admin/splunk-security_content
1
0
Fork 0
mirror of https://github.com/splunk/security_content synced 2026-06-08 17:32:49 +00:00
Code Issues Packages Projects Releases Wiki Activity

Branch was auto-updated.

Browse Source
This commit is contained in:
srv-rr-gh-researchbt
2023-02-07 10:39:50 -08:00
committed by GitHub
parent 6c3e7d21ce 1d456aaf23
commit 5ea61731a3
8 changed files with 1114 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files
detections/experimental/endpoint/detect_suspicious_processnames_using_pretrained_model_in_dsdl.yml
+117
View File
@@ -0,0 +1,117 @@
name: Detect suspicious processnames using a pretrained model in DSDL
id: a15f8977-ad7d-4669-92ef-b59b97219bf5
version: 1
date: "2023-01-23"
author: Abhinav Mishra, Kumar Sharad and Namratha Sreekanta, Splunk
type: Anomaly
datamodel:
- Endpoint
description: The following analytic uses a pre-trained Deep Learning model to
predict whether a processname is suspicious or not. Malwares and malicious
programs such as ransomware often use tactics, techniques, and procedures
(TTPs) such as copying malicious files to the local machine to propagate
themselves across the network. A key indicator of compromise is that after
a successful execution of the malware, it copies itself as an executable file
with a randomly generated filename and places this file in one of the
directories. Such techniques are seen in several malwares such as TrickBot.
We develop machine learning model that uses a Recurrent Neural Network (RNN)
to distinguish between malicious and benign processnames. The model is trained
independently and is then made available for download. We use a character level
RNN to classify malicious vs. benign processnames. The higher is_malicious_prob,
the more likely is the processname to be suspicious (between [0,1]). The threshold
for flagging a processname as suspicious is set as 0.5.
search: '| tstats `security_content_summariesonly` count min(_time) as firstTime
max(_time) as lastTime from datamodel=Endpoint.Processes
by Processes.process_name Processes.parent_process_name Processes.process
Processes.user Processes.dest
| `drop_dm_object_name(Processes)`
| rename process_name as text
| fields text, parent_process_name, process, user, dest
| apply detect_suspicious_processnames_using_pretrained_model_in_dsdl
| rename predicted_label as is_suspicious_score
| rename text as process_name
| where is_suspicious_score > 0.5
| `detect_suspicious_processnames_using_pretrained_model_in_dsdl_filter`'
how_to_implement: 'Steps to deploy detect suspicious processnames model into Splunk App
DSDL. This detection depends on the Splunk app for Data Science and Deep
Learning which can be found here - `https://splunkbase.splunk.com/app/4607/`
and the Endpoint datamodel. The detection uses a pre-trained
deep learning model that needs to be deployed in the DSDL app. Follow the steps
for deployment here - `https://github.com/splunk/security_content/wiki/How-to-deploy-pre-trained-Deep-Learning-models-for-ESCU`.\
* Download the `artifacts .tar.gz` file from the link - `https://seal.splunkresearch.com/detect_suspicious_processnames_using_pretrained_model_in_dsdl.tar.gz`.\
* Download the `detect_suspicious_processnames_using_pretrained_model_in_dsdl.ipynb`
Jupyter notebook from the link - `https://github.com/splunk/security_content/notebooks`.\
* Login to the Jupyter Lab assigned for `detect_suspicious_processnames_using_pretrained_model_in_dsdl`
container. This container should be listed on Containers page for DSDL app.\
* Follow the steps below inside Jupyter Notebook:\
* Upload the `detect_suspicious_processnames_using_pretrained_model_in_dsdl.tar.gz` file
into `app/model/data/detect_suspicious_processnames_using_pretrained_model_in_dsdl/` path using the upload option in the Jupyter Notebook.\
* Untar the artifact `detect_suspicious_processnames_using_pretrained_model_in_dsdl.tar.gz` using
`tar -xf app/model/data/detect_suspicious_processnames_using_pretrained_model_in_dsdl.tar.gz -C app/model/data/detect_suspicious_processnames_using_pretrained_model_in_dsdl/`.\
* Upload `detect_suspicious_processnames_using_pretrained_model_in_dsdl.ipynb` into the Jupyter Notebooks
folder using the upload option in Jupyter Notebook.\
* Save the notebook using the save option in Jupyter Notebook.\
* Upload `detect_suspicious_processnames_using_pretrained_model_in_dsdl.json` into `notebooks/data`
folder.'
known_false_positives: False positives may be present if a suspicious processname
is similar to a benign processname.
