将Numpy数组插入并分类到Django建模数据库EAV模式中

Question

I have a Pandas pivot table of the format:

income_category     age_category      income         age
High                Middle aged       123,564.235    23.456
Medium              Old               18,324.356     65.432

I have a category hierarchy with matching label s in a self-referencing table called dimension . Ie,

dimension_id       label             parent_dimension_id
1                  Age categories
2                  Young             1
3                  Middle aged       1
4                  Old               1

...and similarly for income

I'm really struggling to pick a row at a time and access cells in that row randomly.

I have the parent category id dimension_id (in the code below it is already in cat_id_age ). So I want to iterate through the Numpy array, getting the matching category dimension_id for that row, and insert it into a value table along with its corresponding value. However I've no idea how to do this Pythonically or Djangonically. (There are only a few categories so I think the Dictionary approach below for looking up dimension_id is best.) To my iterative mind the process is:

# populate a Dictionary to find dimension_ids
age_dims = Dimension.objects.filter(parent_id=cat_id_age).values('label', 'id')

for row in Numpy_array:

    dim_id = Dimension.get(row.age_category)

    # Or is the Dict approach incorrect? I'm trying to do: SELECT dimension_id FROM dimension WHERE parent_dimension_id=cat_id_age AND label=row.age_category
    # Djagonically? dim = Dimension.objects.get(parent_id=cat_id_age, label=row.age_category)

    # Then insert categorized value, ie, INSERT INTO float_value (value, dimension_id) VALUES (row.age, dimension_id)
    float_val = FloatValue(value=row.age, dimension_id=dim_id)
    float_val.save()

...then repeat for income_category and income.

However I'm struggling with iterating like this - that may be my only problem but I've included the rest to communicate what I'm trying to do as I often seem a paradigm away Python (eg, sth like cursor.executemany("""insert values(?, ?, ?)""", map(tuple, numpy_arr[x:].tolist())) ?).

Any pointers really appreciated. (I'm using Django 1.7 and Python 3.4.)

Answer 1

Anzel answered the iterating problem here - use the Pandas to_csv() function . My dictionary syntax was also wrong. My final solution was therefore:

# populate a Dictionary to find dimension_ids for category labels
parent_dimension_age = Dimension.objects.get(name='Age')
parent_dimension_income = Dimension.objects.get(name='Income')
dims_age = dict([ (d.name, d.id) for d in Dimension.objects.filter(parent_id=parent_dimension_age.id) ])
dims_income = dict([ (d.name, d.id) for d in Dimension.objects.filter(parent_id=parent_dimension_income.id) ])

# Retrieves a row at a time into a comma delimited string
for line in pandas_pivottable.to_csv(header=False, index=True, sep='\t').split('\n'):
    if line:
        # row[0] = income category, row[1] = age category, row[2] = age, row[3] = income
        row = line.split('\t')
        entity = Entity(name='data pivot row', dataset_id=dataset.id)
        entity.save()
        # dims_age.get(row[1]) gets the ID for the category whose name matches the contents of row[1]
        age_val = FloatValue(value=row[2], entity_id=entity.id, attribute_id=attrib_age.id, dimension_id=dims_age.get(row[1]))
        age_val.save()
        income_val = FloatValue(value=row[3], entity_id=entity.id, attribute_id=attrib_income.id, dimension_id=dims_income.get(row[0]))
        income_val.save()

For more on the Entity-Attribute-Value (EAV) schema see the Wikipedia page , (if you are considering it see the Django-EAV extension ). In the next iteration of this project however, I will be replacing it with postgresql's new JSONB type . This promises to make the data more legible and perform equally or better.