FATE Test¶

A collection of useful tools to running FATE’s test.

quick start¶

(optional) create virtual env

python -m venv venv
source venv/bin/activate
pip install -U pip

install fate_test
```
pip install fate_test
fate_test --help
```

edit default fate_test_config.yaml

# edit priority config file with system default editor
# filling some field according to comments
fate_test config edit

configure FATE-Pipeline and FATE-Flow Commandline server setting

# configure FATE-Pipeline server setting
pipeline init --port 9380 --ip 127.0.0.1
# configure FATE-Flow Commandline server setting
flow init --port 9380 --ip 127.0.0.1

run some fate_test suite

fate_test suite -i <path contains *testsuite.json>

run some fate_test benchmark

fate_test benchmark-quality -i <path contains *benchmark.json>

useful logs or exception will be saved to logs dir with namespace shown in last step

develop install¶

It is more convenient to use the editable mode during development: replace step 2 with flowing steps

pip install -e ${FATE}/python/fate_client && pip install -e ${FATE}/python/fate_test

command types¶

suite: used for running testsuites, collection of FATE jobs
```
fate_test suite -i <path contains *testsuite.json>
```
benchmark-quality used for comparing modeling quality between FATE and other machine learning systems
```
fate_test benchmark-quality -i <path contains *benchmark.json>
```

configuration by examples¶

no need ssh tunnel:

9999, service: service_a
10000, service: service_b

and both service_a, service_b can be requested directly:

work_mode: 1 # 0 for standalone, 1 for cluster
data_base_dir: <path_to_data>
parties:
  guest: [10000]
  host: [9999, 10000]
  arbiter: [9999]
services:
  - flow_services:
    - {address: service_a, parties: [9999]}
    - {address: service_b, parties: [10000]}

need ssh tunnel:

9999, service: service_a
10000, service: service_b

service_a, can be requested directly while service_b don’t, but you can request service_b in other node, say B:

work_mode: 0 # 0 for standalone, 1 for cluster
data_base_dir: <path_to_data>
parties:
  guest: [10000]
  host: [9999, 10000]
  arbiter: [9999]
services:
  - flow_services:
    - {address: service_a, parties: [9999]}
  - flow_services:
    - {address: service_b, parties: [10000]}
    ssh_tunnel: # optional
    enable: true
    ssh_address: <ssh_ip_to_B>:<ssh_port_to_B>
    ssh_username: <ssh_username_to B>
    ssh_password: # optional
    ssh_priv_key: "~/.ssh/id_rsa"

Testsuite¶

Testsuite is used for running a collection of jobs in sequence. Data used for jobs could be uploaded before jobs are submitted, and are cleaned when jobs finished. This tool is useful for FATE’s release test.

command options¶

fate_test suite --help

include:

fate_test suite -i <path1 contains *testsuite.json>

will run testsuites in path1

exclude:

fate_test suite -i <path1 contains *testsuite.json> -e <path2 to exclude> -e <path3 to exclude> ...

will run testsuites in path1 but not in path2 and path3

glob:

fate_test suite -i <path1 contains *testsuite.json> -g "hetero*"

will run testsuites in sub directory start with hetero of path1

replace:
```
fate_test suite -i <path1 contains *testsuite.json> -r '{"maxIter": 5}'
```
will find all key-value pair with key “maxIter” in data conf or conf or dsl and replace the value with 5
skip-data:
```
fate_test suite -i <path1 contains *testsuite.json> --skip-data
```
will run testsuites in path1 without uploading data specified in benchmark.json.
yes:
```
fate_test suite -i <path1 contains *testsuite.json> --yes
```
will run testsuites in path1 directly, skipping double check
skip-dsl-jobs:
```
fate_test suite -i <path1 contains *testsuite.json> --skip-dsl-jobs
```
will run testsuites in path1 but skip all tasks in testsuites. It’s would be useful when only pipeline tasks needed.
skip-pipeline-jobs:
```
fate_test suite -i <path1 contains *testsuite.json> --skip-pipeline-jobs
```
will run testsuites in path1 but skip all pipeline tasks in testsuites. It’s would be useful when only dsl tasks needed.

