Does H2o Driverless ai latest stable 1.8 version support Image Classification

Question

Can we do Image Classification on own dataset using H2o Driverless-ai latest stable 1.8 version

Answer 1

Yes, the upcoming version 1.9.0 will support both classification and regression using image data. Meanwhile, a workaround for 1.8.x LTS using BYOR (custom recipes) found here is possible, please contact H2O.ai customer support if necessary.

The dataset can be prepared similarly to this example: https://h2o-public-test-data.s3.amazonaws.com/bigdata/laptop/images/demo_disney_data.zip First, copy the data to the VM (using scp or similar command) and then ssh to VM to unzip the file in the ~/data/ directory. Under the ~/data/disney_data/ find disney_data.csv file with the paths to the images and labels: path, label. Finally, load dataset from file using File System option by navigating to disney_data.csv.

Answer 2

Doing experiment with sample data as disney_data.csv, some errors come out below.

• experiment log: <experiment_id>.stack Exception: URL fetch failure on https://github.com/Callidior/keras-applications/releases/download/efficientnet/efficientnet-b3_weights_tf_dim_ordering_tf_kernels_autoaugment_notop.h5: None -- [Errno -2] Name or service not known

• <experiment_id>.stack AssertionError: No best ensemble, so no models to use.

• Expert setting : pipeline building recipe = image_model

Urgently, Want to know when comes out this kinds of error and how I can solve problems.