There are a few different assertion modules (polarity, uncertainty, subject, conditional, historyOf). The ClearTK/Liblinear-based model can train and eval all those modules in the same class:
/ctakes-assertion/src/main/java/org/apache/ctakes/assertion/eval/AssertionEvaluation.java If you look in the resources/launch subdirectory there are eclipse launch configurations for preprocessing (reading from gold standard annotations and writing xmis) and for different evaluations (e.g., cross-validation). Even if you're not using eclipse that will help you understand the arguments that class takes. The trickiest part for adapting to your data will probably be reading the gold standard. If you look in the preprocess() method of the eval class above it makes calls to different corpus readers depending on corpus argument. So unless you use the same tool (Knowtator) and schema you will probably need to write your own corpus reader. But you can look there for some examples of how to do so. Sorry, but all the code in that module could use some cleaning. Tim ________________________________ From: Leander Melms <[email protected]> Sent: Tuesday, October 25, 2016 10:25 AM To: [email protected] Subject: Re: cTakes Models Yes, I succcessfully trained a chunker and sentence detection model. However, I'm still not sure how training of the lemmatizer and the semantic role labeler, relation extractor, historyOf, conditional (cTakes assertion), polarity, subject, uncertainity, sideEffect libsvm models has been performed. Any documentation or training scripts are hightly appreciated. Thanks Leander Melms On 25 Oct 2016, at 15:47, lewis john mcgibbney <[email protected]<mailto:[email protected]>> wrote: Hi Leander, Hve you seen the fllowing documentation? https://cwiki.apache.org/confluence/display/CTAKES/cTAKES+3.2+Dictionaries+and+Models<https://urldefense.proofpoint.com/v2/url?u=https-3A__cwiki.apache.org_confluence_display_CTAKES_cTAKES-2B3.2-2BDictionaries-2Band-2BModels&d=DQMFAg&c=qS4goWBT7poplM69zy_3xhKwEW14JZMSdioCoppxeFU&r=Heup-IbsIg9Q1TPOylpP9FE4GTK-OqdTDRRNQXipowRLRjx0ibQrHEo8uYx6674h&m=Vk0sXyyPtZL4Er9pJbyjTS_ZLhia99950n_cFi9BySc&s=Pf9U8Evx-wEzIny7Vp4NZ8c2vyGAG--gzX1duizDB4Y&e=> ? On Tue, Oct 25, 2016 at 12:29 AM, <[email protected]<mailto:[email protected]>> wrote: From: Leander Melms <[email protected]<mailto:[email protected]>> To: [email protected]<mailto:[email protected]> Cc: Date: Tue, 25 Oct 2016 09:21:21 +0200 Subject: cTakes Models Dear cTAKES Dev Community, I'm a doctoral candidate at ZUSE Marburg, an institution for differential diagnosis of rare diseases, and I'm currently working with cTAKES. I'd like to train various models, including but not limited to the pependency parser (mayo-en-dep), the lemmatizer and the semantic role labeler, relation extractor, historyOf, conditional (cTakes assertion), polarity, subject, uncertainity, sideEffect libsvm models l on a German text corpus. Unfortunately, I haven't found much information on how training was performed yet. Any hints towards how the training data was labeled and formatted or even training scripts would be greatly appreciated. We have a mid-sized text corpus of patient discharge summaries and would be excited to integrate cTAKES in differetial diagnoses. Best regards Leander Melms Doctoral candidate and medical student at Philipps-University Marburg [email protected]<mailto:[email protected]>-marburg.de<https://urldefense.proofpoint.com/v2/url?u=http-3A__marburg.de_&d=DQMFAg&c=qS4goWBT7poplM69zy_3xhKwEW14JZMSdioCoppxeFU&r=Heup-IbsIg9Q1TPOylpP9FE4GTK-OqdTDRRNQXipowRLRjx0ibQrHEo8uYx6674h&m=Vk0sXyyPtZL4Er9pJbyjTS_ZLhia99950n_cFi9BySc&s=2BT1NAwxQunl3TH_zR6nYuNgXNPinMgennKsrn5ef3M&e=> -- http://home.apache.org/~lewismc/<https://urldefense.proofpoint.com/v2/url?u=http-3A__home.apache.org_-257Elewismc_&d=DQMFAg&c=qS4goWBT7poplM69zy_3xhKwEW14JZMSdioCoppxeFU&r=Heup-IbsIg9Q1TPOylpP9FE4GTK-OqdTDRRNQXipowRLRjx0ibQrHEo8uYx6674h&m=Vk0sXyyPtZL4Er9pJbyjTS_ZLhia99950n_cFi9BySc&s=yq6Z4bBdJ07YzmbxOZx-Yr5MWsu9slDh2mANcYEx2ws&e=> @hectorMcSpector http://www.linkedin.com/in/lmcgibbney<https://urldefense.proofpoint.com/v2/url?u=http-3A__www.linkedin.com_in_lmcgibbney&d=DQMFAg&c=qS4goWBT7poplM69zy_3xhKwEW14JZMSdioCoppxeFU&r=Heup-IbsIg9Q1TPOylpP9FE4GTK-OqdTDRRNQXipowRLRjx0ibQrHEo8uYx6674h&m=Vk0sXyyPtZL4Er9pJbyjTS_ZLhia99950n_cFi9BySc&s=N5ZNEMGVcd8aicPRIlviSHqRPXQJNfY-KqPTe0vJnbI&e=>