Benchmark Quality¶

Benchmark-quality is used for comparing modeling quality between FATE and other machine learning systems. Benchmark produces a metrics comparison summary for each benchmark job group.

fate_test benchmark-quality -i examples/benchmark_quality/hetero_linear_regression

+-------+--------------------------------------------------------------+
|  Data |                             Name                             |
+-------+--------------------------------------------------------------+
| train | {'guest': 'motor_hetero_guest', 'host': 'motor_hetero_host'} |
|  test | {'guest': 'motor_hetero_guest', 'host': 'motor_hetero_host'} |
+-------+--------------------------------------------------------------+
+------------------------------------+--------------------+--------------------+-------------------------+---------------------+
|             Model Name             | explained_variance |      r2_score      | root_mean_squared_error |  mean_squared_error |
+------------------------------------+--------------------+--------------------+-------------------------+---------------------+
| local-linear_regression-regression | 0.9035168452250094 | 0.9035070863155368 |   0.31340413289880553   | 0.09822215051805216 |
| FATE-linear_regression-regression  | 0.903146386539082  | 0.9031411831961411 |    0.3139977881119483   | 0.09859461093919596 |
+------------------------------------+--------------------+--------------------+-------------------------+---------------------+
+-------------------------+-----------+
|          Metric         | All Match |
+-------------------------+-----------+
|    explained_variance   |    True   |
|         r2_score        |    True   |
| root_mean_squared_error |    True   |
|    mean_squared_error   |    True   |
+-------------------------+-----------+

command options¶

use the following command to show help message

fate_test benchmark-quality --help

include:

fate_test benchmark-quality -i <path1 contains *benchmark.json>

will run benchmark testsuites in path1

exclude:

fate_test benchmark-quality -i <path1 contains *benchmark.json> -e <path2 to exclude> -e <path3 to exclude> ...

will run benchmark testsuites in path1 but not in path2 and path3

glob:

fate_test benchmark-quality -i <path1 contains *benchmark.json> -g "hetero*"

will run benchmark testsuites in sub directory start with hetero of path1

tol:
```
fate_test benchmark-quality -i <path1 contains *benchmark.json> -t 1e-3
```
will run benchmark testsuites in path1 with absolute tolerance of difference between metrics set to 0.001. If absolute difference between metrics is smaller than tol, then metrics are considered almost equal. Check benchmark testsuite writing guide on setting alternative tolerance.
skip-data:
```
fate_test benchmark-quality -i <path1 contains *benchmark.json> --skip-data
```
will run benchmark testsuites in path1 without uploading data specified in benchmark.json.
yes:
```
fate_test benchmark-quality -i <path1 contains *benchmark.json> --yes
```
will run benchmark testsuites in path1 directly, skipping double check

benchmark testsuite¶

Configuration of jobs should be specified in a benchmark testsuite whose file name ends with “*benchmark.json”. For benchmark testsuite example, please refer here.

A benchmark testsuite includes the following elements:

data: list of local data to be uploaded before running FATE jobs
- file: path to original data file to be uploaded, should be relative to testsuite or FATE installation path
- head: whether file includes header
- partition: number of partition for data storage
- table_name: table name in storage
- namespace: table namespace in storage
- role: which role to upload the data, as specified in fate_test.config; naming format is: “{role_type}_{role_index}”, index starts at 0
```
"data": [
    {
        "file": "examples/data/motor_hetero_host.csv",
        "head": 1,
        "partition": 8,
        "table_name": "motor_hetero_host",
        "namespace": "experiment",
        "role": "host_0"
    }
]
```
job group: each group includes arbitrary number of jobs with paths to corresponding script and configuration
- job: name of job to be run, must be unique within each group list
  - script: path to testing script, should be relative to testsuite
  - conf: path to job configuration file for script, should be relative to testsuite
```
"local": {
     "script": "./local-linr.py",
     "conf": "./linr_config.yaml"
}
```
- compare_setting: additional setting for quality metrics comparison, currently only takes relative_tol
  
  If metrics a and b satisfy abs(a-b) <= max(relative_tol * max(abs(a), abs(b)), absolute_tol) (from math module), they are considered almost equal. In the below example, metrics from “local” and “FATE” jobs are considered almost equal if their relative difference is smaller than 0.05 * max(abs(local_metric), abs(pipeline_metric).
```
"linear_regression-regression": {
    "local": {
        "script": "./local-linr.py",
        "conf": "./linr_config.yaml"
    },
    "FATE": {
        "script": "./fate-linr.py",
        "conf": "./linr_config.yaml"
    },
    "compare_setting": {
        "relative_tol": 0.01
    }
}
```

testing script¶

All job scripts need to have Main function as an entry point for executing jobs; scripts should return two dictionaries: first with data information key-value pairs: {data_type}: {data_name_dictionary}; the second contains {metric_name}: {metric_value} key-value pairs for metric comparison.

By default, the final data summary shows the output from the job named “FATE”; if no such job exists, data information returned by the first job is shown. For clear presentation, we suggest that user follow this general guideline for data set naming. In the case of multi-host task, consider numbering host as such:

{'guest': 'default_credit_homo_guest',
 'host_1': 'default_credit_homo_host_1',
 'host_2': 'default_credit_homo_host_2'}

Returned quality metrics of the same key are to be compared. Note that only real-value metrics can be compared.

