1 | #=============================================================================== |
---|
2 | # SpiderLing -- A web spider/crawler for linguistic purpose. |
---|
3 | # Copyright (c) 2011-2017 Vit Suchomel |
---|
4 | # http://corpus.tools/wiki/SpiderLing |
---|
5 | # |
---|
6 | # This program is free software: you can redistribute it and/or modify |
---|
7 | # it under the terms of the GNU General Public License as published by |
---|
8 | # the Free Software Foundation, either version 3 of the License, or |
---|
9 | # (at your option) any later version. |
---|
10 | # |
---|
11 | # This program is distributed in the hope that it will be useful, |
---|
12 | # but WITHOUT ANY WARRANTY; without even the implied warranty of |
---|
13 | # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
---|
14 | # GNU General Public License for more details. |
---|
15 | # |
---|
16 | # You should have received a copy of the GNU General Public License |
---|
17 | # along with this program. If not, see <http://www.gnu.org/licenses/>. |
---|
18 | #=============================================================================== |
---|
19 | |
---|
20 | #=============================================================================== |
---|
21 | # SPIDERLING CONFIGURATION FILE |
---|
22 | # Set the most important configuration options here |
---|
23 | # Sections: general, connection, domains, processing, debug |
---|
24 | #=============================================================================== |
---|
25 | |
---|
26 | |
---|
27 | #== general == |
---|
28 | |
---|
29 | #file directories |
---|
30 | RUN_DIR = u'run' |
---|
31 | LOG_DIR = RUN_DIR |
---|
32 | PIPE_DIR = u'%s/pipes' % RUN_DIR #interprocess communication |
---|
33 | URL_DIR = u'%s/url' % RUN_DIR #URLs to download |
---|
34 | ROBOTS_DIR = u'%s/robots' % RUN_DIR #robots.txt |
---|
35 | REDIRECT_DIR = u'%s/redirect' % RUN_DIR #HTTP redirections |
---|
36 | WPAGE_DIR = u'%s/wpage' % RUN_DIR #raw encoded downloaded documents |
---|
37 | ARC_DIR = u'%s/arc' % RUN_DIR #raw encoded downloaded documents in the arc format |
---|
38 | DOC_META_DIR = u'%s/docmeta' % RUN_DIR #document metadata and extracted URLs |
---|
39 | PREVERT_DIR = u'%s/prevert' % RUN_DIR #processed documents in the prevertical format |
---|
40 | IGNORED_DIR = u'%s/ignored' % RUN_DIR #URLs excluded from further processing |
---|
41 | SAVE_DIR = u'%s/save' % RUN_DIR #saved states |
---|
42 | |
---|
43 | #max crawling time [s] (None for no limit) |
---|
44 | MAX_RUN_TIME = None |
---|
45 | |
---|
46 | #The number of downloaded data processing threads (process.py) to run, |
---|
47 | #2 is enough for very small languages, 16 is enough for huge English crawls. |
---|
48 | #Do not set to more than (<number of your processors> - 2). |
---|
49 | #A single document processor can deal with approx. 1000 MB/hour raw data |
---|
50 | #download speed (a rough estimate, depends on CPU). |
---|
51 | DOC_PROCESSOR_COUNT = 8 |
---|
52 | |
---|
53 | #Interpreter and priority of subprocesses opened by the main script |
---|
54 | DOWNLOADER_EXEC = ['python', 'crawl.py'] |
---|
55 | PROCESSOR_EXEC = ['ionice', '-c3', 'nice', '-n10', 'python', 'process.py'] |
---|
56 | |
---|
57 | |
---|
58 | #== connection == |
---|
59 | |
---|
60 | #You are responsible for properly identifying your crawler. |
---|
61 | #Some sites use the '+' in AGENT_URL to reject bots => it is polite to keep it. |
---|
62 | #The crawler's behaviour and possible masking is __your__ responsibility. |
---|
63 | AGENT = 'SpiderLing' |
---|