references:
- https://www.cisa.gov/uscert/ncas/alerts/aa20-302a
- https://www.splunk.com/en_us/blog/security/random-words-on-entropy-and-dns.html
tags:
analytic_story:
- Suspicious Command-Line Executions
asset_type: Endpoint
cis20:
- CIS 8
- CIS 12
- CIS 13
confidence: 90
context:
- Source:Endpoint
- Stage:Execution
impact: 50
kill_chain_phases:
- Exploitation
message: The process $process$ is running from an unusual place by $user$ on $dest$ with a processname
that appears to be randomly generated.
mitre_attack_id:
- T1059
nist:
- PR.DS
- PR.PT
- DE.AE
- DE.CM
observable:
- name: dest
type: Hostname
role:
- Victim
- name: user
type: User
role:
- Victim
product:
- Splunk Enterprise
- Splunk Enterprise Security
- Splunk Cloud
required_fields:
- _time
- Processes.process
- Processes.parent_process_name
- Processes.process_name
- Processes.parent_process
- Processes.user
- Processes.dest
risk_score: 45
security_domain: Endpoint
detections/experimental/network/detect_suspicious_dns_txt_records_using_pretrained_model_in_dsdl.yml
+13 -12
View File
@@ -35,30 +35,31 @@ search: '| tstats `security_content_summariesonly` count min(_time) as firstTime
how_to_implement: 'Steps to deploy detect suspicious DNS TXT records model into Splunk App
DSDL. This detection depends on the Splunk app for Data Science and Deep
Learning which can be found here - https://splunkbase.splunk.com/app/4607/
Learning which can be found here - `https://splunkbase.splunk.com/app/4607/`
and the Network Resolution datamodel which can be found here -
https://splunkbase.splunk.com/app/1621/. The detection uses a pre-trained
`https://splunkbase.splunk.com/app/1621/`. The detection uses a pre-trained
deep learning model that needs to be deployed in DSDL app. Follow the steps
for deployment here - `https://github.com/splunk/security_content/wiki/How-to-deploy-pre-trained-Deep-Learning-models-for-ESCU`.\
* Download the `artifacts .tar.gz` file from the link - https://seal.splunkresearch.com/detect_suspicious_dns_txt_records_using_pretrained_model_in_dsdl.tar.gz
Download the `detect_suspicious_dns_txt_records_using_pretrained_model_in_dsdl.ipynb`
Jupyter notebook from https://github.com/splunk/security_content/notebooks\
* Download the `artifacts .tar.gz` file from the link - `https://seal.splunkresearch.com/detect_suspicious_dns_txt_records_using_pretrained_model_in_dsdl.tar.gz`.
* Download the `detect_suspicious_dns_txt_records_using_pretrained_model_in_dsdl.ipynb`
Jupyter notebook from `https://github.com/splunk/security_content/notebooks`.\
* Login to the Jupyter Lab assigned for
detect_suspicious_dns_txt_records_using_pretrained_model_in_dsdl container.
`detect_suspicious_dns_txt_records_using_pretrained_model_in_dsdl` container.
This container should be listed on Containers page for DSDL app.\
* Below steps need to be followed inside Jupyter lab \
* Below steps need to be followed inside Jupyter lab.\
* Upload the detect_suspicious_dns_txt_records_using_pretrained_model_in_dsdl.tar.gz file
* Upload the `detect_suspicious_dns_txt_records_using_pretrained_model_in_dsdl.tar.gz` file
into `app/model/data` path using the upload option in the jupyter notebook.\
* Untar the artifact detect_suspicious_dns_txt_records_using_pretrained_model_in_dsdl.tar.gz using `tar -xf app/model/data/detect_suspicious_dns_txt_records_using_pretrained_model_in_dsdl.tar.gz -C app/model/data`\
* Untar the artifact `detect_suspicious_dns_txt_records_using_pretrained_model_in_dsdl.tar.gz` using `tar -xf app/model/data/detect_suspicious_dns_txt_records_using_pretrained_model_in_dsdl.tar.gz -C app/model/data`.\
* Upload detect_suspicious_dns_txt_records_using_pretrained_model_in_dsdl.pynb into Jupyter lab notebooks folder using the upload option in Jupyter lab \
* Upload detect_suspicious_dns_txt_records_using_pretrained_model_in_dsdl.ipynb` into Jupyter lab notebooks folder using the upload option in Jupyter lab.\
* Save the notebook using the save option in jupyter notebook. \
* Save the notebook using the save option in Jupyter notebook.\
* Upload `detect_suspicious_dns_txt_records_using_pretrained_model_in_dsdl.json` into `notebooks/data` folder.'
