Grab the NLU training dataset!

[Serge (Serge B)]

Hi.
I just downloaded and training the smalltalk.md and during training I started to get the warnings:

At runtme, I started to get the following warnings:

• 2019-01-22 04:22:54+0000 [-] /usr/local/lib/python3.5/dist-packages/sklearn/preprocessing/label.py:151: builtins.DeprecationWarning: The truth value of an empty array is ambiguous. Returning False, but in future this will result in an error. Use array.size > 0 to check that an array is not empty.

I noticed something related:

Training on demo-rasa.json results in UndefinedMetricWarning

And in the smalltalk.md there are duplicate intents.

## intent:smart
- study
- you should study better
- you must learn
- be clever
- be more clever
- be smarter
- be smart
- get qualified

and ## intent:smart

• you are so intelligent
• you are a genius
• smart
• brilliant
• clever
• you are clever
• you are so brainy
• you’re really smart
• you’re really brainy
• you know a lot
• you know a lot of things
• you have a lot of knowledge
• you know so much
• how smart you are
• how brainy you are
• how clever you are
• how brilliant you are
• you are intelligent
• you are qualified
• you are really smart
• you’re very smart
• you are so smart
• you are too smart
• you are very clever
• you are very intelligent
• you are very smart
• you’re intelligent
• you’re a genius
• you’re a smart cookie
• you’re clever
• you’re pretty smart
• you’re qualified
• why are you so smart
• you are so clever

Same applies for intent:good - it’s there twice and maybe more.

Is it the duplicates that cause this problem? As the confidence level went to near zero after the retrain with this smalltalk.md:

[HeiooDo (Heiko Dotzauer)]

Hi Juste,

are there also some stories available which are related to the new intents?

Ciao HeiooDo

[unit01 (Jack)]

Hey @fade2black. Are there any better alternatives you’d suggest looking at?

[fade2black (Bayram)]

Hi @unit01, alternative to what?

[unit01 (Jack)]

@fade2black An alternative to Rasa, that doesn’t have the downsides you mentioned in your comment (withdrawn now): Grab the NLU training dataset and starter-packs!. What works better in your experience?

[fade2black (Bayram)]

@unit01 I don’t know any alternatives to RASA. I have looked at several others such as Watson but they do not fit my problem either. My problem is creating a closed-domain QA system where creating a chit-chat bot is not a major problem. I have been dealing with a harder problem concerning a QA system.

If RASA does not solve your problem you could create your own chat bot from scratch or create your own layer between incoming data (sentences) and RASA server where you could preprocess the sentence and either take other actions or simply delegate it to RASA.

[Juste]

Hey @HeiooDo. Not at the moment, but it would be a great community contribution if you would decide to generate some conversations!

[Devan0369 (Devan Nair)]

Juste, I am having trouble with running nlu_model.py as the error indicate cannot find rasa_nlu.converters can you explain where or how I can import rasa_nlu.converters from??

[Mayflower]

@Juste, Hi Juste, I am Jinglong, a Software Engineer from China, Hangzhou.In my company, We want to build a chatbot platform for different kinds of customer ,for example,E-commerce( taobao sellers),Education industry,Medical industry and so on.I have already run some demo by rasa stack.

We want to use your rasa platform module,Can you tell me the price and level?

After I sent message to rasa sales team by Contact Us - Rasa . I received one email from Dominik. Then I wrote some more email to Dominik to get information about rasa platform several days ago, but I have not received reply .

My email is jiaojinglong@qipeng.com .

Could you help me about this problems?

Thanks!

[AllBecomesGood (Philipp)]

After retraining with the newly added data my bot forgot how to tell jokes and doesn’t reply much in general anymore >.>

[vishnunahak23 (Vishnu Nahak)]

@ Juste, I am trying to train nlu but its throwing following error can you please help me on this make: *** No rule to make target `train’. Stop.

[AllBecomesGood (Philipp)]

@vishnunahak23

This doesn’t work because Windows doesn’t know “make”. Instead, if you go into the file of this make command “Makefile” then you can find the following commands which can work on windows (but I recommend using anaconda cmd in windows https://www.anaconda.com/)

example of where “make train_nlu” goes looking for what command to execute (in Makefile): train-nlu: python -m rasa_nlu.train -c nlu_config.yml --data data/nlu_data.md -o models --fixed_model_name nlu --project current --verbose

So just run the python commands.

