Trying to replicate the quick start example (JOSS review 2)

[ArneTillmann]

When I tried to replicate the example from the introduction the google colab instructions were very helpful for me as I have never used Julia so far. However I could not replicate the example. The code threw two errors:

After commenting out line 3

I suggest you rerun the examples including those in the tutorial on a new machine to check whether they work as presented if you haven't done so already. I will double check if you give the go.

[EssamWisam]

Thank you for reporting this issue.

I'm quite certain that I ran all the tutorials and the introduction before publishing them in the documentation. It's also why they are presented with the output shown.

Colab support to Julia is a little iffy (not officially supported and running it requires a hacky solution) but I would be happy to see the specific notebook from which you extracted the screenshots. Can you share it in view only mode and I will take a copy and see what could have went wrong?

[EssamWisam]

Now I understand that you were trying to replicate the introduction but using TableTransforms instead. The fact is that you can't be using multiple interfaces at once without specifying from which the model will come (e.g., both MLJ and TableTransforms) as the same model name is used for each.

[ArneTillmann]

Here is the notebook.

[EssamWisam]

Okay as I expected you need your last cell to be

using Imbalance
using TableTransforms

# Generate imbalanced data
num_rows = 200
num_features = 5
y_ind = 3
Xy, _ = generate_imbalanced_data(num_rows, num_features;
                                 class_probs=[0.5, 0.2, 0.3], insert_y=y_ind, rng=42)

# Initiate SMOTE model
oversampler = Imbalance.TableTransforms.SMOTE(y_ind; k=5, ratios=Dict(0=>1.0, 1=> 0.9, 2=>0.8), rng=42)
Xyover = Xy |> oversampler       # can chain with other table transforms
# equivalently if TableTransforms is used
Xyover, cache = TableTransforms.apply(oversampler, Xy)    # equivalently

Since you are also using MLJ and the same model name is used in the TableTransforms interface for consistency. The assumption here is that the two interfaces are meant to be serving different audiences; it could quite odd for someone using MLJ in their project to prefer using the table transforms interface over the MLJ interface. Under the unliekly occassion that someone uses both, the error should signify that there is a crash in the names and I will also add in the README/main page in the documentation a hint for this.

[jbytecode]

Due to the JOSS submission

https://github.com/openjournals/joss-reviews/issues/6310

[ArneTillmann]

Thank you for clarifying

[EssamWisam]

I may add that it's best for you to try the algorithms offline. Since Colab for Julia is afterall a hacky approach (and too slow), I will likely jettison it from the docs to avoid confusing new users. Will then have it back should support to Julia be ever official.