Merge pull request #1 from h2oai/master

Refresh from original

CHANGES.md

@@ -1,3 +1,100 @@
+#Noether Release (2.8.4.1) - 1/30/15
+
+####Enhancements 
+
+These changes are improvements to existing features (which includes changed default values):
+
+#####UI
+
+- Java POJO scoring implemented for unsupervised models (especially DL Autoencoder) [(PUB-1118)](https://0xdata.atlassian.net/browse/PUB-1118)
+- Added perfbar to Cluster Status [(PUB-296)](https://0xdata.atlassian.net/browse/PUB-296)
+- Raised default stack trace depth for Profiler to 10 [(github)](https://github.com/h2oai/h2o/commit/08604c6d9663cbf0c9d955fff137faf68048eb7d)
+- Reduced linewidth to <90 chars. [(github)](https://github.com/h2oai/h2o/commit/b91aad3bcb4db4428d5e5600a3d17040cf7167fcO)
+- Added warning and listed remedies if the input layer is bigger than 100k neurons [(github)](https://github.com/h2oai/h2o/commit/ae78448b9dabb4dd10ad099cdae7abd5d73c65dd)
+- Added license info [(github)](https://github.com/h2oai/h2o/commit/8f51b83f671fa63ef4e2118836a8f20648c0b282)
+
+#####API
+- Added extra options to `CreateFrame` to make (potentially sparse) binary columns [(github)](https://github.com/h2oai/h2o/commit/7e0d6509fde1733fcfe17f92f5c65abc8597fa8a)
+- Ported changes to `createFrame` to R [(github)](https://github.com/h2oai/h2o/commit/f5b2a4d60076dff61414e8dbc246780ade1132ad)
+- Added chunk summary after `CreateFrame` [(github)](https://github.com/h2oai/h2o/commit/6e83e39ca6797e03db97ce925afe6ad71bcba56f)
+- Added `makeModel` call to R/H2O to allow for hand-made glm models with efficient scoring [(github)](https://github.com/h2oai/h2o/commit/4603531579f9e890df1d0fdaf5b57de3719c5e8e)
+- Raised default value of `-data_max_factor_levels` from 65k to 1M [(github)](https://github.com/h2oai/h2o/commit/452392e17a08489103b2063955761702e637d905)
+- Updated `h2o.createFrame` with optional response column [(github)](https://github.com/h2oai/h2o/commit/372bb4a94a1b404328c4c50ad780a480bf49efa8)
+- Added `many_cols` and `chunk_bytes` to h2o.init [(github)](https://github.com/h2oai/h2o/commit/5edb09f8e74544847f063f17a2ec654803aabada)
+- Added `data_max_factor_levels` to h2o.init() in R [(github)](https://github.com/h2oai/h2o/commit/f7a7defc3411d5e16cd8d9ab6690119c5b94a363)
+- Added `-many_cols` (same as `-chunk_bits` 24), renamed `-chunk_bits` to `-chunk_bytes` [(github)](https://github.com/h2oai/h2o/commit/524bef3102e43a2d09ec337f2c6ebf5e75ebab38)
+- Added glm `disable_line_search` flag to R [(github)](https://github.com/h2oai/h2o/commit/e1211d487f15885f76c1bf9a56c1f37180f8a0e7)
+- Added (hidden) explicit `disable_line_search` flag to allow user to disable line search [(github)](https://github.com/h2oai/h2o/commit/7b346d76f2b474656d2c72bf7c37a8d778acbdac)
+- Added `-data_max_factor_levels` and `-chunk_bits` to Hadoop command line [(github)](https://github.com/h2oai/h2o/commit/e0cb1a8fdfd251d8c27a14dfc295ac24af1b7098)
+- Made chunk size an argument. Default is `-chunk_bits` 24 -> 16MB [(github)](https://github.com/h2oai/h2o/commit/014dda94e2e055dff42e99f9e03897da08d51080)
+- Updated parameters for py API for createFrame [(github)](https://github.com/h2oai/h2o/commit/53ab3196983ffae0df0f73679c0bdbe53052c357)
