Paso 1: Export Data desde W&B (Python Script)

import wandb
import json
import os
from tqdm import tqdm

# Initialize W&B API
api = wandb.Api()

# Target entity/project
ENTITY = "your-team"
PROJECT = "your-project"

# Fetch all runs
print(f"Fetching runs from {ENTITY}/{PROJECT}...")
runs = api.runs(f"{ENTITY}/{PROJECT}")
print(f"Found {len(runs)} runs to export")

exported_data = []
for run in tqdm(runs, desc="Exporting runs"):
    # Extract run metadata
    run_data = {
        "run_id": run.id,
        "run_name": run.name,
        "created_at": run.created_at,
        "state": run.state,
        "config": dict(run.config),  # Hyperparameters
        "summary": dict(run.summary),  # Final metrics
        "tags": run.tags,
        "notes": run.notes,
        "history": [],
        "artifacts": []
    }

    # Export full history (step-by-step metrics)
    history = run.history(pandas=False)
    run_data["history"] = history

    # List artifacts (models, plots, files)
    for artifact in run.logged_artifacts():
        run_data["artifacts"].append({
            "name": artifact.name,
            "type": artifact.type,
            "size": artifact.size,
            "download_url": artifact.file()
        })

    exported_data.append(run_data)

# Save to JSON file
output_file = f"wandb_export_{PROJECT}.json"
with open(output_file, 'w') as f:
    json.dump(exported_data, f, indent=2, default=str)

print(f"✅ Exported {len(exported_data)} runs to {output_file}")
print(f"Total size: {os.path.getsize(output_file) / 1024 / 1024:.2f} MB")

import mlflow
import json
from tqdm import tqdm
from datetime import datetime

# Setup MLflow tracking URI
MLFLOW_TRACKING_URI = "http://your-mlflow-server:5000"
mlflow.set_tracking_uri(MLFLOW_TRACKING_URI)

# Load exported W&B data
with open("wandb_export_your-project.json", 'r') as f:
    wandb_data = json.load(f)

print(f"Loaded {len(wandb_data)} runs from W&B export")

# Create MLflow experiment (equivalent to W&B project)
experiment_name = "migrated_from_wandb_your-project"
try:
    experiment_id = mlflow.create_experiment(experiment_name)
except:
    experiment_id = mlflow.get_experiment_by_name(experiment_name).experiment_id

mlflow.set_experiment(experiment_name)

# Import cada run
for run_data in tqdm(wandb_data, desc="Importing to MLflow"):
    with mlflow.start_run(run_name=run_data["run_name"]) as mlflow_run:
        # Import config (hyperparameters) como params
        for key, value in run_data["config"].items():
            try:
                # MLflow params son strings (convert si necesario)
                mlflow.log_param(key, str(value)[:500])  # Max 500 chars
            except Exception as e:
                print(f"⚠️ Could not log param {key}: {e}")

        # Import summary metrics (final values)
        for key, value in run_data["summary"].items():
            try:
                if isinstance(value, (int, float)):
                    mlflow.log_metric(key, value)
            except Exception as e:
                print(f"⚠️ Could not log metric {key}: {e}")

        # Import history (step-by-step metrics)
        for step_data in run_data.get("history", []):
            step = step_data.get("_step", 0)
            for key, value in step_data.items():
                if key.startswith("_"):
                    continue  # Skip W&B internal fields
                try:
                    if isinstance(value, (int, float)):
                        mlflow.log_metric(key, value, step=step)
                except Exception as e:
                    print(f"⚠️ Could not log history metric {key}: {e}")

        # Import tags
        for tag in run_data.get("tags", []):
            mlflow.set_tag("wandb_tag", tag)

        # Set metadata
        mlflow.set_tag("migrated_from", "wandb")
        mlflow.set_tag("original_run_id", run_data["run_id"])
        mlflow.set_tag("original_created_at", str(run_data["created_at"]))

        # Notes como tag (MLflow no tiene "notes" nativo)
        if run_data.get("notes"):
            mlflow.set_tag("notes", run_data["notes"][:5000])

print(f"✅ Imported {len(wandb_data)} runs to MLflow experiment '{experiment_name}'")