Solve a simple task in Keras, Tensorflow, Python
Budget: $30 – $31 USD
This task is about hierarchical reinforcement learning.
The problem description is here: https://gym.openai.com/envs/Taxi-v3/
This problem is considered as Toy text with easy level as is written here:
https://gym.openai.com/envs/#classic_control
What I need from you is to build a single model with keras/tensorflow which will be able to:
1. Learn from a small dataset which represents only 50 gameplays (25 examples of incomplete games and 25 examples of complete game )
2. Predict correctly if a game is completed or not (finished or not)
Dataset example:
[ {'completed': True, 'game': [[0, 0, 3, 2, 0, 0], [1, 0, 3, 2, 0, -1], [2, 0, 3, 2, 0, -1], ......., [4, 0, 2, 2, 5, 20]]},
{'completed': False, 'game': [[2, 3, 0, 3, 1, -1], [1, 3, 0, 3, 1, -1], [0, 3, 0, 3, 1, -1],......, [3, 4, 4, 3, 2, -1]]},
]
Input: gameplay: [[0, 0, 3, 2, 0, 0], [1, 0, 3, 2, 0, -1], [2, 0, 3, 2, 0, -1], ......., [4, 0, 2, 2, 5, 20]]
Output: status or score ( whatever is better)
Based on output we should know if a game is completed or not, in percentage or whatever is better
A game consists of a list of [ state + action + reward ] , [ ]....more states and actions ....
The model should be able to generalize the correct patterns, and predict with high accuracy without overfit/underfit, so that it should be able to predict unseen games very well.
The problem description is here: https://gym.openai.com/envs/Taxi-v3/
This problem is considered as Toy text with easy level as is written here:
https://gym.openai.com/envs/#classic_control
What I need from you is to build a single model with keras/tensorflow which will be able to:
1. Learn from a small dataset which represents only 50 gameplays (25 examples of incomplete games and 25 examples of complete game )
2. Predict correctly if a game is completed or not (finished or not)
Dataset example:
[ {'completed': True, 'game': [[0, 0, 3, 2, 0, 0], [1, 0, 3, 2, 0, -1], [2, 0, 3, 2, 0, -1], ......., [4, 0, 2, 2, 5, 20]]},
{'completed': False, 'game': [[2, 3, 0, 3, 1, -1], [1, 3, 0, 3, 1, -1], [0, 3, 0, 3, 1, -1],......, [3, 4, 4, 3, 2, -1]]},
]
Input: gameplay: [[0, 0, 3, 2, 0, 0], [1, 0, 3, 2, 0, -1], [2, 0, 3, 2, 0, -1], ......., [4, 0, 2, 2, 5, 20]]
Output: status or score ( whatever is better)
Based on output we should know if a game is completed or not, in percentage or whatever is better
A game consists of a list of [ state + action + reward ] , [ ]....more states and actions ....
The model should be able to generalize the correct patterns, and predict with high accuracy without overfit/underfit, so that it should be able to predict unseen games very well.