[BUG] Errors in benchmark notebook

[xlsrln]

Description

When running the benchmark notebook for movielens, several errors occur.

First, the ALS model crashes on

recommend_k_als(model, test, train, top_k, remove_seen) 100 (dfs_pred[DEFAULT_USER_COL] == train[DEFAULT_USER_COL]) 101 & (dfs_pred[DEFAULT_ITEM_COL] == train[DEFAULT_ITEM_COL]), --> 102 how="outer", 103 ) in /content/benchmark_utils.py

The error is Column itemID#481 are ambiguous. It's probably because you joined several Datasets together, and some of these Datasets are the same.

Secondly, the NCF step fails with error

RuntimeError: tf.placeholder() is not compatible with eager execution.

The error happens at self.user_input = tf.compat.v1.placeholder(tf.int32, shape=[None, 1]) in recommenders/models/ncf/ncf_singlenode.py in _create_model(self)

In which platform does it happen?

Google colab notebook

How do we replicate the issue?

By running https://github.com/microsoft/recommenders/blob/main/examples/06_benchmarks/movielens.ipynb

Expected behavior (i.e. solution)

It should run without errors

Other Comments

[miguelgfierro]

This looks a problem with the TF version, which version are you using?

[xlsrln]

The NCF errors indeed look like they are fixed by using tensorflow 1.

[xlsrln]

But the spark error isn't