ValueError: shapes not aligned

[davemate23]

I have a dataframe with 522 observations, but when trying to use this I'm getting the following error from line 241:
ValueError: shapes (73, 3869) and (522, 1) not aligned: 3869 (dim 1) != 522 (dim 0)

I understand that the error is that the two matrices are not aligned, but I don't know where the (73, 3869) matrix was generated from?

[davemate23]

y has a length of 73, so I assume that is part of the first matrix, if that's any use.

[davemate23]

I'm using quarterly data for y and weekly for X, with no NaN values at all. It appears that the R package supports this, but might this be the issue?

[peterlravn]

For anyone still having this problem, it's caused by the following:

 def generate_conversion_matrix(self):

        if self.conversion ==  "sum":
            conversion_weights = np.repeat(1, self.fr).reshape((1, self.fr))

If you go from monthly to weekly, this command will make a 1x12 matrix, but it will only make a 1x4 matrix if going from year to quarter. Therefore, the matrix will be way too large. Instead, what we want is a 1x3 matrix and not a 1x12. By changing "self.fr" to 3, it works fine again. This can be done for "average", "first" and "last".

There's probably a better way to do this, but this will fix it for now.

[peterlravn]

So the problem is, that the author uses the following code:

fr = len(np.unique(df.grain))

This only works with year to quarter, since there's 4 quarters in a year. However, if you go from quarter to month, it will interpret it as 12 month to a quarter, since there's 12 months in a year. Therefore, the code should instead be:

fr = df["y"].value_counts().tolist()[0]

This looks at how many times a certain number is repeated in the "y"-column.

[dipti-gupta-2105]

This Error is because of following code
y = pd.unique(y_check[nan_mask])
it takes unique values of low frequency data , which can be same for two different period
if you don't have any Nans in your data, you can replace this with following code and it will start working

y=df[['index','y']].drop_duplicates()['y'].values

[tyytony]

Alternative solution to the one suggested by dipti-gupta-2105:
I added a sequence of very small numbers to the "y" column, such as 1e-10, 2e-10,...
This ensures the low frequency data values are unique across periods.
It also doesn't change the functions in the package.