filip/Stalwart
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Manually train and test spam classifier (closes #473 closes #264 closes #257 closes #471)

Browse Source
This commit is contained in:
mdecimus
2024-06-10 13:02:52 +01:00
parent 8d3839a90b
commit 835c7d8c30
13 changed files with 225 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
resources/config/build.py
View File

@@ -37,12 +37,17 @@ scripts = {
"greylist": [
"config.sieve",
"greylist.sieve"
],
"train": [
"config.sieve",
"train.sieve"
]
}
script_names = {
"spam-filter" : "Spam Filter",
"track-replies" : "Track Replies",
"greylist" : "Greylisting"
"greylist" : "Greylisting",
"train": "Train Bayes Classifier"
}
maps = ["spam_config.map",
@@ -69,7 +74,7 @@ def read_file(file):
return f.read() + "\n"
def build_spam_filters(scripts):
spam_filter = "[version]\nspam-filter = \"1.0\"\n\n"
spam_filter = "[version]\nspam-filter = \"1.1\"\n\n"
for script_name, file_list in scripts.items():
script_content = read_and_concatenate(file_list).replace("'''", "\\'\\'\\'")
script_description = script_names[script_name]

74
resources/config/spamfilter.toml
View File

@@ -1,5 +1,5 @@
[version]
spam-filter = "1.0"
spam-filter = "1.1"
[sieve.trusted.scripts.spam-filter]
name = "Spam Filter"
@@ -17,7 +17,7 @@ let "ADD_HEADER_SPAM_RESULT" "key_get('spam-config', 'add-spam-result')";
let "AUTOLEARN_REPLIES_HAM" "key_get('spam-config', 'learn-ham-replies')";
# Whether the bayes classifier should be trained automatically
let "AUTOLEARN_ENABLE" "key_get('spam-config', 'learn-enable')";
let "AUTOLEARN_ENABLE" "key_get('spam-config', 'learn-enable') && !env.test";
# When to learn ham (score >= threshold)
let "AUTOLEARN_HAM_THRESHOLD" "key_get('spam-config', 'learn-ham-threshold')";
@@ -61,7 +61,7 @@ let "urls" "dedup(tokenize(header.subject, 'uri') + body_urls + html_body_urls)"
# Obtain thread name and subject
let "subject_lc" "to_lowercase(header.subject)";
let "subject_clean" "thread_name(header.subject)";
let "body_and_subject" "subject_clean + text_body";
let "body_and_subject" "subject_clean + ' ' + text_body";
# Obtain all recipients
let "recipients" "to_lowercase(header.to:cc:bcc[*].addr[*])";
@@ -2257,7 +2257,7 @@ while "i > 0" {
if eval "is_empty(token_rep)" {
# Set reputation
eval "key_set(SPAM_DB, token_id, [score, 1], 2592000)";
eval "!env.test && key_set(SPAM_DB, token_id, [score, 1], 2592000)";
continue;
}
@@ -2265,7 +2265,7 @@ while "i > 0" {
let "token_score" "token_rep[0]";
let "token_count" "token_rep[1]";
let "updated_score" "(token_count + 1) * (score + 0.98 * token_score) / (0.98 * token_count + 1)";
eval "key_set(SPAM_DB, token_id, [updated_score, token_count + 1], 2592000)";
eval "!env.test && key_set(SPAM_DB, token_id, [updated_score, token_count + 1], 2592000)";
# Assign weight
let "weight" "";
@@ -2343,7 +2343,7 @@ let "ADD_HEADER_SPAM_RESULT" "key_get('spam-config', 'add-spam-result')";
let "AUTOLEARN_REPLIES_HAM" "key_get('spam-config', 'learn-ham-replies')";
# Whether the bayes classifier should be trained automatically
let "AUTOLEARN_ENABLE" "key_get('spam-config', 'learn-enable')";
let "AUTOLEARN_ENABLE" "key_get('spam-config', 'learn-enable') && !env.test";
# When to learn ham (score >= threshold)
let "AUTOLEARN_HAM_THRESHOLD" "key_get('spam-config', 'learn-ham-threshold')";
@@ -2403,7 +2403,7 @@ let "ADD_HEADER_SPAM_RESULT" "key_get('spam-config', 'add-spam-result')";
let "AUTOLEARN_REPLIES_HAM" "key_get('spam-config', 'learn-ham-replies')";
# Whether the bayes classifier should be trained automatically
let "AUTOLEARN_ENABLE" "key_get('spam-config', 'learn-enable')";
let "AUTOLEARN_ENABLE" "key_get('spam-config', 'learn-enable') && !env.test";
# When to learn ham (score >= threshold)
let "AUTOLEARN_HAM_THRESHOLD" "key_get('spam-config', 'learn-ham-threshold')";
@@ -2444,6 +2444,66 @@ if eval "!key_exists(SPAM_DB, triplet)" {
'''
[sieve.trusted.scripts.train]
name = "Train Bayes Classifier"
contents = '''
#### Script config.sieve ####
# Whether to add an X-Spam-Status header
let "ADD_HEADER_SPAM" "key_get('spam-config', 'add-spam')";
# Whether to add an X-Spam-Result header
let "ADD_HEADER_SPAM_RESULT" "key_get('spam-config', 'add-spam-result')";
# Whether message replies from authenticated users should be learned as ham
let "AUTOLEARN_REPLIES_HAM" "key_get('spam-config', 'learn-ham-replies')";
# Whether the bayes classifier should be trained automatically
let "AUTOLEARN_ENABLE" "key_get('spam-config', 'learn-enable') && !env.test";
# When to learn ham (score >= threshold)
let "AUTOLEARN_HAM_THRESHOLD" "key_get('spam-config', 'learn-ham-threshold')";
# When to learn spam (score <= threshold)
let "AUTOLEARN_SPAM_THRESHOLD" "key_get('spam-config', 'learn-spam-threshold')";
# Keep difference for spam/ham learns for at least this value
let "AUTOLEARN_SPAM_HAM_BALANCE" "key_get('spam-config', 'learn-balance')";
# If ADD_HEADER_SPAM is enabled, mark as SPAM messages with a score above this threshold
let "SCORE_SPAM_THRESHOLD" "key_get('spam-config', 'threshold-spam')";
# Discard messages with a score above this threshold
let "SCORE_DISCARD_THRESHOLD" "key_get('spam-config', 'threshold-discard')";
# Reject messages with a score above this threshold
let "SCORE_REJECT_THRESHOLD" "key_get('spam-config', 'threshold-reject')";
# Directory name to use for local domain lookups (leave empty for default)
let "DOMAIN_DIRECTORY" "key_get('spam-config', 'directory')";
# Store to use for Bayes tokens and ids (leave empty for default)
let "SPAM_DB" "key_get('spam-config', 'lookup')";
#### Script train.sieve ####
# Obtain thread name and subject
let "contents" "thread_name(header.subject) + ' ' + body.to_text";
if eval "env.train == 'spam'" {
eval "bayes_train(SPAM_DB, contents, true)";
} elsif eval "env.train == 'ham'" {
eval "bayes_train(SPAM_DB, contents, false)";
} else {
reject "Missing variable 'train'";
}
'''
[lookup]
spam-config = {

