tipredict

📊 Model Info

The models in this package were trained on Alibaba Cluster-Trace-v2018.
Measurements for CPU and memory average are based on the instance metrics provided in the trace dataset.

⚠️ CUDA Installation Notice

Since CUDA wheels are not always installable via standard pip install, you must install torch and torch-geometric manually.

✅ Recommended Versions

These versions are tested and compatible (others might work, but ensure compatibility):

• torch==2.0.0+cu118
• torch-geometric==2.6.1
• Requires Python ≥ 3.10

📦 Manual Installation Instructions

pip install torch==2.0.0+cu118 --extra-index-url https://download.pytorch.org/whl/cu118

pip install torch-scatter -f https://data.pyg.org/whl/torch-2.0.0+cu118.html
pip install torch-sparse -f https://data.pyg.org/whl/torch-2.0.0+cu118.html
pip install torch-cluster -f https://data.pyg.org/whl/torch-2.0.0+cu118.html
pip install torch-spline-conv -f https://data.pyg.org/whl/torch-2.0.0+cu118.html

pip install torch-geometric==2.6.1

🧠 Functions

from tipredict import preprocess_GNN, run_pretrained, preprocess_dagtransformer, normalize

📥 Inputs (Note: CSV files must include headers!)

🔹 Input 1: Workflow CSV

Shape: number_workflows × (34 task-level features × number_tasks)
(Predicted task features should be set to 0.)

Task-Level Features (Aggregated over instances):

• count: Count of instances
• mean_ca, var_ca, max_ca, min_ca, med_ca, skew_ca, kurt_ca: Stats for CPU average usage
• mean_cm, var_cm, max_cm, min_cm, med_cm, skew_cm, kurt_cm: Stats for CPU max usage
• mean_ma, var_ma, max_ma, min_ma, med_ma, skew_ma, kurt_ma: Stats for Memory average usage
• mean_mm, var_mm, max_mm, min_mm, med_mm, skew_mm, kurt_mm: Stats for Memory max usage
• mean_t, var_t, max_t, min_t: Stats for running time
• maxtime: Actual runtime

🔹 Input 2: DAG CSV

Shape: number_workflows × ((2 × number_tasks + 1) × number_tasks)

Each task includes 15 features representing graph connectivity.

DAG Feature Example (for 3-task workflow):

(If a node has multiple incoming/outgoing edges, the feature value is 10 / number_of_edges)

# Task 1
1o_1 0         # no edge to 1
1o_2 5         # edge to 2
1o_3 5         # edge to 3
1i_1 0         # no incoming from 1
1i_2 0
1i_3 0

# Task 2
2o_1 0
2o_2 0
2o_3 10        # edge to 3
2i_1 10        # incoming from 1
2i_2 0
2i_3 0

# Task 3
3o_1 0
3o_2 0
3o_3 0
3i_1 5
3i_2 5
3i_3 0

⚙️ Applications

# GNN preprocessing
preprocess_GNN(num_tasks, feature_csv, dag_csv)

# DAGTransformer preprocessing
preprocess_dagtransformer(num_tasks, feature_csv, dag_csv)

# Normalize task-level features (tasks × 34)
normalize(features)

# Run pretrained models
run_pretrained(
    model="GAT",                    # or "GIN", "GraphSAGE", "GNN_ensemble", "DAGTransformer", "DT&GNN"
    num_tasks=7,
    memory=True,                   # if False, predicts CPU
    data_GNN=...,                  # output of preprocess_GNN
    data_DAGTransformer=...,       # output of preprocess_dagtransformer
    penultimateLayer=False         # if True, returns penultimate layer output
)