type object 'MissingDependency' has no attribute 'load'

[debuggedcoder]

So I was trying to load a saved model and Got this error..I am adding my written code and snippet here. This error happened while I was trying things on my local machine

new_document_store = FAISSDocumentStore.load("./faiss.index")
new_retriever = DensePassageRetriever.load("./retriever.pt/", document_store=new_document_store)

Another Error I Got While Working with google colab.. Both of these error is regarding Loading saved files and index.

[bglearning]

Hi,

For the first part

faiss is part of the optional dependencies. So during installation you'll need to specify it also be installed.

pip install farm-haystack[faiss]

Let me know if you had already done it but still get this issue. My guess is you ran a notebook with [faiss] included but this notebook doesn't do it when installing haystack (?)

For the second part

What is probably happening is that the faiss_document_store.db file is missing during your second run.
Solution: You need to save that too into Drive. For this, you can pass the location as FAISSDocumentStore(sql_url="sqlite:///content/drive/MyDrive/Dataset/faiss_document_store.db", ...) when first creating it.

And do the same when creating it again for loading.

Background and more details:

• faiss_document_store.db (the default sqlite location) stores the actual documents
• When we do faiss_document_store.save('index.faiss') it saves the embeddings as index.faiss (and index.json for metadata)

So for FAISSDocumentStore.load(...) to work properly (without mismatch in num-docs and num-embeddings) we need to have all three files. Actually, the two files + the db connection (which by default is saved on disk with sqlite).

One option is to directly initialize instead of calling load and skip check with validate_index_sync=False. So you'd do:

FAISSDocumentStore(faiss_index_path="<your-path>.faiss", 
                   faiss_config_path="<your-path>.json", 
                   validate_index_sync=False)

but I guess having index/embeddings without documents doesn't make sense.

[debuggedcoder]

[debuggedcoder]

Tried when first creating it. Showing The Same Error

[bglearning]

Hi @debuggedcoder
Now, it's the reverse problem. It found the documents but not the embeddings/index.

To rectify you can pass in the paths for the index.faiss and index.json too:

FAISSDocumentStore(sql_url="<folder-path>/faiss_document_store.db", 
                   faiss_index_path="<folder-path>/index.faiss", 
                   faiss_config_path="<folder-path>/index.json", ...) 

[debuggedcoder]

I am really struggling to solve it, can we talk on discord ??

[bglearning]

Okay, will be in the office-hours voice channel