Yes, as you can run virtually any code on Valohai so it can do almost anything, with varying degrees of required effort. We offer a lot of helpful tooling around data science workloads like data preprocessing, training, data management and the rest of the use-cases mentioned above.
Here are some use-cases that Valohai doesn’t automatically help you with:
Not for: Building Your Actual Model Logic:
Valohai doesn’t offer AutoML capabilities or drag-n-drop interfaces to build predictive models. Valohai users must provide actual program logic in their programming language of choosing like Python, R or C++. Valohai supports all programming languages, frameworks and development tools.
Not for: Interactive Big Data Exploration:
Valohai workers are ephemeral; they download/stream your data, do the instructed work and the runtime environment is destroyed along with the temporary data version. Depending on your data volume, you should use Jupyter Notebooks or something similar to interactively explore your dataset or a slice of it.
Not for: Data Acquisition:
We integrate with all the major cloud-based binary data sources and you should use those to ingest your data. Valohai itself doesn’t provide features to acquire new data samples. After the data is in AWS S3, Azure Store, Google Cloud Store, OpenStack Swift or on a local mount, you can begin using Valohai.
Not for: Data Labeling:
Valohai workers do have an Internet connection, but workers cannot be used to reliably host web servers or other services. This is by design; they are meant to be ephemeral. Theoretically you could host a labeling service on top of Valohai Deployment but none of the tools are built with labeling in mind. There are other good labeling tools available, such as for example Labelbox.