known_false_positives: False positives may be present if DNS TXT record contents
@@ -113,4 +114,4 @@ tags:
- DNS.dest
- DNS.answer
risk_score: 45
security_domain: network
security_domain: network
lookups/__mlspl_detect_suspicious_processnames_using_pretrained_model_in_dsdl.mlmodel
+2
View File
@@ -0,0 +1,2 @@
algo,model,options
MLTKContainer,"{""__mlspl_type"": [""mltkc.MLTKContainer"", ""MLTKContainer""], ""dict"": {""endpoint_url"": ""https://localhost:58216"", ""out_params"": {""params"": {""mode"": ""stage"", ""algo"": ""detect_suspicious_processnames_using_pretrained_model_in_dsdl""}, ""args"": [""label"", ""text""], ""target_variable"": [""label""], ""feature_variables"": [""text""], ""model_name"": ""detect_suspicious_processnames_using_pretrained_model_in_dsdl"", ""algo_name"": ""MLTKContainer"", ""mlspl_limits"": {""handle_new_cat"": ""default"", ""max_distinct_cat_values"": ""100"", ""max_distinct_cat_values_for_classifiers"": ""100"", ""max_distinct_cat_values_for_scoring"": ""100"", ""max_fit_time"": ""600"", ""max_inputs"": ""100000"", ""max_memory_usage_mb"": ""4000"", ""max_model_size_mb"": ""30"", ""max_score_time"": ""600"", ""use_sampling"": ""true""}, ""kfold_cv"": null}, ""feature_variables"": [""text""], ""target_variable"": ""label""}}","{""params"": {""mode"": ""stage"", ""algo"": ""detect_suspicious_processnames_using_pretrained_model_in_dsdl""}, ""args"": [""label"", ""text""], ""target_variable"": [""label""], ""feature_variables"": [""text""], ""model_name"": ""detect_suspicious_processnames_using_pretrained_model_in_dsdl"", ""algo_name"": ""MLTKContainer"", ""mlspl_limits"": {""handle_new_cat"": ""default"", ""max_distinct_cat_values"": ""100"", ""max_distinct_cat_values_for_classifiers"": ""100"", ""max_distinct_cat_values_for_scoring"": ""100"", ""max_fit_time"": ""600"", ""max_inputs"": ""100000"", ""max_memory_usage_mb"": ""4000"", ""max_model_size_mb"": ""30"", ""max_score_time"": ""600"", ""use_sampling"": ""true""}, ""kfold_cv"": null}"
lookups/__mlspl_detect_suspicious_processnames_using_pretrained_model_in_dsdl.yml
+4
View File
@@ -0,0 +1,4 @@
description: Detect a suspicious processname using Pretrained Model in DSDL
filename: __mlspl_detect_suspicious_processnames_using_pretrained_model_in_dsdl.mlmodel
name: __mlspl_detect_suspicious_processnames_using_pretrained_model_in_dsdl
case_sensitive_match: 'false'
notebooks/detect_suspicious_processnames_using_pretrained_model_in_dsdl.ipynb
+973
View File
@@ -0,0 +1,973 @@
{
"cells": [
{
"cell_type": "markdown",
"metadata": {
"tags": []
},
"source": [
"# Deep Learning model to detect suspiciously named files"
]
},
{
"cell_type": "markdown",
"metadata": {
"tags": []
},
"source": [
"This notebook uses a pre-trained Deep Learning model to predict whether a processname is suspicious or not.\n",
"\n",
"Malwares and malicious programs such as ransomware often use tactics, techniques, and procedures (TTPs) such as copying malicious files to the local machine to propagate themselves across the network. A key indicator of compromise is that after a successful execution of the malware, it copies itself as an executable file with a randomly generated filename and places this file in one of the directories. Such techniques are seen in several malwares such as TrickBot. \n",
"\n",
"We develop machine learning model that uses a Recurrent Neural Network (RNN) to distinguish between malicious and benign processnames. The model is trained independently and is then made available for download. We use a character level RNN to classify malicious vs. benign processnames. Higher the `is_malicious_prob`, the more likely is the processname to be suspicious (between `[0,1]`). The threshold for flagging a processname as suspicious is set at `0.5`."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Stage 0 - import libraries\n",
"At stage 0 we define all imports necessary to run our subsequent code depending on various libraries."
