Training Fails When Time Ordering Enabled

sdfungayi
sdfungayi Dataiku DSS Core Designer, Dataiku DSS & SQL, Dataiku DSS ML Practitioner, Registered Posts: 19 ✭✭✭✭✭
edited July 16 in Using Dataiku

I am trying to train models with "Time ordering" enabled on the attached dataset I get the error message below, but training sails successfully when "Time ordering" is not enabled. The file is a merger of the "train", "store" and "features" files from https://www.kaggle.com/c/walmart-recruiting-store-sales-forecasting/data.

What I am doing wrong? I changed meanings variables to no avail.

Training failed
Read the logs
Failed to train : <class 'dataiku.doctor.preprocessing.dataframe_preprocessing.DkuDroppedMultiframeException'> : ['target'] values all empty or with unknown classes (you may need to recompute the training set)

The log is included in the attached text file.

 

Answers

  • AlexandreL
    AlexandreL Dataiker, Registered Posts: 36 Dataiker

    Hi,

    I'm trying to reproduce your issue with the data you provided but managed to successfully train a model with time-ordering split. Could you try the following ?

    1. In your model settings > Basic > Target click on the "redetect settings button"
    2. relaunch your training, on the displayed popup tick the "Drop existing sets, recompute new ones" option

    If this doesn't solve your issue could you please provide us additional information ?

    1. Your dss version
    2. A screenshot of your dataset's schema (open the dataset > Settings > Schema)
    3. As I'm not sure to understand the following part "I changed meanings variables to no avail." could you please elaborate on what settings you've changed ?

    Thanks in advance for your feedback and have a good day,

    Alex

  • sdfungayi
    sdfungayi Dataiku DSS Core Designer, Dataiku DSS & SQL, Dataiku DSS ML Practitioner, Registered Posts: 19 ✭✭✭✭✭

    Hi Alexandre

    I tried your suggestion but still got the same error.

    I am using DSS version 9.0.1.

    I have attached the schema of the dataset and images of the error message and small portion of the error log.

  • AlexandreL
    AlexandreL Dataiker, Registered Posts: 36 Dataiker

    Hi,

    I've tried to reproduce your issue with the data you provided and the settings I could see in the logs but unfortunately without success. We might need to investigate a bit more to understand where this issue comes from. Could you please contact our technical support by email at support -at- dataiku -dot- com ? Our support team will guide you but you can start by providing them the logs and data sample you sent me.

    Sorry for not being able to find your issue and good luck.

    Regards,

    Alex

  • acoronel
    acoronel Partner, Dataiku DSS Core Designer, Dataiku DSS ML Practitioner, Dataiku DSS Adv Designer, Registered Posts: 2 Partner

    Hi @sdfungayi

    Pretty old post but.....were you able to solve this issue?
    I was having a similar issue and came across your post. I was having the error:
    dataiku.doctor.preprocessing.dataframe preprocessing.dkudroppedmultiframeexception'> : ['target'] values all empty, infinity or with unknown classes (you may need to recompute the training set)

    I sorted my issue by making sure that storage type and meaning were both discrete in my target variable.
    It was originally storage type = string , meaning = decimal
    Trying to use random forest and logistic regression.

    Once I changed my target variable to
    storage type = string , meaning = Integer
    The ML models were able to run.
    I kept storage type as string, as needed to identify the undefined records vs values (0 and 1)

    In your post you mentioned you tried also changing meanings, any success with that?
    It worked for mine, hope yours as well!!!!!

Setup Info
    Tags
      Help me…