+- Updated create frame arguments [(github)](https://github.com/h2oai/h2o/commit/ddf0da8abddeba5a1da5d71a1490d0deb465f8b8)
+- Added `runifSplit` to Frame [(github)](https://github.com/h2oai/h2o/commit/8a3b7d5cfa02ed32ea454293b79677bb66cda8fa)
+- Added support for `holdout_fraction` to SRF/DRF/GBM/DL in R [(github)](https://github.com/h2oai/h2o/commit/47a491070bd111c1904eca5a0f3dbf00b8f6e890)
+- Backported rbind v2 from h2o-dev [(github)](https://github.com/h2oai/h2o/commit/f0ff6f027e4aa9e3faebcd5f192bd8781d47c5f1)
+- Added `frozenType()` override to speed up histogram deserialization [(github)](https://github.com/h2oai/h2o/commit/b19d95774d008990441d27bcbc88df4be4eb0e79)
+
+#####Algorithms
+
+- Automatically call SplitFrame from the ModelBuilder page [(PUB-1103)](https://0xdata.atlassian.net/browse/PUB-1103)
+- PUB-1103: Added option to automatically call SplitFrame from model builder age if holdout_fraction is > 0 [(github)](https://github.com/h2oai/h2o/commit/e054707a2c627944d8441a6f5562bf2ec216a459)
+- Increased epochs to 30 and increased tolerance to .4 [(github)](https://github.com/h2oai/h2o/commit/8ceb7acf88fbe059f5f6034b297aeb74de9f6454)
+- Enabled variable importance by default [(github)](https://github.com/h2oai/h2o/commit/de5457971a956318c52e6c15b06b95932e3d4d30)
+- Changed GLM printout to avoid errors with modified glm models [(github)](https://github.com/h2oai/h2o/commit/13d4c73887976e4ed2fdc25e249a86e4e9c63f11)
+- Added survfit class designation to h2o.coxph survfit output [(github)](https://github.com/h2oai/h2o/commit/b504526916179aa4265f87767a548d7978ea7ccb)
+- Changed predict method for CoxPH to return the risk as opposed to the cumulative hazard and associated standard error [(github)](https://github.com/h2oai/h2o/commit/fa8af63a02b5ba9a0bbe1df5f1e1e1cea955fcd2)
+- Added N-fold CV, use GBM for MNIST [(github)](https://github.com/h2oai/h2o/commit/60a1957c5f4a0734a1d179b93c77bf5de323fde0)
+- Added grid parallelism parameter to h2o.gbm [(github)](https://github.com/h2oai/h2o/commit/6e9a3577351e606a6da5543580ddca7ff864622c)
+- Updated HDFS deletion command, and run GBM before DRF [(github)](https://github.com/h2oai/h2o/commit/28c499084019bc7566826a2b933c3720091b12df)
+- Added script for running GLM/RF/GBM on 15M rows, 2.2k cols [(github)](https://github.com/h2oai/h2o/commit/b01ea0d5b61d7687b0eae12f15fed3b54689a2b1)
+- Updated R docs for SpeeDRF [(github)](https://github.com/h2oai/h2o/commit/e79c9589efad3d930dc05a69a34aeb4dc4e16db8)
+- Added `class_sampling_factors` to GBM/DRF [(github)](https://github.com/h2oai/h2o/commit/cdd4cad0c748d19874c057f4a9d9aad3bc167588)
+
+#####System 
+
+- Added -extraJavaOpts option to the Hadoop H2O Driver [(github)](https://github.com/h2oai/h2o/commit/77bf6728a791fe4a7ad517ef1a88df9c0a755759)
+- Background the deletion of ICE directory at startup [(github)](https://github.com/h2oai/h2o/commit/814c82217ff3c02cf72523fe7756e7fe1d163cd7)
+- Added Hadoop options (as baseline) [(github)](https://github.com/h2oai/h2o/commit/e8c97e32f7ca1f5e3f6b1c295d1c77275db84e22)
+- Added Hadoop cut and paste example for each distribution [(github)](https://github.com/h2oai/h2o/commit/8e224483df4db29e342abf35fe6f008d0affa2e4)
+- Added R script for Next.ML [(github)](https://github.com/h2oai/h2o/commit/69c1d614d3a37e4428b516f2121d19caa3ded922)