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {
"deletable": false,
"name": "mltkc_import"
},
"outputs": [],
"source": [
"# This definition exposes all python module imports that should be available in all subsequent commands\n",
"\n",
"import pandas as pd\n",
"import numpy as np\n",
"import random\n",
"import string\n",
"import torch\n",
"import torch.nn as nn\n",
"\n",
"# global constants\n",
"MODEL_DIRECTORY = \"/srv/app/model/data/detect_suspicious_processnames_using_pretrained_model_in_dsdl/detect_suspicious_processnames_using_pretrained_model_in_dsdl.pt\"\n",
"\n",
"# model parameters\n",
"all_letters = string.ascii_lowercase\n",
"n_letters = len(all_letters)\n",
"n_hidden = 32\n",
"n_categories = 2\n",
"learning_rate = 5e-4"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"numpy version: 1.23.3\n",
"pandas version: 1.5.0\n"
]
}
],
"source": [
"# THIS CELL IS NOT EXPORTED - free notebook cell for testing or development purposes\n",
"print(\"numpy version: \" + np.__version__)\n",
"print(\"pandas version: \" + pd.__version__)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Stage 1 - get a data sample from Splunk\n",
"There are currently 2 ways to retrieve data from Splunk: Option 1 is to interactively pull data from Splunk into the DSDL Jupyter Lab environment. This is useful when the Splunk REST API is accessible from the Jupyter environment and a valid Splunk auth token is defined in the DSDL app. This option has advantages to quickly experiment with different Splunk SPL queries and further interactively work with the search results in Jupyter.\n",
"\n",
"### Option 1 - pull data from Splunk"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [],
"source": [
"from dsdlsupport import SplunkSearch as SplunkSearch"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "07142129edfe4bb6b4159c9fbc7ec0d8",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"VBox(children=(HBox(children=(Textarea(value='| makeresults count=10 \\n| streamstats c as i \\n| eval s = i%3 \\…"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"search = SplunkSearch.SplunkSearch()"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>_time</th>\n",
" <th>feature_0</th>\n",
" <th>feature_1</th>\n",
" <th>feature_2</th>\n",
" <th>i</th>\n",
" <th>s</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>2022-10-12T15:58:22.000+02:00</td>\n",
" <td>0.2428163280710578</td>\n",
" <td>0.2446760144084692</td>\n",
" <td>0.878678847104311</td>\n",
" <td>1</td>\n",
" <td>1</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>2022-10-12T15:58:22.000+02:00</td>\n",
" <td>0.458672939799726</td>\n",
" <td>0.4403486642986536</td>\n",
" <td>0.5951908086426556</td>\n",
" <td>2</td>\n",
" <td>2</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>2022-10-12T15:58:22.000+02:00</td>\n",
" <td>0.4268499561585486</td>\n",
" <td>0.70259142620489</td>\n",
" <td>0.6676120953634381</td>\n",
" <td>3</td>\n",
" <td>0</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>2022-10-12T15:58:22.000+02:00</td>\n",
" <td>0.9911085437051952</td>\n",
" <td>0.7537638866342604</td>\n",
" <td>0.8423880040645599</td>\n",
" <td>4</td>\n",
" <td>1</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>2022-10-12T15:58:22.000+02:00</td>\n",
" <td>0.2864162167534232</td>\n",
" <td>0.4240174447186291</td>\n",
" <td>0.7597004366107285</td>\n",
" <td>5</td>\n",
" <td>2</td>\n",
" </tr>\n",
" <tr>\n",
" <th>5</th>\n",
" <td>2022-10-12T15:58:22.000+02:00</td>\n",
" <td>0.9403447071090341</td>\n",
" <td>0.10794003773480654</td>\n",
" <td>0.22778158355504274</td>\n",
" <td>6</td>\n",
" <td>0</td>\n",
" </tr>\n",
" <tr>\n",
" <th>6</th>\n",
" <td>2022-10-12T15:58:22.000+02:00</td>\n",
" <td>0.6541405455209315</td>\n",
" <td>0.7657363642938435</td>\n",
" <td>0.07344387145712972</td>\n",
" <td>7</td>\n",
" <td>1</td>\n",
" </tr>\n",
" <tr>\n",
" <th>7</th>\n",
" <td>2022-10-12T15:58:22.000+02:00</td>\n",
" <td>0.8196326880715787</td>\n",
" <td>0.34731235541403294</td>\n",
" <td>0.8556739939376712</td>\n",
" <td>8</td>\n",
" <td>2</td>\n",
" </tr>\n",
" <tr>\n",
" <th>8</th>\n",
" <td>2022-10-12T15:58:22.000+02:00</td>\n",
" <td>0.3521619248203933</td>\n",
" <td>0.04771727975457907</td>\n",
" <td>0.07742329593747854</td>\n",
" <td>9</td>\n",
" <td>0</td>\n",
" </tr>\n",
" <tr>\n",
" <th>9</th>\n",
" <td>2022-10-12T15:58:22.000+02:00</td>\n",
" <td>0.2476972616277635</td>\n",