Edit: Actually it could also be cause you execute from the data folder folder. go up a level into \starter-pack-rasa-nlu>

[Aerosoul]

Hi, I just installed Rasa Core, NLU and had a few error messages when linking Spacy, although the command seem to end fine and said I could now link to ‘en’. However, when trying to train rasa_nlu i get this:

python -m rasa_nlu.train -c nlu_config.yml --data data/nlu_data.md -o models --fixed_model_name nlu --project current --verbose

C:\Users\RC\AppData\Local\Continuum\anaconda3\lib\runpy.py:125: RuntimeWarning: ‘rasa_nlu.train’ found in sys.modules after import of package ‘rasa_nlu’, but prior to execution of ‘rasa_nlu.train’; this may result in unpredictable behaviour warn(RuntimeWarning(msg)) 2019-04-26 15:17:54 INFO rasa_nlu.utils.spacy_utils - Trying to load spacy model with name ‘en’ Traceback (most recent call last): File “C:\Users\RC\AppData\Local\Continuum\anaconda3\lib\runpy.py”, line 193, in _run_module_as_main “main”, mod_spec) File “C:\Users\RC\AppData\Local\Continuum\anaconda3\lib\runpy.py”, line 85, in _run_code exec(code, run_globals) File “C:\Users\RC\AppData\Local\Continuum\anaconda3\lib\site-packages\rasa_nlu\train.py”, line 173, in num_threads=cmdline_args.num_threads) File “C:\Users\RC\AppData\Local\Continuum\anaconda3\lib\site-packages\rasa_nlu\train.py”, line 136, in train trainer = Trainer(nlu_config, component_builder) File “C:\Users\RC\AppData\Local\Continuum\anaconda3\lib\site-packages\rasa_nlu\model.py”, line 150, in init self.pipeline = self._build_pipeline(cfg, component_builder) File “C:\Users\RC\AppData\Local\Continuum\anaconda3\lib\site-packages\rasa_nlu\model.py”, line 162, in build_pipeline component = component_builder.create_component(component_cfg, cfg) File “C:\Users\RC\AppData\Local\Continuum\anaconda3\lib\site-packages\rasa_nlu\components.py”, line 448, in create_component component_config, cfg) File “C:\Users\RC\AppData\Local\Continuum\anaconda3\lib\site-packages\rasa_nlu\registry.py”, line 186, in create_component_by_config return component_class.create(component_config, config) File “C:\Users\RC\AppData\Local\Continuum\anaconda3\lib\site-packages\rasa_nlu\utils\spacy_utils.py”, line 64, in create nlp = spacy.load(spacy_model_name, disable=[‘parser’]) File "C:\Users\RC\AppData\Local\Continuum\anaconda3\lib\site-packages\spacy_init.py", line 21, in load return util.load_model(name, **overrides) File “C:\Users\RC\AppData\Local\Continuum\anaconda3\lib\site-packages\spacy\util.py”, line 119, in load_model raise IOError(Errors.E050.format(name=name)) OSError: [E050] Can’t find model ‘en’. It doesn’t seem to be a shortcut link, a Python package or a valid path to a data directory.

Any idea how to resolve this? Thanks a ton!

[Juste]

Hi @Aerosoul. Are you sure the linking was successful? After installing the model, did you get the message the the model is now linked and can be loaded using ‘en’? If you get an error, then on Windows you might need to run your cmd as an administrator for the linking to actually take place

[Juste]

Hey @amt. Good point on the differences between the repos! They should for sure not be different. I would say bot are good ones to start - Sara (rasa demo) is a good ‘best practices’ example where the startepack is something you can take, quickly build and then change it all to build your own bot

[Juste]

Hey @AllBecomesGood. Hahaha, that’s not a good thing Did you add any new stories as well? That could be the case - you need at least a few stories which would make use of the newly added intents.

[AllBecomesGood (Philipp)]

Hello to you, too @Juste ! I haven’t added new stories, that could be it!

I’m trying something different now tho, I wanna build a bot that mainly asks the user questions rather than only reacting to input. I can relax the requirements by sometimes not needing to care what the user said (only that they said anything really) and then ask the next question. It’s also meant to happen in a specific order and all that confuses the heck out of my brain currently lol! I haven’t been through all tutorials yet tho, so hopefully I’ll learn more stuff that’ll help me with that

[Juste]

Sounds awesome! Keep us updated with your experience along the way!

[Nasnl (Nas)]

I’m running into the same thing, and installing make in windows doesn’t provide a solution… it seems. Maybe I’m doing something wrong, but after installing make in windows, I get the same error. Is there something else I should do?

[Aerosoul]

Hi, yes - I am running under Anaconda commandline and had to run under Administrator on Win to get the linking done properly. It said it had linked properly but apparently something failed. When running Anaconda commandline as Admin all works fine. Thanks!