FATE script: Main should have three inputs:
- config: job configuration, JobConfig object loaded from “fate_test_config.yaml”
- param: job parameter setting, dictionary loaded from “conf” file specified in benchmark testsuite
- namespace: namespace suffix, user-given namespace or generated timestamp string when using namespace-mangling
non-FATE script: Main should have one or two inputs:
- param: job parameter setting, dictionary loaded from “conf” file specified in benchmark testsuite
- (optional) config: job configuration, JobConfig object loaded from “fate_test_config.yaml”

Note that Main in FATE & non-FATE scripts can also be set to take zero input argument.

data¶

Data sub-command is used for upload or delete dataset in suite’s.

command options¶

fate_test data --help

include:

fate_test data [upload|delete] -i <path1 contains *testsuite.json>

will upload/delete dataset in testsuites in path1

exclude:

fate_test data [upload|delete] -i <path1 contains *testsuite.json> -e <path2 to exclude> -e <path3 to exclude> ...

will upload/delete dataset in testsuites in path1 but not in path2 and path3

glob:
```
fate_test data [upload|delete] -i <path1 contains *testsuite.json> -g "hetero*"
```
will upload/delete dataset in testsuites in sub directory start with hetero of path1

full command options¶

fate_test¶

A collection of useful tools to running FATE’s test.

fate_test [OPTIONS] COMMAND [ARGS]...

Options

-b, --backend <backend>¶: Manual specify backend, 0 for eggroll, 1 for spark

-w, --work-mode <work_mode>¶: Manual specify work mode, 0 for local, 1 for cluster

-y, --yes¶: Skip double check

-nm, --namespace-mangling¶: Mangling data namespace

-n, --namespace <namespace>¶: Manual specify fate_test namespace

-c, --config <config>¶: Manual specify config file

benchmark-quality¶

process benchmark suite, alias: bq

fate_test benchmark-quality [OPTIONS]

Options

-i, --include <include>¶: Required include *benchmark.json under these paths

-e, --exclude <exclude>¶: exclude *benchmark.json under these paths

-g, --glob <glob>¶: glob string to filter sub-directory of path specified by <include>

-t, --tol <tol>¶: tolerance (absolute error) for metrics to be considered almost equal. Comparison is done by evaluating abs(a-b) <= max(relative_tol * max(abs(a), abs(b)), absolute_tol)

--skip-data¶: skip uploading data specified in benchmark conf

config¶

fate_test config

fate_test config [OPTIONS] COMMAND [ARGS]...

check¶

check connection

fate_test config check [OPTIONS]

edit¶

edit fate_test config file

fate_test config edit [OPTIONS]

new¶

create new fate_test config temperate

fate_test config new [OPTIONS]

show¶

show fate_test default config path

fate_test config show [OPTIONS]

data¶

upload or delete data in suite config files

fate_test data [OPTIONS] COMMAND [ARGS]...

delete¶

delete data defined in suite config files

fate_test data delete [OPTIONS]

Options

-i, --include <include>¶: Required include *benchmark.json under these paths

-e, --exclude <exclude>¶: exclude *benchmark.json under these paths

-g, --glob <glob>¶: glob string to filter sub-directory of path specified by <include>

-s, --suite-type <suite_type>¶

Required suite type

Options: testsuite|benchmark

upload¶

upload data defined in suite config files

fate_test data upload [OPTIONS]

Options

-i, --include <include>¶: Required include *benchmark.json under these paths

-e, --exclude <exclude>¶: exclude *benchmark.json under these paths

-g, --glob <glob>¶: glob string to filter sub-directory of path specified by <include>

-s, --suite-type <suite_type>¶

Required suite type

Options: testsuite|benchmark

-r, --role <role>¶: role to process, default to all. use option likes: guest_0, host_0, host

suite¶

process testsuite

fate_test suite [OPTIONS]

Options

-i, --include <include>¶: Required include *testsuite.json under these paths

-e, --exclude <exclude>¶: exclude *testsuite.json under these paths

-r, --replace <replace>¶: a json string represents mapping for replacing fields in data/conf/dsl

-g, --glob <glob>¶: glob string to filter sub-directory of path specified by <include>

--skip-dsl-jobs¶: skip dsl jobs defined in testsuite

--skip-pipeline-jobs¶: skip pipeline jobs defined in testsuite

--skip-data¶: skip uploading data specified in testsuite

--data-only¶: upload data only

--disable-clean-data¶

--enable-clean-data¶