" <td>0.5393188064917922</td>\n",
" <td>0.7553935945034027</td>\n",
" <td>10</td>\n",
" <td>1</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"</div>"
],
"text/plain": [
" _time feature_0 feature_1 \\\n",
"0 2022-10-12T15:58:22.000+02:00 0.2428163280710578 0.2446760144084692 \n",
"1 2022-10-12T15:58:22.000+02:00 0.458672939799726 0.4403486642986536 \n",
"2 2022-10-12T15:58:22.000+02:00 0.4268499561585486 0.70259142620489 \n",
"3 2022-10-12T15:58:22.000+02:00 0.9911085437051952 0.7537638866342604 \n",
"4 2022-10-12T15:58:22.000+02:00 0.2864162167534232 0.4240174447186291 \n",
"5 2022-10-12T15:58:22.000+02:00 0.9403447071090341 0.10794003773480654 \n",
"6 2022-10-12T15:58:22.000+02:00 0.6541405455209315 0.7657363642938435 \n",
"7 2022-10-12T15:58:22.000+02:00 0.8196326880715787 0.34731235541403294 \n",
"8 2022-10-12T15:58:22.000+02:00 0.3521619248203933 0.04771727975457907 \n",
"9 2022-10-12T15:58:22.000+02:00 0.2476972616277635 0.5393188064917922 \n",
"\n",
" feature_2 i s \n",
"0 0.878678847104311 1 1 \n",
"1 0.5951908086426556 2 2 \n",
"2 0.6676120953634381 3 0 \n",
"3 0.8423880040645599 4 1 \n",
"4 0.7597004366107285 5 2 \n",
"5 0.22778158355504274 6 0 \n",
"6 0.07344387145712972 7 1 \n",
"7 0.8556739939376712 8 2 \n",
"8 0.07742329593747854 9 0 \n",
"9 0.7553935945034027 10 1 "
]
},
"execution_count": 7,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"df = search.as_df()\n",
"df"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Option 2 - push data from Splunk\n",
"In Splunk run a search to pipe a dataset into your notebook environment. You utilize the `mode=stage` flag in the in the `| fit` command to do this. The search results are accessible then as csv file with the same model name that is defined in the `into app:<modelname>` part of the fit statement. Additionally, meta data is retrieved and accessible as json file. In the same way you can further work with the meta data object as it is exposed in the fit and apply function definitions below in stage 3 and 4."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"| makeresults count=10<br>\n",
"| streamstats c as i <br>\n",
"| eval s = i%3 <br>\n",
"| eval feature_{s}=0 <br>\n",
"| foreach feature_* [eval \\<\\<FIELD\\>\\>=random()/pow(2,31)]<br>\n",
"| fit MLTKContainer mode=stage algo=barebone_template _time feature_* i into app:barebone_template<br>"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"After you run this search your data set sample is available as a csv inside the container to develop your model. The name is taken from the into keyword (\"barebone_model\" in the example above) or set to \"default\" if no into keyword is present. This step is intended to work with a subset of your data to create your custom model."
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {
"deletable": false,
"name": "mltkc_stage"
},
"outputs": [],
"source": [
"# this cell is not executed from MLTK and should only be used for staging data into the notebook environment\n",
"def stage(name):\n",
" with open(\"data/\"+name+\".csv\", 'r') as f:\n",
" df = pd.read_csv(f)\n",
" with open(\"data/\"+name+\".json\", 'r') as f:\n",
" param = json.load(f)\n",
" return df, param"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {},
"outputs": [],
"source": [
"# THIS CELL IS NOT EXPORTED - free notebook cell for testing or development purposes\n",
"df, param = stage(\"barebone_template\")"
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>_time</th>\n",
" <th>i</th>\n",
" <th>feature_1</th>\n",
" <th>feature_2</th>\n",
" <th>feature_0</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>1665565121</td>\n",
" <td>1</td>\n",
" <td>0.749423</td>\n",
" <td>0.861980</td>\n",
" <td>0.063747</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>1665565121</td>\n",
" <td>2</td>\n",
" <td>0.328421</td>\n",
" <td>0.494055</td>\n",
" <td>0.578063</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>1665565121</td>\n",
" <td>3</td>\n",
" <td>0.456588</td>\n",
" <td>0.135485</td>\n",
" <td>0.935327</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>1665565121</td>\n",
" <td>4</td>\n",
" <td>0.034384</td>\n",
" <td>0.457241</td>\n",
" <td>0.043726</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>1665565121</td>\n",
" <td>5</td>\n",
" <td>0.319659</td>\n",
" <td>0.867864</td>\n",
" <td>0.595515</td>\n",
" </tr>\n",
" <tr>\n",
