Re: How to import POSIXct variables in a h2o data frame

[Tom Kraljevic]

[ Sending to h2ostream, the open source community forum… ]

Hi,

Convert the column to a Date first using as.Date().

H2O does not understand POSIXct...

Tom

On Dec 9, 2015, at 4:28 AM, Stéphane Tufféry <tuffery....@gmail.com> wrote:

how to import POSIXct variables in a h2o data frame?

Example below:

 

> str(Test2)

'data.frame':   500000 obs. of  3 variables:

$ num_carte   : num  4.98e+15 4.98e+15 4.98e+15 4.98e+15 4.98e+15 ...

$ code_reponse: num  0 0 0 0 0 0 0 0 0 0 ...

$ Date_Ok     : POSIXct, format: "2015-02-11 19:41:26" "2015-02-12 19:30:32" "2015-02-13 12:19:19" "2015-02-20 15:05:25" ...

> Test_h2o <- as.h2o(Test2)

 

ERROR: Unexpected HTTP Status code: 412 Precondition Failed (url = http://localhost:54321/3/Parse)

 

water.exceptions.H2OIllegalArgumentException

[1] "water.parser.ParseSetup.strToColumnTypes(ParseSetup.java:135)"                       

 [2] "water.api.ParseHandler.parse(ParseHandler.java:14)"                                  

 [3] "sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)"                         

 [4] "sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)"       

 [5] "sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)"

[6] "java.lang.reflect.Method.invoke(Method.java:497)"                                    

 [7] "water.api.Handler.handle(Handler.java:64)"                                           

 [8] "water.api.RequestServer.handle(RequestServer.java:644)"                              

 [9] "water.api.RequestServer.serve(RequestServer.java:585)"                               

[10] "water.JettyHTTPD$H2oDefaultServlet.doGeneric(JettyHTTPD.java:617)"                   

[11] "water.JettyHTTPD$H2oDefaultServlet.doPost(JettyHTTPD.java:565)"                      

[12] "javax.servlet.http.HttpServlet.service(HttpServlet.java:755)"                        

[13] "javax.servlet.http.HttpServlet.service(HttpServlet.java:848)"                        

[14] "org.eclipse.jetty.servlet.ServletHolder.handle(ServletHolder.java:684)"              

 

Error in .h2o.doSafeREST(h2oRestApiVersion = h2oRestApiVersion, urlSuffix = page,  :

  Provided column type c("POSIXct", "POSIXt") is unknown.  Cannot proceed with parse due to invalid argument.

>

 

Thanks a lot in advance!

 

Best regards,

 

Stéphane

 

[Stéphane Tufféry]

Hi,

 

Thanks, but I want to use functions such as difftime on date-time variables, not only on date variables.

 

Stéphane

 

 

De : Tom Kraljevic [mailto:to...@h2o.ai]
Envoyé : mercredi 9 décembre 2015 17:43
À : Stéphane Tufféry
Cc : H2O Open Source Scalable Machine Learning - h2ostream
Objet : Re: How to import POSIXct variables in a h2o data frame

[Tom Kraljevic]

H2O stores the value as a millis number.

So, you can do math on seconds if you need to.

I’m not sure the as.Date can handle seconds, but the string form definitely does.

Tom

[Stéphane Tufféry]

Hi Tom,

 

Thank you for your quick reply J

 

So I get something like that:

 

> class(Test$Date_Heure)

[1] "character"

> head(Test$Date_Heure)

[1] "2015-02-11 19:41:26" "2015-02-12 19:30:32" "2015-02-13 12:19:19" "2015-02-20 15:05:25" "2015-02-22 15:36:58"

[6] "2015-02-28 08:22:49"

> Test_h2o <- as.h2o(Test[, c("num_carte", "code_reponse", "Date_Heure")])

> head(Test_h2o)

    num_carte code_reponse          Date_Heure

1 4.97783e+15            0 2015-02-11 19:41:26

2 4.97783e+15            0 2015-02-12 19:30:32

3 4.97783e+15            0 2015-02-13 12:19:19

4 4.97783e+15            0 2015-02-20 15:05:25

5 4.97783e+15            0 2015-02-22 15:36:58

6 4.97783e+15            0 2015-02-28 08:22:49

> 

> Test_h2o$Date_Heure[2]-Test_h2o$Date_Heure[1]

  Date_Heure

1        NaN

 

[1 rows x 1 columns]

> Test_h2o$code_reponse[2]-Test_h2o$code_reponse[1]

  code_reponse

1            0

 

[1 rows x 1 columns]

> 

 

What do you mean by “H2O stores the value as a millis number”?

How can I compute the number of seconds between 2015-02-12 19:30:32 and 2015-02-11 19:41:26?

 

Thanks a lot!

 

Stéphane

 

 

De : Tom Kraljevic [mailto:to...@h2o.ai]
Envoyé : mercredi 9 décembre 2015 22:09

[Tom Kraljevic]

use h2o.importFile("myfile.csv") rather than as.h2o() so the h2o parser can guess the right types.

tom

Sent from my iPhone

[Matthew Landowski]

My hack for this is to convert my date/time objects to character objects.

df$datetime <- as.character(df$datetime)
df.h2o <- as.h2o(df)


Another option, I think someone mentioned, is to write the data frame to a csv file and then upload the csv file.