recommenders | Best Practices on Recommendation Systems | Recommender System library

The Pandas Documentation states:

While pivot() provides general purpose pivoting with various data types (strings, numerics, etc.), pandas also provides pivot_table() for pivoting with aggregation of numeric data

Make sure the column is numeric. Without seeing how you create trainingtaken I can't provide more specific guidance. However the following may help:

1. Make sure you handle "empty" values in that column. The Pandas guide is a very good place to start. Pandas points out that "a column of integers with even one missing values is cast to floating-point dtype".
2. If working with a dataframe, the column can be cast to a specific type via your_df.your_col.astype(int) or for your example, pd.trainingtaken.astype(int)

You seem to be preprocessing your data incorrectly. For example, you cannot use tf.strings.to_number to convert 00001446-da5f-4d17 into some number. It will throw an error, since the string contains more than just numbers. Also, each sample in your dataset was an array instead of a single sample: Channel 1, for example, was not 1, but [1]. This was the cause of the original problem in your question. Here is a simplified working example based on your code:

Solution

• I switched to AWS EC2 SageMaker "Python [conda env:tensorflow2_p36] " so this is the exact pre made environment "tensorflow2_p36"
• As I ahev read it in some palces it is probably library collision maybe with NumPy.

You can use tf.strings.to_hash_bucket_strong to hash your strings to indices, as long as you don't care about the mapping order.

Example:

I have checked the code on my end an it seems to be a syntax error when calling the method. As per the library documentation the first parameter should be a request object or None and the parent parameter has to be passed by name. Changing the line as follows I didn't have the error:

I assume that you are using MS SQL based on your Syntax. In that case you can't reference a column_alias in another column:

column_alias can be used in an ORDER BY clause. However, it cannot be used in a WHERE, GROUP BY, or HAVING clause. If the query expression is part of a DECLARE CURSOR statement, column_alias cannot be used in the FOR UPDATE clause.

I think it's relatively easy to work around this with an CTE:

This seems to be related to the Java used to run the Ant script.

In the run configuration, in the JRE tab, make sure a JRE/JDK is selected (for an existing Ant run configuration it can happen that the previously selected JRE/JDK is lost due to changes in the preferences Java > Installed JREs or when updating Java). If a JRE/JDK is already selected, try another JRE/JDK setting here (if there are VM arguments specify, try it without them, since they might be illegal for the chosen VM).

TLDR; The model_selection documentation in Surprise indicates a "refit" method, that will fit data on a full trainset, however it explicitly doesn't work with predefined folds.

Another major issue: oyyablokov's comment on this issue suggests you cannot fit a model with data that has NaNs. So even if you have a full trainset, how does one create a full prediction matrix to calculate things like prediction coverage, which requires all users and item combinations with or without ratings?

My workaround was to create 3 Surprise datasets.

1. The dataset from predefined folds to compute best_params
2. The full dataset of ratings (combining all folds outside of Surprise)
3. The full prediction matrix dataset including all possible combinations of users and items (with or without ratings).

After you find your best paramaters with grid search cross validation, you can find your predictions and coverage with something like this:

I would look at this example.

Namely, if you render the template like it is shown in the example, you should be able to do