Defining seperate validation dataframe for defining validation image paths and labels

Part 1 (2019)
1

I have a folder of bunch of images and a dataframe that holds the list of training image paths and labels. I have also a similar dataframe for validation.

I am able to define a data source with this code below with splitting the training data by random in two pieces. But I have to define the validation data from an other dataframe.

# This is my lacking data source
src = (
        ImageList.from_df(path=im_path, df=df_filtered)
        .split_by_rand_pct(0.2)
        .label_from_df(cols=['float_label_1','float_label_2'],label_cls=FloatList)
       )

Is it possible to define the data source in this way?

2

This is a two dimensional regression model and I solved my problem above like this;

tfms = get_transforms(flip_vert=True, max_lighting=0.1) # some transformations here

src = (
          ImageList.from_df(path=im_path, df=df_filtered)
          .split_by_idx(list(range(40000,40500))) # select rows from 40000 to 40500 for validation set
          .label_from_df(cols=['float_label_1','float_label_2'], label_cls=FloatList)
         )

data = (
        src.transform(tfms, size=224)
        .databunch(num_workers=0, bs=256) #num_workers =0 is for cpu efficiency in windows
        .normalize(imagenet_stats)
       )