vecml-automl（向量自动）

VecML AutoML — One-Command ML Pipeline

Train a model from any CSV in one command. No setup, no notebooks, no boilerplate.

Setup (one time)

export VECML_API_KEY="vml_your_key_here"

Train a Model

Just point it at a CSV and tell it which column to predict:

python3 ~/.openclaw/workspace/skills/vecml-automl/vecml-pipeline.py train data.csv --target Survived

That's it. It will:

1. Auto-detect categorical vs numeric columns
2. Split features and labels
3. Create the project on VecML
4. Upload the data (base64 encoded)
5. Wait for labels to attach (avoids the async race bug)
6. Train the model
7. Show validation metrics (accuracy, AUC, F1, precision, recall)
8. Show feature importance with visual bars

Options

python3 vecml-pipeline.py train data.csv \
  --target target_column \
  --task classification          # or regression
  --mode balanced                # high_speed | balanced | high_accuracy
  --project my_project           # default: openclaw_automl
  --collection my_dataset        # default: auto-generated from filename
  --model my_model_v1            # default: auto-generated

Run Predictions

python3 vecml-pipeline.py predict new_data.csv --model my_model --collection my_dataset

Saves results to new_data_predictions.csv automatically.

List Models

python3 vecml-pipeline.py models --collection my_dataset

Feature Importance

python3 vecml-pipeline.py importance --model my_model --collection my_dataset

Example Output

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  🧠 VecML AutoML Training Pipeline
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  📄 File:       titanic.csv
  🎯 Target:     Survived
  📊 Task:       classification
  ⚡ Mode:       balanced

  [1/6] Creating project...         ✅
  [2/6] Uploading features...       ✅ done! (1.2s)
  [3/6] Attaching labels...         ✅ done! (0.8s)
  [4/6] Training model...           ✅ done! (3.5s)
  [5/6] Validation metrics:
   │ accuracy               0.8101  ████████████████
   │ auc                    0.8798  █████████████████
   │ macro_f1               0.7947  ███████████████
  [6/6] Feature importance:
   │ 🥇 Fare                0.7294  ██████████████
   │ 🥈 Age                 0.6019  ████████████
   │ 🥉 Sex                 0.2732  █████

  ✅ DONE! Accuracy: 81.01%  AUC: 87.98%
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

For OpenClaw Agent Usage

When a user sends a CSV file or asks to train a model, run:

export VECML_API_KEY="vml_your_key_here"
python3 ~/.openclaw/workspace/skills/vecml-automl/vecml-pipeline.py train /path/to/their/file.csv --target their_target_column

If the user doesn't specify a target column, read the CSV headers first and ask which column they want to predict:

head -1 /path/to/file.csv

Then run the pipeline with their chosen target.