" <th>...</th>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" <td>...</td>\n",
" </tr>\n",
" <tr>\n",
" <th>995</th>\n",
" <td>1665565121</td>\n",
" <td>996</td>\n",
" <td>0.171754</td>\n",
" <td>0.162497</td>\n",
" <td>0.923288</td>\n",
" </tr>\n",
" <tr>\n",
" <th>996</th>\n",
" <td>1665565121</td>\n",
" <td>997</td>\n",
" <td>0.500971</td>\n",
" <td>0.899659</td>\n",
" <td>0.607057</td>\n",
" </tr>\n",
" <tr>\n",
" <th>997</th>\n",
" <td>1665565121</td>\n",
" <td>998</td>\n",
" <td>0.209019</td>\n",
" <td>0.751987</td>\n",
" <td>0.469225</td>\n",
" </tr>\n",
" <tr>\n",
" <th>998</th>\n",
" <td>1665565121</td>\n",
" <td>999</td>\n",
" <td>0.163475</td>\n",
" <td>0.825084</td>\n",
" <td>0.158101</td>\n",
" </tr>\n",
" <tr>\n",
" <th>999</th>\n",
" <td>1665565121</td>\n",
" <td>1000</td>\n",
" <td>0.208661</td>\n",
" <td>0.545487</td>\n",
" <td>0.872765</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"<p>1000 rows × 5 columns</p>\n",
"</div>"
],
"text/plain": [
" _time i feature_1 feature_2 feature_0\n",
"0 1665565121 1 0.749423 0.861980 0.063747\n",
"1 1665565121 2 0.328421 0.494055 0.578063\n",
"2 1665565121 3 0.456588 0.135485 0.935327\n",
"3 1665565121 4 0.034384 0.457241 0.043726\n",
"4 1665565121 5 0.319659 0.867864 0.595515\n",
".. ... ... ... ... ...\n",
"995 1665565121 996 0.171754 0.162497 0.923288\n",
"996 1665565121 997 0.500971 0.899659 0.607057\n",
"997 1665565121 998 0.209019 0.751987 0.469225\n",
"998 1665565121 999 0.163475 0.825084 0.158101\n",
"999 1665565121 1000 0.208661 0.545487 0.872765\n",
"\n",
"[1000 rows x 5 columns]"
]
},
"execution_count": 11,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"df"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'options': {'params': {'mode': 'stage', 'algo': 'barebone_template'},\n",
" 'args': ['_time', 'feature_*', 'i'],\n",
" 'feature_variables': ['_time', 'feature_*', 'i'],\n",
" 'model_name': 'barebone_template',\n",
" 'algo_name': 'MLTKContainer',\n",
" 'mlspl_limits': {'handle_new_cat': 'default',\n",
" 'max_distinct_cat_values': '100',\n",
" 'max_distinct_cat_values_for_classifiers': '100',\n",
" 'max_distinct_cat_values_for_scoring': '100',\n",
" 'max_fit_time': '600',\n",
" 'max_inputs': '100000',\n",
" 'max_memory_usage_mb': '4000',\n",
" 'max_model_size_mb': '30',\n",
" 'max_score_time': '600',\n",
" 'use_sampling': 'true'},\n",
" 'kfold_cv': None},\n",
" 'feature_variables': ['_time', 'feature_1', 'feature_2', 'feature_0', 'i']}"
]
},
"execution_count": 8,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"param"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Stage 2 - create and initialize a model"
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {
"deletable": false,
"name": "mltkc_init"
},
"outputs": [],
"source": [
"# initialize your model\n",
"# available inputs: data and parameters\n",
"# returns the model object which will be used as a reference to call fit, apply and summary subsequently\n",
"\n",
"class ProcessnameClassifier(nn.Module):\n",
" def __init__(self, input_size, hidden_size, output_size):\n",
" super(ProcessnameClassifier, self).__init__()\n",
" self.hidden_size = hidden_size\n",
" self.i2h = nn.Linear(input_size + hidden_size, hidden_size)\n",
" self.i2o = nn.Linear(input_size + hidden_size, output_size)\n",
" self.softmax = nn.LogSoftmax(dim=1)\n",
"\n",
" def forward(self, input, hidden):\n",
" combined = torch.cat((input, hidden), 1)\n",
" hidden = self.i2h(combined)\n",
" output = self.i2o(combined)\n",
" output = self.softmax(output)\n",
" return output, hidden\n",
"\n",
" def initHidden(self):\n",
" return torch.zeros(1, self.hidden_size)\n",
"\n",
" \n",
"def init(df, param):\n",
" model = ProcessnameClassifier(n_letters, n_hidden, n_categories)\n",
" model.load_state_dict(torch.load(MODEL_DIRECTORY, map_location=torch.device('cpu')))\n",
" model.eval()\n",
" \n",
" return model\n",
" "
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {
"scrolled": true
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{'hyperparameter': 42.0}\n"
]
}
],
"source": [
"# THIS CELL IS NOT EXPORTED - free notebook cell for testing or development purposes\n",
"model = init(df,param)\n",
"print(model)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Stage 3 - fit the model"
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {
"deletable": false,
"name": "mltkc_fit"
},
"outputs": [],
"source": [
"# train your model\n",