+- Use output of `boot2docker ip` instead of localhost [(github)](https://github.com/h2oai/h2o/commit/9687cd0b443e4ac23a772535bed2bd16e9bf25b7)
+
+
+####Bug Fixes
+
+These changes are to resolve incorrect software behavior: 
+
+#####UI
+- Fixed a bug in an example due to createFrame change [(github)](https://github.com/h2oai/h2o/commit/afc93a9734ea22efa9afcec46bbc7e26b2faee98)
+- Import file hangs at 99% for past 45 mins [(HEX-2003)](https://0xdata.atlassian.net/browse/HEX-2003)
+
+#####API
+- as.h2o returns NA when character field has many unique values [(PUB-1108)](https://0xdata.atlassian.net/browse/PUB-1108)
+- Fixed R for createFrame updates [(github)](https://github.com/h2oai/h2o/commit/f52f6a4679a2f5cacc0825c271f3c899cca43c03)
+
+#####Algorithms
+- Removed restrictions on glm upper/lower bounds to be non-negative/non-positive. The only restriction is lb <= ub [(github)](https://github.com/h2oai/h2o/commit/5980d62c436922e6c3c73cbfa0b83c1bcebff7d8)
+- Fixed PUB-1118 - AutoEncoder POJO scoring. Also fixed a bug in initializing the activation layer for categoricals. [(github)](https://github.com/h2oai/h2o/commit/6505a1ffc0a99c211a6e14807ff280058ab9c0e7)
+- Variable Importance for DRF slow for large numbers of columns [(PUB-1101)](https://0xdata.atlassian.net/browse/PUB-1101)
+- Fix PUB-1101 by using GBM's variable importance method for DRF [(github)](https://github.com/h2oai/h2o/commit/0026200d4ba1e76f759cb4df5743555855b92c42)
+- Bugfix for case where N is not divisible by num CV folds [(github)](https://github.com/h2oai/h2o/commit/34e271760b70fe6f384e106d84f18c7f0adb8210)
+- Fixed typo in parameter validation [(github)](https://github.com/h2oai/h2o/commit/f6c6996594894c5ecdcd2ad112953b92844a3848)
+- Don't shuffle intra-chunk after stratified sampling [(github)](https://github.com/h2oai/h2o/commit/f0ff6f027e4aa9e3faebcd5f192bd8781d47c5f1)
+- Fixed intercept adjustment by prior probability in logistic regression [(github)](https://github.com/h2oai/h2o/commit/cdd8a6b54a35e83a5f9aebb419ada476593ab144)
+- DL: DL build model running for past 12 hrs doesn't complete [(HEX-1997)](https://0xdata.atlassian.net/browse/HEX-1997)
+- GLM fails simpleCheckGLM in multinode when cross-validation is on [(HEX-1867)](https://0xdata.atlassian.net/browse/HEX-1867)
+
+#####System
+
+- Parse creates too many keys for large numbers of columns [(PUB-1102)](https://0xdata.atlassian.net/browse/PUB-1102)
+- Don't allow pure number enums [(github)](https://github.com/h2oai/h2o/commit/fb95e198db3a6186ef5ed408bd31f29dadae60a5)
+- Fixed argument order [(github)](https://github.com/h2oai/h2o/commit/2a174063ca5ffdc4bd117a7206e45d89fa56317b)
+- Excluded factors that contain lower-case UUIDs that look like they are in scientific notation with exponent: `^\d+e\d+$`[(github)](https://github.com/h2oai/h2o/commit/7fc6f1c23f254156a60ce343df4ca6b5b73f3156)
+- Prevented deadlock during startup by adding UniqueFrameId to TypeMap [(github)](https://github.com/h2oai/h2o/commit/7e6fb375cd1c45c5a771b79add73f697409a312a)
+- Fixed sparse chunk to avoid binary search for prefix/suffix zeros [(github)](https://github.com/h2oai/h2o/commit/48d4dfa9ad12461970c7d3c2f75907a27cb7d6f0)
+
+
+
 Release 2.6.1.1 (Lambert build 1)
 ==================================
 

