file.

Test Run

Due to less amount of data, reponses are sometimes too vague.

python Got_Bot.py --m=chat
Initialize new model
Create placeholders
Inference, SampleSoftmax
Creating loss...
Time: 3.61780786514
Create optimizer...
Loading parameters for the Chatbot
After Output--------------------------------
GOT-BOT: Message Limit: 16
--> Who are you
Output Logits: [[-0.30675367 6.68275833 -0.25434929 ..., 0.01963711 -0.30866829
-0.34872866]]
Output Greedy: [151, 4, 1737, 5, 215, 4, 3, 4, 3, 22, 144, 4, 3, 3, 4, 3, 3, 4, 3]
Lannister . Tarth the Yes .
--> Who rules kings landing
Output Logits: [[ 0.45411149 15.40985966 -0.24345914 ..., -0.08361083 0.32688218 0.35243195]]
Output Greedy: [2989, 83, 27, 11, 1232, 450, 3, 7, 590, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3]
Valar men with a Moon Kingdoms
--> where is my army of unsullied
Output Logits: [[-0.78256309 4.53607988 -0.84746158 ..., -0.29136604 -0.5404861 -0.06462321]]
Output Greedy: [13, 8, 36, 83, 13, 35, 14, 748, 39, 5, 2787, 4, 3, 3, 4, 3, 3, 4, 3]
i I have men i He s standing on the battlefield .
--> 'The man who passes the sentence should swing the sword.
Output Logits: [[ 0.45567238 11.5950985 -1.02363682 ..., -0.29802227 -0.66476578 0.07089426]]
Output Greedy: [378, 15, 340, 4, 3, 109, 15, 300, 4, 3, 3, 27, 1731, 4, 3, 4, 3, 3, 3] under his side .
--> The things I do for love.
Output Logits: [[-1.81315196 7.67702293 -1.41315424 ..., -0.63991416 -1.00333619 -1.17256224]]
Output Greedy: [5, 130, 10, 193, 7, 52, 5, 70, 59, 1, 3, 7, 5, 107, 1, 14, 289, 4, 3]
the way you So to do the man Lord
--> It's the family name that lives on. It's all that lives on.
Output Logits: [[-0.94194341 -1.72485757 -0.64410913 ..., 0.37948987 -0.76213503
0.36827829]]
Output Greedy: [48, 469, 14, 2540, 1, 45, 1488, 560, 560, 19, 1102, 778, 109, 5, 537, 4, 3, 87, 14]
And someone s proved they offered which which he commands .. by the air .
--> I learned how to die a long time ago.
Output Logits: [[ 0.43588781 1.38688314 -0.53811967 ..., 0.85255432 -0.06748818 0.17954454]]
Output Greedy: [13, 24, 137, 28, 27, 5, 1, 13, 13, 181, 29, 4534, 4, 3, 3, 3, 3, 13, 3]
i You come him with the i i This is bleed .
--> Turn us away, and we will burn you first. Output Logits: [[ -0.60744536 12.11123943 -0.68960094 ..., -0.04841948 -0.99147111 -0.6090138 ]]
Output Greedy: [1523, 6, 34, 126, 41, 3, 4, 3, 80, 11, 1, 3, 4, 3, 3, 3, 7, 3, 3]
Enough , my lord she
--> A girl gives a man his own name?
Output Logits: [[-0.22489437 3.75130343 0.36403471 ..., 1.052912 -0.42520836
0.18865156]]
Output Greedy: [9, 1641, 10, 228, 238, 56, 3, 6, 3, 91, 22, 3, 4, 3, 3, 4, 3, 3, 4]
and surely you might keep ?
--> A dragon is not a slave.
Output Logits: [[ 0.76116204 5.55503654 -0.32142043 ..., -0.41793767 0.18629865 0.2971383 ]]
Output Greedy: [378, 72, 124, 177, 4, 3, 4, 3, 22, 177, 4, 3, 3, 27, 22, 177, 4, 3, 3]
under their good head .
--> The Lannisters send their regards
Output Logits: [[ 0.29623079 11.3139143 -0.40246141 ..., -0.67723441 -0.5055514 -0.78701621]]
Output Greedy: [5, 117, 6, 1, 184, 4, 3, 26, 11, 100, 3, 3, 4, 3, 3, 3, 3, 3, 3]
the only , again .

Instructions

• Download Pretrained Model and extract in the checkpoints folder
• Run with 'python Got_Bot.py --m=chat'
• There are 3 modes, train, chat and api(for rest api using flask)