Metaflow

Search…
Metaflow
Overview
​Metaflow is a framework created by Netflix for creating and running ML workflows.
This integration lets users apply decorators to Metaflow steps and flows to automatically log parameters and artifacts to W&B.
Decorating a step will enable or disable logging for certain types within that step.
Decorating the flow will enable or disable logging for every step in the flow.
Quickstart
Install W&B and login
Notebook
Command Line
1
!pip install -Uqqq metaflow fastcore wandb
2
​
3
import wandb
4
wandb.login()
Copied!
1
pip install -Uqqq metaflow fastcore wandb
2
wandb login
Copied!
Decorate your flows and steps
Step
Flow
Both flow and steps
Decorating a step will enable or disable logging for certain types within that Step.
In this example, all datasets and models in start will be logged
1
from wandb.integration.metaflow import wandb_log
2
​
3
class WandbExampleFlow(FlowSpec):
4
    @wandb_log(datasets=True, models=True, settings=wandb.Settings(...))
5
    @step
6
    def start(self):
7
        self.raw_df = pd.read_csv(...).    # pd.DataFrame -> upload as dataset
8
        self.model_file = torch.load(...)  # nn.Module    -> upload as model
9
        self.next(self.transform)
Copied!
Decorating a flow is equivalent to decorating all the constituent steps with a default.
In this case, all steps in WandbExampleFlow will log datasets and models by default -- the same as decorating each step with @wandb_log(datasets=True, models=True)
1
from wandb.integration.metaflow import wandb_log
2
​
3
@wandb_log(datasets=True, models=True)  # decorate all @step 
4
class WandbExampleFlow(FlowSpec):
5
    @step
6
    def start(self):
7
        self.raw_df = pd.read_csv(...).    # pd.DataFrame -> upload as dataset
8
        self.model_file = torch.load(...)  # nn.Module    -> upload as model
9
        self.next(self.transform)
Copied!
Decorating the flow is equivalent to decorating all steps with a default. That means if you later decorate a Step with another @wandb_log, you will override the flow-level decoration.
In the example below:
start and mid will log datasets and models, but
end will not log datasets or models.
1
from wandb.integration.metaflow import wandb_log
2
​
3
@wandb_log(datasets=True, models=True)  # same as decorating start and mid
4
class WandbExampleFlow(FlowSpec):
5
  # this step will log datasets and models
6
  @step
7
  def start(self):
8
    self.raw_df = pd.read_csv(...).    # pd.DataFrame -> upload as dataset
9
    self.model_file = torch.load(...)  # nn.Module    -> upload as model
10
    self.next(self.mid)
11
​
12
  # this step will also log datasets and models
13
  @step
14
  def mid(self):
15
    self.raw_df = pd.read_csv(...).    # pd.DataFrame -> upload as dataset
16
    self.model_file = torch.load(...)  # nn.Module    -> upload as model
17
    self.next(self.end)
18
​
19
  # this step is overwritten and will NOT log datasets OR models
20
  @wandb_log(datasets=False, models=False)
21
  @step
22
  def end(self):
23
    self.raw_df = pd.read_csv(...).    
24
    self.model_file = torch.load(...)
Copied!
Where is my data? Can I access it programmatically?
You can access the information we've captured in three ways: inside the original Python process being logged using the wandb client library, via the web app UI, or programmatically using our Public API. Parameters are saved to W&B's config and can be found in the Overview tab. datasets, models, and others are saved to W&B Artifacts and can be found in the Artifacts tab. Base python types are saved to W&B's summary dict and can be found in the Overview tab. See our guide to the Public API for details on using the API to get this information programmatically from outside .
Here's a cheatsheet:
Data
Client library
UI
Parameter(...)
wandb.config
Overview tab, Config 
datasets, models, others
wandb.use_artifact("{var_name}:latest")
Artifacts tab
Base Python types (dict, list, str, etc.)
wandb.summary
Overview tab,  Summary
wandb_log kwargs
kwarg
Options
datasets
True: Log instance variables that are a dataset
False
models
True: Log instance variables that are a model
False
others
True: Log anything else that is serializable as a pickle
False
settings
wandb.Settings(...): Specify your own wandb settings for this step or flow
None: Equivalent to passing wandb.Settings()
By default, if:
settings.run_group is None, it will be set to {flow_name}/{run_id}
settings.run_job_type is None, it will be set to {run_job_type}/{step_name}
Frequently Asked Questions
What exactly do you log? Do you log all instance and local variables?
wandb_log only logs instance variables.  Local variables are NEVER logged.  This is useful to avoid logging unnecessary data.  
Which data types get logged?
We currently support these types:
Logging Setting
Type
default (always on)
dict, list, set, str, int, float, bool
datasets
pd.DataFrame
pathlib.Path
models
nn.Module
sklearn.base.BaseEstimator
others
Anything that is pickle-able​
Examples of logging behaviour
Kind of Variable
Behaviour
Example
Data Type
Instance
Auto-logged
self.accuracy
float
Instance
Logged if datasets=True
self.df
pd.DataFrame
Instance
Not logged if datasets=False
self.df
pd.DataFrame
Local
Never logged
accuracy
float
Local
Never logged
df
pd.DataFrame
Does this track artifact lineage?
Yes!  If you have an artifact that is an output of step A and an input to step B, we automatically construct the lineage DAG for you.
For an example of this behaviour, please see this notebook and its corresponding W&B Artifacts page​
Kubeflow
MMF
Last modified 2mo ago
Copy link
Contents
Overview
Quickstart
Install W&B and login
Decorate your flows and steps
Where is my data? Can I access it programmatically?
wandb_log kwargs
Frequently Asked Questions
What exactly do you log? Do you log all instance and local variables?
Which data types get logged?
Examples of logging behaviour
Does this track artifact lineage?