Make it easier to manually create DataFrames

We often need to create Minimal Complete Verifiable Examples with Dask DataFrames for blog posts, questions, etc.

Here’s how Dask allows you to make these examples currently:

import pandas as pd

pandas_df = pd.DataFrame(
    {"num1": [1, 2, 3, 4], "num2": [7, 8, 9, 10]},
)
df = dd.from_pandas(pandas_df, npartitions=2)

This is confusing for some users. They are not sure why pandas needs to be imported and can struggle making the pandas => Dask connection.

The following syntax would make the MCVEs shorter and easier to understand:

import dask.dataframe as dd

ddf = dd.DataFrame(
    {"num1": [1, 2, 3, 4], "num2": [7, 8, 9, 10]},
    npartitions=2,
)

Let me know if this suggestion sounds good and I can submit a PR. Or perhaps there is already a way to directly create a Dask DataFrame in a similar manner?

[jrbourbeau]

Thanks @MrPowers, I think this is a reasonable request but wouldn’t want to see us point users directly to the dd.DataFrame constructor as it’s a fairly low-level class constructor and we prefer to route users towards more friendly methods like dd.read_parquet, dd.from_delayed, etc.

That said, adding a new dd.from_dict method similar to pandas’ pd.from_dict seems like a nice way to accomplish what you’re after while also sticking with an existing DataFrame API. Adding this new method would be my personal preference

[MrPowers]

@jrbourbeau - dd.from_dict sounds like a great approach to me. I’ll prep a PR. Thanks!