2
resources/config/spamfilter/scripts/config.sieve
View File

@@ -8,7 +8,7 @@ let "ADD_HEADER_SPAM_RESULT" "key_get('spam-config', 'add-spam-result')";
let "AUTOLEARN_REPLIES_HAM" "key_get('spam-config', 'learn-ham-replies')";
# Whether the bayes classifier should be trained automatically
let "AUTOLEARN_ENABLE" "key_get('spam-config', 'learn-enable')";
let "AUTOLEARN_ENABLE" "key_get('spam-config', 'learn-enable') && !env.test";
# When to learn ham (score >= threshold)
let "AUTOLEARN_HAM_THRESHOLD" "key_get('spam-config', 'learn-ham-threshold')";

2
resources/config/spamfilter/scripts/prelude.sieve
View File

@@ -13,7 +13,7 @@ let "urls" "dedup(tokenize(header.subject, 'uri') + body_urls + html_body_urls)"
# Obtain thread name and subject
let "subject_lc" "to_lowercase(header.subject)";
let "subject_clean" "thread_name(header.subject)";
let "body_and_subject" "subject_clean + text_body";
let "body_and_subject" "subject_clean + ' ' + text_body";
# Obtain all recipients
let "recipients" "to_lowercase(header.to:cc:bcc[*].addr[*])";

4
resources/config/spamfilter/scripts/reputation.sieve
View File

@@ -41,7 +41,7 @@ while "i > 0" {
if eval "is_empty(token_rep)" {
# Set reputation
eval "key_set(SPAM_DB, token_id, [score, 1], 2592000)";
eval "!env.test && key_set(SPAM_DB, token_id, [score, 1], 2592000)";
continue;
}
@@ -49,7 +49,7 @@ while "i > 0" {
let "token_score" "token_rep[0]";
let "token_count" "token_rep[1]";
let "updated_score" "(token_count + 1) * (score + 0.98 * token_score) / (0.98 * token_count + 1)";
eval "key_set(SPAM_DB, token_id, [updated_score, token_count + 1], 2592000)";
eval "!env.test && key_set(SPAM_DB, token_id, [updated_score, token_count + 1], 2592000)";
# Assign weight
let "weight" "";

12
resources/config/spamfilter/scripts/train.sieve Normal file
View File

@@ -0,0 +1,12 @@
# Obtain thread name and subject
let "contents" "thread_name(header.subject) + ' ' + body.to_text";
if eval "env.train == 'spam'" {
eval "bayes_train(SPAM_DB, contents, true)";
} elsif eval "env.train == 'ham'" {
eval "bayes_train(SPAM_DB, contents, false)";
} else {
reject "Missing variable 'train'";
}