Makefile

@@ -289,6 +289,7 @@ dw_3:
 	mkdir -p $(BUILD_WEBSITE_DIR)/bits/hadoop
 	cp -p hadoop/README.txt $(BUILD_WEBSITE_DIR)/bits/hadoop
 	cp -p docs/H2O_on_Hadoop_0xdata.pdf $(BUILD_WEBSITE_DIR)/bits/hadoop
+	cp -p docs/sparkling_water_meetup.pdf $(BUILD_WEBSITE_DIR)/bits
 	cp -p docs/h2o_datasheet.pdf $(BUILD_WEBSITE_DIR)/bits
 	cp -p docs/H2ODeveloperCookbook.pdf $(BUILD_WEBSITE_DIR)/bits
 	mkdir -p $(BUILD_WEBSITE_DIR)/bits/ec2
@@ -300,6 +301,7 @@ dw_3:
 	sed -i -e "s/SUBST_RELEASE_NAME/$(RELEASE_NAME)/g; s/SUBST_PROJECT_VERSION/$(PROJECT_VERSION)/g; s/SUBST_BUILD_NUMBER/$(BUILD_NUMBER)/g" $(BUILD_WEBSITE_DIR)/deployment/hadoop_tutorial.html
 	sed -i -e "s/SUBST_RELEASE_NAME/$(RELEASE_NAME)/g; s/SUBST_PROJECT_VERSION/$(PROJECT_VERSION)/g; s/SUBST_BUILD_NUMBER/$(BUILD_NUMBER)/g" $(BUILD_WEBSITE_DIR)/Ruser/Rinstall.html
 	sed -i -e "s/SUBST_RELEASE_NAME/$(RELEASE_NAME)/g; s/SUBST_PROJECT_VERSION/$(PROJECT_VERSION)/g; s/SUBST_BUILD_NUMBER/$(BUILD_NUMBER)/g" $(BUILD_WEBSITE_DIR)/deployment/ec2_build_ami.html
+	sed -i -e "s/SUBST_RELEASE_NAME/$(RELEASE_NAME)/g; s/SUBST_PROJECT_VERSION/$(PROJECT_VERSION)/g; s/SUBST_BUILD_NUMBER/$(BUILD_NUMBER)/g" $(BUILD_WEBSITE_DIR)/deployment/H2O_Hadoop_Mapr.html
 
 #
 # Set appropriately for your data size to quickly try out H2O.

R/.gitignore

@@ -4,6 +4,7 @@ h2o-package/man/h2o-package.Rd
 h2o-package/h2o_package.pdf
 h2o-package/inst/java
 h2o-package/inst/branch.txt
+h2o-package/inst/buildnum.txt
 h2o-package/inst/source_code_repository_info.txt
 tests/.RData
 

R/H2OTestDemo.R

@@ -8,7 +8,7 @@ localH2O = new("H2OClient", ip = "localhost", port = 54321)
 # h2o.checkClient(localH2O)
 
 # Test using prostate cancer data set
-prostate.hex = h2o.importURL(localH2O, path = "https://raw.github.com/0xdata/h2o/master/smalldata/logreg/prostate.csv", key = "prostate.hex")
+prostate.hex = h2o.importURL(localH2O, path = "https://raw.github.com/h2oai/h2o/master/smalldata/logreg/prostate.csv", key = "prostate.hex")
 prostate.sum = summary(prostate.hex)
 print(prostate.sum)
 prostate.glm = h2o.glm(y = "CAPSULE", x = c("AGE","RACE","PSA","DCAPS"), data = prostate.hex, family = "binomial", nfolds = 10, alpha = 0.5)
@@ -38,13 +38,13 @@ for(i in 1:length(glm_test.hex))
 
 
 #Test of GLMGrid using prostate cancer data set
-prostate.hex = h2o.importURL(localH2O, path = "https://raw.github.com/0xdata/h2o/master/smalldata/logreg/prostate.csv", key = "prostate.hex")
+prostate.hex = h2o.importURL(localH2O, path = "https://raw.github.com/h2oai/h2o/master/smalldata/logreg/prostate.csv", key = "prostate.hex")
 prostate.sum = summary(prostate.hex)
 prostate.glmgrid = h2o.glmgrid(y = "CAPSULE", x = c("AGE","RACE","PSA","DCAPS"), data = prostate.hex, family = "binomial", nfolds = 10, alpha = c(0.2,0.5,1),lambda=c(1e-4,1))
 print(prostate.glmgrid)
 
 # Test of PCA using prostate cancer data set
-prostate.hex = h2o.importURL(localH2O, path = "https://raw.github.com/0xdata/h2o/master/smalldata/logreg/prostate.csv", key = "prostate.hex")
+prostate.hex = h2o.importURL(localH2O, path = "https://raw.github.com/h2oai/h2o/master/smalldata/logreg/prostate.csv", key = "prostate.hex")
 prostate.pca = h2o.prcomp(prostate.hex)
 print(prostate.pca)
 summary(prostate.pca)

