merge

.gitattributes

@@ -1,2 +1 @@
-# Auto detect text files and perform LF normalization
-* text=auto
+case/secret_tundra.json filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -1,5 +1,6 @@
 # Data
 *.json
+*.psd
 
 # Logs
 logs

README.md

@@ -1,33 +1,19 @@
-# RoadSign for DOTA 2
+## T-Foresight
 
-## How to start it
+#### This is the repository for submission T-Foresight: Interpret Moving Strategies based on Context-Aware Trajectory Prediction.
 
-1. install `NodeJS`
-2. install `yarn`
-3. run `yarn install` command
-4. run `yarn run dev` command
-5. visit `http://127.0.0.1:5173`
+This repository provides the source codes.
 
-## Main techniques used
+### How to run it
 
-- [React](https://react.dev/): UI framework
-- [MobX](https://mobx.js.org/README.html): Data management
-- [MUI](https://mui.com/material-ui/getting-started/): UI elements
-- [Konva](https://konvajs.org/docs/react/Intro.html): Canvas drawing
-- [I18N](https://react.i18next.com/): Change the language (we need an English version in our paper, and a Chinese version for user study)
+We have a frontend (T-Foresight, in the folder `frontend\`) and a backend (the implementation of our proposed analytics workflow, in the folder `backend\`).
+To save your review time, we provide the analysis progress of our domain experts for review (in the folder `case\`), so that you can just run the frontend and explore our system.
 
-## Project structures
+**Frontend**. Our frontend is built based on Vite, React, and MobX. You can run it in three steps.
+1. *Install dependencies*. Please run `cd frontend` to change the root directory as `frontend\` and then run `yarn install` to install necessary dependencies.
+2. *Run the interface*. Run `run dev` to start the user interface and open `http://localhost:5173` in your explorer (Chrome is recommended). Now, you can see the user interface.
+3. *Load the data*. At the app bar, click on `IMPORT DATA` and select the file `case\secret_tundra.json` to load the game. Then click on `IMPORT CASE` and select one of the other two files, either `secret_tundra-case1.json` or `secret_tundra-case2.json`, to load the analysis progress. Now, you can explore the case!
 
-- `src/components` is for some common and reusable UI elements
-- `src/model` declares the data structure
-- `src/store` declares the data center
-- `src/utils` consists of some tool functions
-- `src/views` contains the main UI designs
-
-## Example files
-
-Here are some example files with detailed comments, which demonstrate how the code works.
-
-- `src/store/store.js` demonstrates how we manage the data
-- `src/store/App.jsx` demonstrates how we render the views based on data
-- `src/model/D2Data.d.ts` demonstrates how we define data types
+**Backend**. Our backend is build based on FastAPI. You can use it in two steps.
+1. *Install dependencies*. Please run `cd backend` to change the root directory as `backend\` and run `pip install -r requirements.text` to install necessary packages.
+2. *Run the server*. Please run `uvicorn main:app --reload` to start the server. But we cannot upload our dataset since it is so large. So you cannot use the functions of the backend.

backend/main.py

@@ -0,0 +1,53 @@
+from fastapi import FastAPI
+
+from src.alg.data import search_inst, search_similar_inst
+from src.alg.interpret import aggregate_workers, proj, extend_stage
+from src.model.model import Model
+
+app = FastAPI()
+
+
+@app.post("/prediction")
+async def root(data):
+    game_name = data["gameName"]
+    team_id = data["teamId"]
+    player_id = data["playerId"]
+    frame = data["frame"]
+    context_limit = data["contextLimit"]
+
+    inst = search_inst('./data', game_name, team_id, player_id, frame)
+    similar_inst = search_similar_inst('./data', inst, team_id, player_id, context_limit)
+
+    model = Model.load('./model/v3')
+    feature_vectors = [[] for _ in model.workers]
+    stages = []
+    for simi_inst in similar_inst:
+        predictions = model.predict(simi_inst)
+        feature_vectors_4_single_inst = [[] for _ in model.workers]
+        for i, worker_pred in enumerate(predictions):
+            fv = []
+            for g_name in worker_pred['attention']:
+                for i_name in worker_pred['attention'][g_name]:
+                    fv.append(worker_pred['attention'][g_name][i_name])
+            for pos in worker_pred['trajectory']:
+                fv.append(pos[0])
+                fv.append(pos[1])
+            feature_vectors[i].extend(fv)
+            feature_vectors_4_single_inst[i].extend(fv)
+
+        pred_groups = aggregate_workers(feature_vectors_4_single_inst)
+        extend_stage(stages, pred_groups)
+
+    pred_groups = aggregate_workers(feature_vectors)
+    pred_projection = proj(pred_groups, feature_vectors)
+
+    return {
+        "predictions": model.predict(inst),
+        "predGroups": pred_groups,
+        "predProjection": pred_projection,
+        "predInstances": {
+            "stages": stages,
+            "numPredictors": model.K,
+            "totalInstances": len(similar_inst)
+        },
+    }

backend/requirements.txt

@@ -0,0 +1,12 @@
+fastapi==0.11.0
+numpy==1.16.3
+numba==0.43.1
+scipy==1.2.1
+matplotlib==3.0.3
+scikit-learn==0.20.3
+
+--find-links https://download.pytorch.org/whl/cu118
+
+torch==2.2.2
+torchvision==0.17.2
+torchaudio==2.2.2