"# returns a fit info json object and may modify the model object\n",
"def fit(model,df,param):\n",
" # model.fit()\n",
" info = {\"message\": \"model trained\"}\n",
" return info"
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{'message': 'model trained'}\n"
]
}
],
"source": [
"# THIS CELL IS NOT EXPORTED - free notebook cell for testing or development purposes\n",
"print(fit(model,df,param))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Stage 4 - apply the model"
]
},
{
"cell_type": "code",
"execution_count": 17,
"metadata": {
"deletable": false,
"name": "mltkc_apply"
},
"outputs": [],
"source": [
"# apply your model\n",
"# returns the calculated results\n",
"\n",
"# Find letter index from all_letters\n",
"def letterToIndex(letter):\n",
" return all_letters.find(letter)\n",
"\n",
"\n",
"# Turn a line into a <line_length x 1 x n_letters>,\n",
"# or an array of one-hot letter vectors\n",
"def lineToTensor(line):\n",
" tensor = torch.zeros(len(line), 1, n_letters)\n",
" for li, letter in enumerate(line):\n",
" tensor[li][0][letterToIndex(letter)] = 1\n",
" return tensor\n",
"\n",
"\n",
"def evaluate(model, line_tensor):\n",
" hidden = model.initHidden()\n",
"\n",
" for i in range(line_tensor.size()[0]):\n",
" output, hidden = model(line_tensor[i], hidden)\n",
"\n",
" return output\n",
"\n",
"\n",
"def preprocess_txt(input_line):\n",
" input_line = input_line.split('/')[-1]\n",
" input_line = input_line.lower()\n",
" input_line = input_line.replace('.exe','')\n",
" input_line = ''.join(filter(str.islower, input_line))\n",
" \n",
" return input_line\n",
"\n",
"\n",
"def apply(model, df, param):\n",
" is_malicious_prob_lst = []\n",
" for idx, row in df.iterrows():\n",
" input_line = row['text']\n",
" input_line = preprocess_txt(input_line)\n",
" \n",
" if len(input_line) == 0:\n",
" is_mal_prob = 1\n",
" else:\n",
" with torch.no_grad():\n",
" output = evaluate(model, lineToTensor(input_line))\n",
"\n",
" output = torch.exp(output)\n",
" is_mal_prob = round(output[0][1].item(), 4)\n",
"\n",
" is_malicious_prob_lst.append(is_mal_prob)\n",
" \n",
" output = pd.DataFrame()\n",
" output['is_malicious_prob'] = is_malicious_prob_lst\n",
"\n",
" return output"
]
},
{
"cell_type": "code",
"execution_count": 18,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
" index\n",
"0 0\n",
"1 1\n",
"2 2\n",
"3 3\n",
"4 4\n",
".. ...\n",
"995 995\n",
"996 996\n",
"997 997\n",
"998 998\n",
"999 999\n",
"\n",
"[1000 rows x 1 columns]\n"
]
}
],
"source": [
"# THIS CELL IS NOT EXPORTED - free notebook cell for testing or development purposes\n",
"print(apply(model,df,param))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Stage 5 - save the model"
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {
"deletable": false,
"name": "mltkc_save"
},
"outputs": [],
"source": [
"# save model to name in expected convention \"<algo_name>_<model_name>\"\n",
"def save(model,name):\n",
" with open(MODEL_DIRECTORY + name + \".json\", 'w') as file:\n",
" json.dump(model, file)\n",
" return model"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Stage 6 - load the model"
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {
"deletable": false,
"name": "mltkc_load"
},
"outputs": [],
"source": [
"# load model from name in expected convention \"<algo_name>_<model_name>\"\n",
"def load(name):\n",
" model = ProcessnameClassifier(n_letters, n_hidden, n_categories)\n",
" model.load_state_dict(torch.load(MODEL_DIRECTORY, map_location=torch.device('cpu')))\n",
" model.eval()\n",
" return model "
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Stage 7 - provide a summary of the model"
]
},
{
"cell_type": "code",
"execution_count": 17,
"metadata": {
"deletable": false,
"name": "mltkc_summary"
},
"outputs": [],
"source": [
"# return a model summary\n",
"def summary(model=None):\n",
" returns = {\"version\": {\"numpy\": np.__version__, \"pandas\": pd.__version__} }\n",
" return returns"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"After implementing your fit, apply, save and load you can train your model:<br>\n",
"| makeresults count=10<br>\n",
"| streamstats c as i<br>\n",
"| eval s = i%3<br>\n",
"| eval feature_{s}=0<br>\n",
"| foreach feature_* [eval &lt;&lt;FIELD&gt;&gt;=random()/pow(2,31)]<br>\n",
"| fit MLTKContainer algo=barebone s from feature_* into app:barebone_model<br>"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Or apply your model:<br>\n",
