Creating a new dask df using columns from 2 dataframes and keeping the index of the first

Hello! I have 2 dask dfs of same number of rows and partitions. I want to create another one with one column of the first and some columns from the second one, I need to keep the index values of the first dask df. As it is in the code below. It apparently works, Then I do a merge with another dask df, and do a pivot table of the merge .
when I try to save the dask merge using parquet , it doesn’t work and show me the message of “ValueError: Length of values (1769167) does not match length of index (1784153)”. It looks like to me that the creation of pca_comps_df index is not ok. It dosné provide an error, but loos it doesn’t have an index.
Can anyone help me on that?
I also tried to save the dask array of pivot table of the merge andI got the same error above.

pca_comps_df_index = df_toPCA[['orig_index']].copy()
n_compon = 3
comps_sel = ['c'+str(x) for x in range(1,n_compon+1)]
for c in comps_sel:
       pca_comps_df_index[c] = pca_components_df2[c].valuespca_components_df_merge= 

df_toPCA_coords.merge(pca_comps_df_index, how='left', shuffle="tasks")img_comp_df[c]= 
pca_components_df_merge.pivot_table(index='coords_0', columns='coords_1', values=c)

#img_comp_dic[c] = [](

img_comp_dic[c] = img_comp_df.values

[]('df_parquet', engine='pyarrow')

Thanks in advance for any help.

Hi @flaviaas,

It’s really hard to help, as the provided code is not well formated and hard to read. Coul you please try to rework it? Ideally, a minimal reproducible example would be the best.

I don’t see a lot of Dask related code, and also some inplace assignment which is really not recommended with Dask.

Hello @guillaumeeb ,

I am going to try format the code better. Just to provide more information, I have 2 dask dataframes with the same total number of rows (111456171) and same number of partitions (63). Where I want to add one column (orig_index) from the first in the second. After that I want to merge it to df_toPCA_coords, which have the same number of partitions but more rows (111513600).
#apply the same repartition of df_toPCA to pca_components_df2
pca_components_df2 = pca_components_df2.repartition(npartitions=npart)

#create a new dask df: pca_comp_df_index

orig_index = df_toPCA[‘orig_index’].to_dask_array(lengths=True)
pca_comps_df_index[‘orig_index’] = orig_index

pca_components_df_merge= df_toPCA_coords.merge(pca_comps_df_index, how=‘left’, on=‘orig_index’, shuffle=“tasks”)


#save the merge df to parquet files
pca_components_df_merge.to_parquet(‘/merge_files’, engine=‘pyarrow’)

The code show an error when I try to save to parquet files:
ValueError: Length of values (1784153) does not match length of index (1769167)

I checked that pca_comps_df_index also show this error when I tried some comands, for instance, I tryed to verify the number of rows running:
print (pca_comps_df_index.index.pipe(len)
and it show the same error:
ValueError: Length of values (1770994) does not match length of index (1769167)

I think there is a problem in pca_comps_df_index. I tried to create it in different ways and I always have the same error.

Thanks in advance for your help.