loudml-grafana-app
loudml-grafana-app copied to clipboard
vsergeyev
Loud ML Graph Panel Unsigned on grafana 7.3 and problems with Loud ML graph panel
Also I get an error data_1.FieldColorMode is undefined all the time if I use Loud ML Graph Panel My config.yml
List buckets (TSDB data stores) here.
An empty list is the preferred option if you want to populate this list
at run time using the REST APIs.
Uncomment the next line to use an empty bucket list.
buckets: []
Another option is to define static buckets that will immediately become
visible when the Loud ML server starts.
Uncomment the next lines and fine tune the parameters based on your
specific TSDB settings in order to define static buckets.
buckets:
Output bucket
- name: loudml type: influxdb addr: localhost:8086 database: output measurement: loudml retention_policy: autogen create_database: true max_series_per_request: 2000 annotation_db: loudmlannotations
Input bucket
- name: influxdb1 type: influxdb addr: localhost:8086 database: Nueva_Joya measurement: loudml
create_database: true
retention_policy: autogen
- name: elastic
type: elasticsearch
addr: localhost:9200
index: myindex
doc_type: doc
max_series_per_request: 2000
storage defines where Loud ML will save trained model
information.
storage.path: /var/lib/loudml
server defines the TCP host and port address that the
Loud ML server will listen to.
listen: Use 0.0.0.0:8077 to listen to all IP address available
on the host. This setting should be set to localhost:8077 if you
are using a reverse proxy eg nginx to proxy incoming requests to Loud ML.
workers: sets the number of worker process. Use default for CPU
hardwares. Use num_cpu_cores * 4 * num_gpus for GPU configurations.
maxtasksperchild: sets how many tasks a worker process is allowed to do
before being replaced.
jobs_max_ttl: sets how long a job result will remain available
in GET /jobs/ when the job is done. Unit in seconds.
server: listen: 0.0.0.0:8077
workers: 16
maxtasksperchild: 100
jobs_max_ttl: 60
inference defines the TensorFlow cores used to predict
output data from trained models.
inference: num_cpus: 1
num_gpus: 0
training defines the TensorFlow cores used to train new models.
The minimum number for num_cpus is one.
Fine tune these settings according to your hardware configuration.
GPUs offload compute intensive tasks. One GPU typically provides 4x the
compute capacity of a regular CPU.
training: num_cpus: 1
num_gpus: 0
scheduled_jobs automate regular training and inference tasks.
They use standard REST APIs. Refer to the API documentation
for more information.
scheduled_jobs hacks #1:
Uncomment the following lines to perform a one day forecast
and update this forecast every ten minutes:
#scheduled_jobs:
- name: "forecast(test-model) every five minutes"
relative_url: "/models/test-model/_forecast"
method: post
params:
from: "now"
to: "now+1d"
every:
count: 5
unit: minutes
scheduled_jobs hacks #2:
Uncomment the following lines to perform a one day forecast
and update this forecast every ten minutes for all models.
Note the {{model_name}} placeholder:
#scheduled_jobs:
- name: "forecast({{model_name}}) every five minutes"
relative_url: "/models/{{model_name}}/_forecast"
method: post
params:
from: "now"
to: "now+1d"
every:
count: 5
unit: minutes
scheduled_jobs hacks #3:
Uncomment the following lines to update all saved models and train
each model every night, every Sunday, or every 28 days.
Note the time ranges used to fetch data points in each job:
#scheduled_jobs:
- name: "train({{model_name}}) every day at 1am"
relative_url: "/models/{{model_name}}/_train"
method: post
params:
from: "now-1d"
to: "now"
continue: True
every:
count: 1
unit: day
at: "01:00"
- name: "train({{model_name}}) every sunday at 2am"
relative_url: "/models/{{model_name}}/_train"
method: post
params:
from: "now-7d"
to: "now"
every:
count: 1
unit: sunday
at: "02:00"
- name: "train({{model_name}}) every 28 days at 3am"
relative_url: "/models/{{model_name}}/_train"
method: post
params:
from: "now-28d"
to: "now"
every:
count: 28
unit: days
at: "03:00"
#metrics:
enable: True
To reproduce the problem, the environment is based on Debian 9, Grafana 7.3.4
Hello @adelvalle62 , @maniac0r
Will take a look at this most likely on this weekend.
Thank you for using app and a lot of kudos for you to reporting bugs!
V.
Hello;
When I compile the code I get an error on GraphPanel.tsx, and then I change the line 8: import { LegendDisplayMode } from '@grafana/ui'; //src/components/Legend/Legend';
After that I get the next error in the panel at Grafana 7.3.6:
Cannot read property 'Fixed' of undefined
I don't know if it will help anyone, but I had some issues with running the plugin in grafana. I tried 6.x, 7.x in Docker, but finally managed to install plugin in latest Grafana running local (multiple reinstallations). What I managed to run this plugin. First of all, you you LoudML pluging just to create, train and deploy model. I create empty panel with graph, select my influxdb fields and select the time range. Thank I click "Stat" and than LoudML. That is only way for me to see data in LoudML plugin. Next step is to create baseline, train model and start (without saving or exiting). After it is done, I save the panel only to be able to retrain model. To visualise data, I create new panel with graph and mixed sourced queries. 1st is used for input data db and second is output bucket from LoudML.
I am struggling with forecast, as my output is always in late compared to live data, therefore not a forecast.
Seems like the Fixed issue is because they change the API in 7.2 for FieldColorMode: https://grafana.com/docs/grafana/v7.2/packages_api/data/fieldcolormode/
EDIT: Yes, they renamed that enum to FieldColorModeId: https://grafana.com/docs/grafana/v7.2/packages_api/data/fieldcolormodeid/