R/H2O_Load.R

@@ -1,5 +1,5 @@
 # Change this global variable to match your own system's path
-ROOT.PATH <- "/Users/anqi_fu/Documents/workspace/h2o/R/h2o-package/R/"
+ROOT.PATH <- "/Users/spencer/master/h2o/R/h2o-package/R/"
 src <-
 function() {
   warning("MAY NOT WORK ON YOUR SYSTEM -- **TRY TO CHANGE `ROOT.PATH`!**")

R/README.txt

@@ -1,108 +1,36 @@
 H2O in R
-------------
-
+--------
 
 These instructions assume you are using R 2.13.0 or later.  
 
-**STEP 1**
-
-The download package can be obtained by clicking on the button "Download H2O" at http://0xdata.com/h2o <http://0xdata.com/h2o.
-
-Unzip the downloaded h2o zip file
-
-
-**STEP 2: Console Users and Studio Users should follow the same steps: **
-
-In the R console install the package by 
-
-1. Visiting http://0xdata.com/downloadtable/
-2. Choosing the version of H2O appropriate for their environment 
-3. Copy and pasting the R command shown below the downloadable zip file on the download page for the version of their choice into their R console. 
-
-
-Correctly following the above steps will return output similar to the following: 
-
-trying URL 'http://s3.amazonaws.com/h2o-release/h2o/master/1247/R/bin/macosx/contrib/3.0/h2o_2.3.0.1247.tgz'
-Content type 'application/x-tar; charset=binary' length 36702378 bytes (35.0 Mb)
-opened URL
-==================================================
-downloaded 35.0 Mb
-
-
-**STEP 3**
-
-Start an instance of H2O. If you have questions about how to do this see the notes provided at the bottom of the page for starting from a zip file.
-
-If users choose to not start an instance of H2O prior to attempting to connect to H2O through R, an instance will be started automatically for them at ip: localhost, port: 54321.
-
-*Users should be aware that in order for H2O to successfully run through R, an instance of H2O must also simultaneously be running. If the instance of H2O is stopped, the R program will no longer run, and work done will be lost.* 
-
-
-**STEP 4** 
-
-call the H2O package in the R environment, start the connection between R and H2O at ip: localhost and port: 54321
-
-
-  >library(h2o)
-
-  >localH2O = h2o.init()
-
-
 
-**STEP 6** 
+**STEP 1: Unzip**
 
-Here is an example of using the above object in an H2O call in R
+If you are reading this, you probably already downloaded an h2o zip file and unzipped it.
+(Note:  Obtain a zip file from the Download button at http://h2o.ai)
 
 
+**STEP 2: Install the H2O R Package**
 
-  >irisPath = system.file("extdata", "iris.csv", package="h2o")
-  >iris.hex = h2o.importFile(localH2O, path = irisPath, key = "iris.hex")
-  >summary(iris.hex)
+In a terminal window, type:
 
+R CMD INSTALL h2o*.gz
 
 
+**STEP 3: Start a connection to H2O**
 
+Load the H2O package in the R environment.  Start the connection between R and H2O (with defaults at ip: localhost and port: 54321).
+Look at the help for h2o.init() for additional information about how to start and connect to H2O.
 
-Getting started from a zip file
--------------------------------
+  > library(h2o)
+  > localH2O = h2o.init()
 
 
-1. Download the latest release of H2O as a .zip file from the H2O website http://0xdata.com/h2O/.
+**STEP 4: Drive H2O from R** 
 