64 | AGENT_URL = '+http://nlp.fi.muni.cz/projects/biwec/' |
---|
65 | USER_AGENT = 'Mozilla/5.0 (compatible; SpiderLing (a SPIDER for LINGustic research); %s)' % AGENT_URL |
---|
66 | #Ignore the robots protocol if the crawler failed to fetch or parse robots.txt. |
---|
67 | IGNORE_ROBOTS_WHEN_FAILED = True #False |
---|
68 | #Give up waiting for robots.txt for a domain in time [s] (None to turn off). |
---|
69 | #Apply the ignore decision (see above). |
---|
70 | MAX_ROBOT_WAITING_TIME = 3*3600 |
---|
71 | |
---|
72 | #https support, fall back to http if disabled |
---|
73 | #Used ssl.SSLContext functions require Python >= 2.7.9. |
---|
74 | #Disabling results in ignoring webs requiring SSL. |
---|
75 | SSL_ENABLED = True |
---|
76 | |
---|
77 | #http response size constraints |
---|
78 | DOC_SIZE_MIN = 200 |
---|
79 | DOC_SIZE_MAX = 10*1024*1024 #do not set to less than 10000 |
---|
80 | |
---|
81 | #max number of open connections (max: ulimit -n), must be >> OPEN_AT_ONCE |
---|
82 | MAX_OPEN_CONNS = 8000 |
---|
83 | #max number of connections opened at once |
---|
84 | OPEN_AT_ONCE = 800 |
---|
85 | #abandon open connection when not responding after time [s] |
---|
86 | NO_ACTIVITY_TIMEOUT = 40 |
---|
87 | |
---|
88 | #max number of urls waiting to be downloaded, decrease to save the RAM |
---|
89 | MAX_URL_QUEUE = 5000000 |
---|
90 | #max number of documents waiting to be sent |
---|
91 | MAX_WPAGE_QUEUE = 5000 |
---|
92 | |
---|
93 | #period [s] of connecting to the same IP |
---|
94 | IP_CONN_INTERVAL = 1 |
---|
95 | #period [s] of connecting to the same host =~ crawl delay (20 means 3/min) |
---|
96 | HOST_CONN_INTERVAL = 20 |
---|
97 | #The crawler's behaviour is __your__ responsibility, |
---|
98 | #setting these values too low will result in banning your IP by target sites |
---|
99 | #and can lead to blocking your internet connection by your ISP! |
---|
100 | |
---|
101 | |
---|
102 | #== domains == |
---|
103 | |
---|
104 | #min docs downloaded from a domain before yield rate threshold applies |
---|
105 | MIN_DOCS_DOWNLOADED = 6 |
---|
106 | #min bytes downloaded from a domain before yield rate threshold applies |
---|
107 | MIN_BYTES_DOWNLOADED = 256*1024 |
---|
108 | #max (soft limit) cleaned non-empty docs from a domain (None for no limit) |
---|
109 | MAX_DOCS_CLEANED = None |
---|
110 | |
---|
111 | """ |
---|
112 | The yield rate threshold function sets the threshold to stop crawling a domain |
---|
113 | when it is no more efficient. |
---|
114 | """ |
---|
115 | from math import log |
---|
116 | def yield_rate_threshold(doc_count): |
---|
117 | #standard threshold: 10: 0 %, 100: 1 %, 1000: 2 %, 10000: 3 %, 100000: 4 % |
---|
118 | #return 0.01 * (log(doc_count, 10) - 1) |
---|
119 | return 0.001 * (log(doc_count, 10) - 1) #small languages |
---|
120 | #return 0.0 #threshold disabled |
---|
121 | |
---|
122 | #The web domain data (paths within the domain) is the most memory consuming |
---|
123 | #structure. A rough estimate is 500 MB - 5 GB per million domains, it depends |
---|
124 | #on the number of urls within a stored domain. The more domains the crawler has |
---|
125 | #to store, the more memory allocated. After reaching MAX_DOMS_READY domains, |
---|
126 | #additional domains in state "ready" are written into domains_oversize file |
---|
127 | #and never examined again. Therefore do not set MAX_DOMS_READY to a low value. |
---|
128 | #Domains in state "new" (IP not determined yet) consume memory in the same way. |
---|
129 | #The size of that queue is not limited. Therefore your free memory can be |
