[CODING QUESTION] 'Data' object has no attribute 'is_ok' about ax

[NoobIamNoob]

I found a quick fix solution :

1 - at the begining of your srcipt you import the Ok class :

from ax.utils.common.result import Ok

2 - In your custom metric class, for the fetch_trial_data function, you return the Ok object initialized with your Data. In my case the fetch_trial_data function looks like this now :

def fetch_trial_data(self, trial, **kwargs):
    records = []
    for arm_name, arm in trial.arms_by_name.items():
        # Extracting the result from trial metadata
        result = retrieve_result(trial.index, arm_name, self.result_directory)

        records.append({
            "arm_name": arm_name,
            "metric_name": self.name,
            "mean": result,
            "sem": 0.0,  # Standard error of the mean
            "trial_index": trial.index,
        })
    data = Data(df=pd.DataFrame.from_records(records))
    return Ok(data)

Good afternoon a-rouxel!

I don't know how it works but, It perfectly worked!
Thank you for your help!
Have a good day today! : )

I added

from ax.utils.common.result import Ok
and
for arm_name, arm in trial.arms_by_name.items():
in the metric function

for calling the AX functions
##Call AX functions 
from ax import (
    ComparisonOp,
    ParameterType,
    RangeParameter,
    ChoiceParameter,
    FixedParameter,
    SearchSpace,
    Experiment,
    OutcomeConstraint,
    OrderConstraint,
    SumConstraint,
    OptimizationConfig,
    Objective,
    Metric,
    Data
)
from ax.utils.notebook.plotting import render, init_notebook_plotting

init_notebook_plotting()
from ax.utils.common.result import Ok #I added this part

for metric function,

#Create Metric
from ax import Metric
import pandas as pd

class Discrep_Metric(Metric):
    def fetch_trial_data(self, trial):
    
        records = []

        for arm_name, arm in trial.arms_by_name.items():        
       
            #Call MSE calculation that is defined from global env   
            mse = MSE_CALCULATION() 

            # Record
            records.append({
                "arm_name": arm_name,
                "metric_name": self.name,
                "mean": mse,  # MSE 값
                "sem": 0.0, 
                "trial_index": trial.index,
            })
            
        data = Data(df=pd.DataFrame.from_records(records))
        return Ok(data)

[a-rouxel]

I have the same issue here. In the docs, the fetch_trial_data function from the Custom metric should return a Data object, however, in the base function from the metric.py file, it is indicated that it should return a MetricFetchResult object that is defined like this : Result[Data, MetricFetchE] .... There is something wrong here but i am not sure how to fix it. I'll keep you updated if i find a good solution !

[a-rouxel]

I found a quick fix solution :

1 - at the begining of your srcipt you import the Ok class :

from ax.utils.common.result import Ok

2 - In your custom metric class, for the fetch_trial_data function, you return the Ok object initialized with your Data.
In my case the fetch_trial_data function looks like this now :

def fetch_trial_data(self, trial, **kwargs):
    records = []
    for arm_name, arm in trial.arms_by_name.items():
        # Extracting the result from trial metadata
        result = retrieve_result(trial.index, arm_name, self.result_directory)

        records.append({
            "arm_name": arm_name,
            "metric_name": self.name,
            "mean": result,
            "sem": 0.0,  # Standard error of the mean
            "trial_index": trial.index,
        })
    data = Data(df=pd.DataFrame.from_records(records))
    return Ok(data)