hello all Stephen Darlington: I read https://www.gridgain.com/docs/latest/installation-guide/deployment-modes#choosing-a-client and since I'm developing in Java and I need more than SQL and key-values (I need for sure ML) I opted for thick client
Jeremy McMillan: I'm using linear regression actually. Since I'm pretty sure my scenario will evolve in making forecasts about some events distribution in the future (next N month or year) according to the saved data (the training models) I was studying how to use time series algorithms. DJL seems interesting to me and I wanted to deep dive into it but I need to understand how to fit all pieces together in my scenario where I'm using a thick client. In any case I really thank you all for these very useful tips and discussions Angelo Il giorno ven 5 gen 2024 alle ore 15:51 Jeremy McMillan < jeremy.mcmil...@gridgain.com> ha scritto: > To answer the OP question, maybe linear regression is sufficient for > making predictions in your data. > > Ignite isn't really designed for exploratory data analysis, so it really > helps to understand the character of your data. Linear models are usually a > good place to start. Does a regression line make sense if you plot one on > one of your time value columns in a spreadsheet? If so, you may not need to > deploy any additional libraries. > > > https://ignite.apache.org/docs/latest/machine-learning/regression/linear-regression > > On Fri, Jan 5, 2024, 08:25 Stephen Darlington <sdarling...@apache.org> > wrote: > >> Normally we recommend using thin-clients if you can. Though, in this >> case, using a thick-client makes your life easier. Thick clients can deploy >> Java code for you. >> >> There are a few different ways to do it. The "easy" option is to just >> deploy the JAR files to the server nodes "manually." You could also >> consider peer class loading ( >> https://ignite.apache.org/docs/2.11.1/code-deployment/peer-class-loading), >> which is where the client automatically sends classes to the remote nodes. >> Or UriDeployment ( >> https://ignite.apache.org/docs/2.11.1/code-deployment/deploying-user-code), >> where Ignite copies the Jar files from a central location. GridGain's >> Control Center (not open source) is also able to deploy code. >> >> On Fri, 5 Jan 2024 at 14:04, Angelo Immediata <angelo...@gmail.com> >> wrote: >> >>> hello Gianluca and all >>> >>> Regarding to thin client, in my architecture I avoided to use thin >>> clients; I'm using thick clients; so if python is supported only in "thin >>> client" mode, I'd prefer to avoid it >>> >>> Regarding distributed computing, I didn't see it but it seems to be >>> interesting but something is missing me. Let's suppose I want to use djl >>> https://djl.ai/ and its timeseries support ( >>> https://djl.ai/extensions/timeseries/) I can use the distributed >>> computing; as far as I understood the distributed computing allows to me to >>> distribute computations across all my cluster nodes. Now I'm using thick >>> clients, this means my java application is remotely connected to the apache >>> ignite "master nodes"; in distributed computing I should execute the >>> computation on master nodes but if I use a custom dependency (e.g. djl) how >>> can these master remote nodes execute the computation if they don't have >>> the libraries? >>> Am I missing anything? >>> >>> Thank you >>> Angelo >>> >>> Il giorno ven 5 gen 2024 alle ore 14:24 Gianluca Bonetti < >>> gianluca.bone...@gmail.com> ha scritto: >>> >>>> Hello Jagat >>>> >>>> There are Ignite thin clients for a number of languages, including >>>> Python. >>>> For a full list of functionalities and comparison, please always refer >>>> to the official documentation. >>>> >>>> https://ignite.apache.org/docs/latest/thin-clients/getting-started-with-thin-clients >>>> >>>> All thin clients should perform around the same in tasks such as >>>> storing and retrieving data as they use the Apache Ignite binary protocol. >>>> As you know performance also varies case by case, because of different >>>> setups, configurations, and versions of software/frameworks/libraries >>>> being used, and of course the performance of the code that you will write >>>> yourself. >>>> >>>> For my specific use cases, Apache Ignite always performed extremely >>>> well. >>>> As I don't know anything about your project, there are far too many >>>> possible variables to be able to reduce to a yes/no answer. >>>> The advice is to run your own benchmarks on your infrastructure to get >>>> some meaningful figures for your specific project and infrastructure. >>>> >>>> Cheers >>>> Gianluca Bonetti >>>> >>>> On Fri, 5 Jan 2024 at 12:40, Jagat Singh <jagatsi...@gmail.com> wrote: >>>> >>>>> Hi Gianluna, >>>>> >>>>> Does the Python client miss any functionality or performance compared >>>>> to Java? >>>>> >>>>> Thanks >>>>> >>>>> On Fri, 5 Jan 2024 at 15:55, Gianluca Bonetti < >>>>> gianluca.bone...@gmail.com> wrote: >>>>> >>>>>> Hello Angelo >>>>>> >>>>>> It seems to be an interesting use case for Ignite. >>>>>> >>>>>> However, you should consider what Ignite is, and what is not. >>>>>> Essentially, Ignite is a distributed in-memory >>>>>> database/cache/grid/etc... >>>>>> It also has some distributed computing API capabilities. >>>>>> >>>>>> You can store data easily in Ignite, and consume data by your code >>>>>> written in Java. >>>>>> You can also use Python since there is a Python Ignite Client >>>>>> available if it makes your time series analysis easier. >>>>>> You can also use the Ignite Computing API to execute code on your >>>>>> cluster >>>>>> https://ignite.apache.org/docs/latest/distributed-computing/distributed-computing >>>>>> but in this case I think Python is not supported. >>>>>> >>>>>> Cheers >>>>>> Gianluca Bonetti >>>>>> >>>>>> On Fri, 5 Jan 2024 at 08:52, Angelo Immediata <angelo...@gmail.com> >>>>>> wrote: >>>>>> >>>>>>> I'm pretty new to Apache Ignite >>>>>>> >>>>>>> >>>>>>> I asked this also on stackoverflow ( >>>>>>> https://stackoverflow.com/questions/77667648/apache-ignite-time-series-forecasting) >>>>>>> but I received no answer >>>>>>> >>>>>>> I need to make some forecasting analysis >>>>>>> >>>>>>> Basically I can collect data in Ignite in real time. Ignite will >>>>>>> store data in its own caches >>>>>>> >>>>>>> Now I need to make some forecasting showing me the distribution of >>>>>>> data in the next X months/years by starting from observed and collected >>>>>>> data. >>>>>>> >>>>>>> As far as I know, this kind of forecasting can be realized by time >>>>>>> series forecasting. In Ignite I see no time series based algorithm. Am I >>>>>>> right? >>>>>>> >>>>>>> If I'm correct I may use or tensor flow or Deep Java Library. But in >>>>>>> this case what I don't understand is: where should I use these >>>>>>> libraries? >>>>>>> Inside my thick client microservice or should I write an Ignite plugin >>>>>>> in >>>>>>> order to use the scalability feature provided by Ignite? >>>>>>> >>>>>>> Thank you >>>>>>> >>>>>>> Angelo >>>>>>> >>>>>>