---|
130 | #depleted by "new" domains despite you limited "ready" domains. |
---|
131 | |
---|
132 | #max number of domains ready for crawling kept in memory |
---|
133 | #1M is enough for small languages (expect ~2 GB RAM required), |
---|
134 | #50M should be enough for huge English crawls (allocates up to ~200 GB RAM). |
---|
135 | MAX_DOMS_READY = 5*1000*1000 |
---|
136 | #max domains selected for crawling at the same time |
---|
137 | MAX_DOMS_SELECTED = 500*1000 |
---|
138 | #max time [s] no data received from a selected domain |
---|
139 | MAX_IDLE_TIME = 6*3600 #1800 |
---|
140 | #period [s] of domain refresh (adding new, removing old, evaluating efficiency) |
---|
141 | UPDATE_DOMAINS_PERIOD = 600 |
---|
142 | #max number of domains with the same IP selected for crawling at the same time |
---|
143 | MAX_IP_REPEAT = 50 |
---|
144 | |
---|
145 | #Number of urls sent at once by the scheduler to the crawler. Decrease the |
---|
146 | #minimum value to 10 if crawling a small language. Set the minimum starting |
---|
147 | #value to 1 if crawling a single web domain. Increase the maximum value to |
---|
148 | #allow wider domain coverage in the downloader's url queue. |
---|
149 | #Must be < MAX_DOMS_SELECTED. |
---|
150 | MIN_URL_SELECT_START = 100 |
---|
151 | MIN_URL_SELECT = 1000 |
---|
152 | MAX_URL_SELECT = 100000 |
---|
153 | #Decrease to improve domain variety and thus crawling speed, useful only when |
---|
154 | #there is more than ~500 domains crawled concurrently (big languages) |
---|
155 | MAX_URL_SELECT_PER_DOMAIN = 100 |
---|
156 | |
---|
157 | #hostname -> IP mapping file (two space separated columns), None by default |
---|
158 | #Use recent DNS mappings only since hosts tend to move occasionally. |
---|
159 | DNS_TABLE_PATH = u'url/dns_table' #None |
---|
160 | |
---|
161 | |
---|
162 | #== processing == |
---|
163 | |
---|
164 | #Set the languages of documents to recognise, e.g. ['Norwegian', 'Danish']. |
---|
165 | #This enables recognising unwanted languages similar to target languages. |
---|
166 | #Requires |
---|
167 | # - a plaintext in that languages: ./util/lang_samples/Norwegian |
---|
168 | # - a jusText wordlist for that languages: ./util/justext_wordlists/Norwegian |
---|
169 | # - a chared model for that languages: ./util/chared_models/Norwegian |
---|
170 | #The most similar language is selected (if below lang diff thresholds). |
---|
171 | LANGUAGES = ['Norwegian_Bokmal', 'Norwegian_Nynorsk', 'Danish', 'English'] |
---|
172 | #Set which of recognised languages to accept, e.g. ['Norwegian']. |
---|
173 | LANGUAGES_ACCEPT = ['Norwegian_Bokmal', 'Norwegian_Nynorsk'] |
---|
174 | #max allowed difference to the lang model for document and paragraphs |
---|
175 | #(0.0 = same as the model); it does not work well with paragraphs, the reported |
---|
176 | #similarity is usually low |
---|
177 | LANG_DIFF_THRESHOLD_DOC = 0.7 |
---|
178 | LANG_DIFF_THRESHOLD_PAR = 0.99 |
---|
179 | #disable encoding detection and force an encoding (e.g. 'utf-8'), None by default |
---|
180 | FORCE_ENCODING = None |
---|
181 | #switch to unigram models (useful for Chinese, Japanese, Korean) |
---|
182 | UNIGRAM_MODELS = False |
---|
183 | |
---|
184 | #Allowed top level domains regexp -- crawling is restricted to these, e.g.: |
---|
185 | #English + general web: u'\.(?:uk|us|au|ca|com|org|net|info|biz|edu|name)$' |
---|
186 | #German + general web: u'\.(?:de|at|ch|com|org|net|info|eu|edu|name)$' |
---|
187 | #Czech only: u'\.cz$' |
---|
188 | #no restriction (all): u'' |
---|
189 | from re import compile |