-2. From your terminal change your working directory to the same directory where your .zip file is saved.
-
-3. From your terminal, unzip the .zip file.  For example:
-
-
-  unzip h2o-1.7.0.520.zip
-
-4. At the prompt enter the following commands.  (Choose a unique name (use the -name option) for yourself if other people might be running H2O in your network.)
-
-
-  cd h2o-1.7.0.520 
-  java -Xmx1g -jar h2o.jar -name mystats-cloud
-
-5. Wait a few moments and the output similar to the following will appear in your terminal window:
-
-
-
-  03:05:45.311 main      INFO WATER: ----- H2O started -----
-  03:05:45.312 main      INFO WATER: Build git branch: master
-  03:05:45.312 main      INFO WATER: Build git hash: f253798433c109b19acd14cb973b45f255c59f3f
-  03:05:45.312 main      INFO WATER: Build git describe: f253798
-  03:05:45.312 main      INFO WATER: Build project version: 1.7.0.520
-  03:05:45.313 main      INFO WATER: Built by: 'jenkins'
-  03:05:45.313 main      INFO WATER: Built on: 'Thu Sep 12 00:01:52 PDT 2013'
-  03:05:45.313 main      INFO WATER: Java availableProcessors: 8
-  03:05:45.321 main      INFO WATER: Java heap totalMemory: 0.08 gb
-  03:05:45.321 main      INFO WATER: Java heap maxMemory: 0.99 gb
-  03:05:45.322 main      INFO WATER: ICE root: '/tmp/h2o-tomk'
-  03:05:45.364 main      INFO WATER: Internal communication uses port: 54322
-  +                                  Listening for HTTP and REST traffic on  http://192.168.1.52:54321/
-  03:05:45.409 main      INFO WATER: H2O cloud name: 'mystats-cloud'
-  03:05:45.409 main      INFO WATER: (v1.7.0.520) 'mystats-cloud' on /192.168.1.52:54321, discovery address /236.151.114.91:60567
-  03:05:45.411 main      INFO WATER: Cloud of size 1 formed [/192.168.1.52:54321]
-  03:05:45.543 main      INFO WATER: Log dir: '/tmp/h2o-tomk/h2ologs'
+  > irisPath = system.file("extdata", "iris.csv", package="h2o")
+  > iris.hex = h2o.importFile(localH2O, irisPath)
+  > summary(iris.hex)
 
 
 
@@ -126,16 +54,3 @@ memory).  For best performance, Xmx should be 4x the size of your
 data, but never more than the total amount of memory on your
 computer. For larger data sets, running on a server or service 
 with more memory available for computing is recommended. 
-
-
-
-
-
-
-
-
-
-
-
-
-

R/ensemble/README.md

@@ -7,13 +7,13 @@ The `h2oEnsemble` R package provides functionality to create ensembles from the
 The `h2oEnsemble` package can be installed using either of the following methods.
 - Clone the main h2o repository and install the package:
 ```
-git clone https://github.com/0xdata/h2o.git
+git clone https://github.com/h2oai/h2o.git
 R CMD INSTALL h2o/R/ensemble/h2oEnsemble-package
 ```
 - Install in R using `devtools::install_github`:
 ```
 library(devtools)
-install_github("0xdata/h2o/R/ensemble/h2oEnsemble-package")
+install_github("h2oai/h2o/R/ensemble/h2oEnsemble-package")
 ```
 
 ## Create Ensembles

R/ensemble/h2oEnsemble-package/DESCRIPTION

@@ -1,14 +1,14 @@
 Package: h2oEnsemble
 Type: Package
 Title: H2O Ensemble Learning
-Version: 0.0.2
-Date: 2014-11-17
+Version: 0.0.3
+Date: 2014-12-28
 Author: Erin LeDell
 Maintainer: Erin LeDell <[email protected]>
-Description: This package implements the super learner ensemble (stacking) algorithm using the H2O R interface to provide base leanrning algorithms.   
+Description: This package implements the super learner ensemble (stacking) algorithm using the H2O R interface to provide base learning algorithms.   
 License: Apache License (== 2.0)
 Depends: R (>= 2.13.0), h2o
 NeedsCompilation: no
 SystemRequirements: Java (>= 1.6)
 Suggests: SuperLearner, cvAUC
-URL: http://www.0xdata.com
+URL: https://github.com/h2oai/h2o/tree/master/R/ensemble/h2oEnsemble-package

R/ensemble/h2oEnsemble-package/NEWS

@@ -1,6 +1,11 @@
 # News for the h2oEnsemble package. #
 
 
+h2oEnsemble 0.0.3 (2014-12-28)
+----------------
+* Fixed bug in `h2o.ensemble` that occurred when the number of observations is not divisible by the number of internal CV folds.
+
+
 h2oEnsemble 0.0.2 (2014-11-17)
 ----------------
 * Removed rounding in `h2o.ensemble.predict`.