"| makeresults count=10<br>\n",
"| streamstats c as i<br>\n",
"| eval s = i%3<br>\n",
"| eval feature_{s}=0<br>\n",
"| foreach feature_* [eval &lt;&lt;FIELD&gt;&gt;=random()/pow(2,31)]<br>\n",
"| apply barebone_model as the_meaning_of_life"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Send data back to Splunk HEC\n",
"When you configured the Splunk HEC Settings in the DSDL app you can easily send back data to an index with [Splunk's HTTP Event Collector (HEC)](https://docs.splunk.com/Documentation/Splunk/latest/Data/UsetheHTTPEventCollector). Read more about data formats and options in the [documentation](https://docs.splunk.com/Documentation/Splunk/latest/Data/FormateventsforHTTPEventCollector#Event_metadata).\n",
"\n",
"### Use cases\n",
"- you want to offload longer running, possibly distributed computations that need to deliver results asynchroneously back into Splunk. \n",
"- you might not want to present results back into the search pipeline after your `| fit` or `| apply` command. \n",
"- you can easily utilize this approach for any logging purposes or other profiling tasks in your ML code so you can actively monitor and analyze your processes.\n",
"\n",
"### Example"
]
},
{
"cell_type": "code",
"execution_count": 18,
"metadata": {},
"outputs": [],
"source": [
"from dsdlsupport import SplunkHEC as SplunkHEC\n",
"hec = SplunkHEC.SplunkHEC()"
]
},
{
"cell_type": "code",
"execution_count": 19,
"metadata": {},
"outputs": [],
"source": [
"# example to send 10 hello world events\n",
"response = hec.send_hello_world(10)"
]
},
{
"cell_type": "code",
"execution_count": 20,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"HEC endpoint http://host.docker.internal:8088/services/collector/event \n",
"returned with status code 200 \n",
"and response message: {\"text\":\"Success\",\"code\":0}\n"
]
}
],
"source": [
"print(\"HEC endpoint %s \\nreturned with status code %s \\nand response message: %s\" % (response.url, response.status_code, response.text))"
]
},
{
"cell_type": "code",
"execution_count": 21,
"metadata": {},
"outputs": [],
"source": [
"# example to send a JSON object, e.g. to log some data\n",
"from datetime import datetime\n",
"response = hec.send({'event': {'message': 'operation done', 'log_level': 'INFO' }, 'time': datetime.now().timestamp()})"
]
},
{
"cell_type": "code",
"execution_count": 22,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"HEC endpoint http://host.docker.internal:8088/services/collector/event \n",
"returned with status code 200 \n",
"and response message: {\"text\":\"Success\",\"code\":0}\n"
]
}
],
"source": [
"print(\"HEC endpoint %s \\nreturned with status code %s \\nand response message: %s\" % (response.url, response.status_code, response.text))"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## End of Stages\n",
"All subsequent cells are not tagged and can be used for further freeform code"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.9.13"
}
},
"nbformat": 4,
"nbformat_minor": 4
}
notebooks/detect_suspicious_processnames_using_pretrained_model_in_dsdl.json
+1
View File
@@ -0,0 +1 @@
{"options": {"params": {"mode": "stage", "algo": "detect_suspicious_processnames_using_pretrained_model_in_dsdl"}, "args": ["label", "text"], "target_variable": ["label"], "feature_variables": ["text"], "model_name": "detect_suspicious_processnames_using_pretrained_model_in_dsdl", "algo_name": "MLTKContainer", "mlspl_limits": {"handle_new_cat": "default", "max_distinct_cat_values": "100", "max_distinct_cat_values_for_classifiers": "100", "max_distinct_cat_values_for_scoring": "100", "max_fit_time": "600", "max_inputs": "100000", "max_memory_usage_mb": "4000", "max_model_size_mb": "30", "max_score_time": "600", "use_sampling": "true"}, "kfold_cv": null}, "feature_variables": ["text"], "target_variables": ["label"]}
notebooks/detect_suspicious_processnames_using_pretrained_model_in_dsdl.yml
+3
View File
@@ -0,0 +1,3 @@
description: A Jupyter notebook for detecting suspicious processnames using a pre-trained model in DSDL
filename: detect_suspicious_processnames_using_pretrained_model_in_dsdl.ipynb
name: detect_suspicious_processnames_using_pretrained_model_in_dsdl
requirements.txt
+1 -1
View File
@@ -7,7 +7,7 @@ mock==4.0.3
psutil==5.9.4
pycvesearch==1.2
pydantic==1.10.4
pytest==7.2.0
pytest==7.2.1
PyYAML==5.4.1
questionary==1.10.0
requests==2.28.2