---|
190 | TLD_WHITELIST_RE = compile(u'') |
---|
191 | #Country/language native TLDs, e.g. u'\.(?:uk|us|au|ca)$' or u'\.cz$' or u''. |
---|
192 | #If EXTRACT_EMPTY_PAGE_EXTERNAL_LINKS == True, a matched cross domain link |
---|
193 | #will be followed. u'' is the default and results in following all links. |
---|
194 | TLD_NATIVE_RE = compile(u'\.no$') |
---|
195 | #TLD blacklist > TLD whitelist |
---|
196 | TLD_BLACKLIST_RE = compile(u'\.(?:da|se)$') |
---|
197 | |
---|
198 | #file containing unwanted web domains (one per line), None by default |
---|
199 | DOMAIN_BLACKLIST_PATH = u'url/domain_blacklist' #None |
---|
200 | #file containing allowed web domains (one per line), None by default |
---|
201 | #priority: blacklist > whitelist > allowed TLDs |
---|
202 | DOMAIN_WHITELIST_PATH = None |
---|
203 | |
---|
204 | #extract links from web pages containing no clean text |
---|
205 | #useful to decrease crawling from domains in unwanted languages |
---|
206 | #which is dealt with by the trigram model |
---|
207 | #disabling this may increase crawling efficiency (yield rate) |
---|
208 | #at the cost of omitting some web domains (or visiting them later) |
---|
209 | EXTRACT_EMPTY_PAGE_INTERNAL_LINKS = True |
---|
210 | #extract links from pages with no clean text going out from the domain |
---|
211 | #priority: EXTRACT_EMPTY_PAGE_LINKS > EXTRACT_EMPTY_PAGE_EXTERNAL_LINKS |
---|
212 | EXTRACT_EMPTY_PAGE_EXTERNAL_LINKS = False |
---|
213 | |
---|
214 | #justext paragraph heuristic configuration |
---|
215 | #character count < length_low => bad or short |
---|
216 | JUSTEXT_LENGTH_LOW = 50 #justext default = 70 |
---|
217 | #character count > length_high => good |
---|
218 | JUSTEXT_LENGTH_HIGH = 100 #justext default = 200 |
---|
219 | #number of words frequent in the language >= stopwords_low => neargood |
---|
220 | JUSTEXT_STOPWORDS_LOW = 0.2 #justext default = 0.3 |
---|
221 | #number of words frequent in the language >= stopwords_high => good or neargood |
---|
222 | JUSTEXT_STOPWORDS_HIGH = 0.3 #justext default = 0.32 |
---|
223 | #density of link words (words inside the <a> tag) > max_link_density => bad |
---|
224 | JUSTEXT_MAX_LINK_DENSITY = 0.4 #justext default = 0.2 |
---|
225 | #short/near-good headings in the distance of <value> characters before a good paragraph => good |
---|
226 | JUSTEXT_MAX_HEADING_DISTANCE = 200 #justext default = 200 |
---|
227 | |
---|
228 | #path to justext wordlists, use None for the justext default |
---|
229 | #JUSTEXT_WORDLIST_DIR/LANGUAGE must exist for all LANGUAGES |
---|
230 | JUSTEXT_WORDLIST_DIR = u'util/justext_wordlists' |
---|
231 | #path to chared models |
---|
232 | #CHARED_MODEL_DIR/LANGUAGE must exist for all LANGUAGES |
---|
233 | CHARED_MODEL_DIR = u'util/chared_models' |
---|
234 | #allow conversion of binary formats (pdf, doc) to text (not tested, |
---|
235 | #the conversion is not reliable, may result in the presence of garbage text |
---|
236 | CONVERSION_ENABLED = False |
---|
237 | |
---|
238 | |
---|
239 | #== debug == |
---|
240 | |
---|
241 | import logging |
---|
242 | LOG_LEVEL = logging.DEBUG #DEBUG|INFO|WARNING|ERROR |
---|
243 | LOG_FORMAT = u'%(asctime)-15s %(threadName)s %(message)s' |
---|
244 | #period [s] of writing out some debug info, 0 to disable |
---|
245 | INFO_PERIOD = 3600 |
---|
246 | #period [s] of saving state of the crawler (all domain data), 0 to disable |
---|
247 | SAVE_PERIOD = 3600*24 |
---|
248 | #log stream buffering (0 = none, 1 = line, -1 = fully/system, n = n bytes) |
---|
249 | LOG_BUFFERING = 1 |
---|