Upload README.md

README.md

@@ -0,0 +1,89 @@
+<h1 align="center">
+Commonsene Object Affordance Task [COAT]
+</h1>
+  
+<p align="center">
+  <br>
+  <a href="https://openreview.net/pdf?id=xYkdmEGhIM">OpenReview</a> | <a href="https://drive.google.com/drive/u/4/folders/1reH0JHhPM_tFzDMcAaJF0PycFMixfIbo">Datasets</a>
+</p>
+
+<p align="center">
+<img src="https://github.com/com-phy-affordance/COAT/blob/main/utility-intro(1).png" alt="Paper Summary Flowchart">
+<em>A 3 level framework adumbrating human commonsense style reasoning for estimating object affordance for various tasks</em>
+</p>
+
+### Experimental Setup:
+- Task List: [tasks](https://github.com/com-phy-affordance/com-affordance/blob/main/tasks.json)
+- Object List: [objects](https://github.com/com-phy-affordance/com-affordance/blob/main/concepts.json)
+- Utility List[^1]: [utilities](https://github.com/com-phy-affordance/com-affordance/blob/main/concepts.json)
+- Variables Used:
+  ```temperature```, ```mass```, ```material```, ```already-in-use```, ```condition```
+  
+### Utility Level Pruning:
+This gives us ```Utility```to``Object`` mappings also called ```utility objects```
+- GT Object-Utility Mappings : [utility-mappings](https://github.com/com-phy-affordance/com-affordance/blob/main/objects.json)
+
+### Task-u(Utility Level):
+Here we evaluate models on their ability to prune out appropriate objects on the basis of Utility. 
+- GT (Utility)-(Object) Mappings: [utility-objects](https://github.com/com-phy-affordance/com-affordance/blob/main/objects.json)
+- Task-u Dataset: [4 Variations](https://drive.google.com/drive/folders/1JJSIicKGp0a7ThsenKl0XWKsTtPL_b5z?usp=sharing)
+  
+### Task-0(Context Level):
+Here we evaluate models on their ability to prune out appropriate objects on the basis of Context. This gives us ```(Task,Utility)```to``Object`` mappings also called ```context objects```
+- GT (Task-Utility)-(Object) Mappings: [context-objects](https://github.com/com-phy-affordance/com-affordance/blob/main/oracle.json)
+- Task-0 Dataset: [4 Variations](https://drive.google.com/drive/folders/1reH0JHhPM_tFzDMcAaJF0PycFMixfIbo?usp=sharing)
+
+### Task-1(Physical State Level):
+Here we evaluate models on their ability to prune out the ```ideal``` configuration when presented with a number of ```context object``` configurations. (Something that is pretty obvious to humans)
+- All Possible Common Configurations: [possible configurations](https://github.com/com-phy-affordance/com-affordance/blob/main/task-1/possible_configurations_v1.json)
+- Ideal Configurations: [ideal configurations](https://github.com/com-phy-affordance/com-affordance/blob/main/task-1/pouch_config_oracle.json)
+- Commonsense Common Occurence Variables: [common variables values](https://github.com/com-phy-affordance/com-affordance/blob/main/task-1/common_var_responses.json)
+- Task-1 Dataset: [12 Variations](https://drive.google.com/drive/folders/1reH0JHhPM_tFzDMcAaJF0PycFMixfIbo?usp=sharing)
+
+### Task-2(Physical State Level):
+Here we evaluate models on their ability to prune out the most appropriate```sub-optimal``` configuration when presented with a number of sub-optimal configurations of ```context objects```. (Something that is pretty obvious to humans)
+- Suboptimal Configurations: [suboptimal configurations](https://github.com/com-phy-affordance/com-affordance/blob/main/task-2/pouch_suboptimal.json)
+- Human Preference Material Order: [material preference](https://github.com/com-phy-affordance/com-affordance/blob/main/task-2/material_preference.json)
+- Task-2 Dataset: [14 Variations](https://drive.google.com/drive/folders/1reH0JHhPM_tFzDMcAaJF0PycFMixfIbo?usp=sharing)
+---------------------------------------------------------------------------------------------------------------
+
+### Finetuning Datasets 
+
+Please refer to [Appendix F.1](https://openreview.net/pdf?id=xYkdmEGhIM) for dataset details
+
+- Finetuning Dataset for Object Level Selection : [Google Drive Link](https://drive.google.com/drive/folders/1GtrGQxTTtYEczYK1ytB71Y2HGxM1TEu5?usp=drive_link)
+- Finetuning Dataset for Physical State Level Selection : [Google Drive Link](https://drive.google.com/drive/folders/1FiZc8u_G8wUrN4NroZmIgmcTe0jor72T?usp=drive_link)
+
+### Full Pipeline Evaluation Datasets
+
+Please refer to [Appendix F.2](https://openreview.net/pdf?id=xYkdmEGhIM) for dataset deatails
+
+- Ideal Object Choice Datasets : [Google Drive Link](https://drive.google.com/drive/folders/1SMM2TU1BKH32oKtfmW0gS3QfyUA68IZ0?usp=drive_link)
+- Moderate Object Choice Datasets : [Google Drive Link](https://drive.google.com/drive/folders/1SlZQBp4Iao3VHnmOFZMKfzn_LWOctnVE?usp=drive_link)
+
+
+<h3>Prompts Used</h3>
+<p>
+  
+  [Quantitative Examples](https://giant-licorice-a62.notion.site/Prompts-for-Appendix-Examples-d58e0184d1c546bd8632024de3f7ac25)
+</p>
+
+### Implementations For Language Models:
+- PaLM/GPT3.5-Turbo: API
+- LLama13B: huggingface text generation pipeline [link](https://huggingface.co/blog/llama2)
+- Vicuna13B: lmsys [link](https://github.com/lm-sys/FastChat)
+- Vicuna7B: lmsys [link](https://github.com/lm-sys/FastChat)
+- Mistral-7B: huggingface [link](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1)
+- ChatGLM-6B: huggingface [link](https://huggingface.co/THUDM/chatglm-6b)
+- ChatGLM2-6B: huggingface [link](https://github.com/THUDM/ChatGLM2-6B)
+
+[^1]: For the purpose of datasets, we've used `concept and utility` interchangeably.
+----------------------------------------------------------------------------------------------------------------
+### Upcoming Stuff:
+- generating object, task, utility jsons for your purpose 
+- generating task-0 datasets for your own task list, object list, utility lists
+- generating task-1, task-2 datasets for your own variables, your preferred possible configurations, handcrafted penalty schema and your own preferences.
+
+> play around, create more variables, go for more comprehensive reward structures, go in any depth you wish. Let's create more agents capable of physical commonsense reasoning!
+
+PS: If you need any help experimenting with this data or curating your own datasets, feel free to create an Issue.