From 4c71eec3cd5f5f36c1cdc6d2284f6dd93facc193 Mon Sep 17 00:00:00 2001
From: Niclas Dobbertin <niclas.dobbertin@mailbox.org>
Date: Wed, 1 Nov 2023 11:20:45 +0100
Subject: put analysis functions into own py file

---
 experiment/analysis/analysis.org | 57 ++++++----------------------------------
 1 file changed, 8 insertions(+), 49 deletions(-)

(limited to 'experiment/analysis/analysis.org')

diff --git a/experiment/analysis/analysis.org b/experiment/analysis/analysis.org
index 7f6a58d..e726046 100644
--- a/experiment/analysis/analysis.org
+++ b/experiment/analysis/analysis.org
@@ -4,10 +4,11 @@
 * Imports
 #+begin_src python :results none
 import pandas as pd
-import pickle
 from pathlib import Path
 from pprint import pprint
 
+import tools
+
 #+end_src
 
 * Constants
@@ -18,13 +19,6 @@ procedures = ["1", "2", "3", "4", "5", "6", "overall"]
 #+end_src
 
 * Import Data
-#+begin_src python :results none
-def unpickle(pkl):
-    with open(pkl, "rb") as f:
-        data = pickle.load(f)
-    return data
-#+end_src
-
 ** Conditions
 #+begin_src python
 conditions = [x.stem for x in data_path.iterdir() if x.is_dir()]
@@ -35,59 +29,24 @@ conditions
 | random | fixed | blocked |
 
 ** Data
-#+begin_src python
+#+begin_src python :results none
 data = {}
 for condition in conditions:
     data[condition] = {}
     for vp in (data_path / condition).iterdir():
-        data[condition][vp.stem] = unpickle(vp / "vp.pkl")
-#+end_src
+        data[condition][vp.stem] = tools.unpickle(vp / "vp.pkl")
 
-#+RESULTS:
-: None
+data_train, data_test = tools.train_test_split(data)
+#+end_src
 
 * Basic statistics
 ** Total percent correct
 To find out how well VP solved the tasked, we calculate the accuracy for train
 and test phase.
 
-#+begin_src python :results none
-def count_correct(vp, trials):
-    trials_correct = {}
-    for proc in procedures:
-        trials_correct[proc] = 0
-    for sample in trials:
-        for proc in vp[sample]["procedure_order"]:
-            vp_ans = vp[sample][proc]["answer"]
-            for c in vp_ans:
-                if not c.isdigit():
-                    vp_ans = vp_ans.replace(c, "")
-            vp_ans = int(vp_ans)
-            if vp_ans == vp[sample]["water_sample"][proc][0]:
-                trials_correct[proc] += 1
-    return trials_correct
-#+end_src
-
-#+begin_src python :results none
-def total_accuracy(vp):
-    train_total = len(train) * len(vp[train[0]]["procedure_order"])
-    test_total = len(test) * len(vp[test[0]]["procedure_order"])
-
-    acc_train = count_correct(vp, train)
-    acc_test = count_correct(vp, test)
-
-    acc_train = sum([acc_train[x] for x in acc_train.keys()]) / train_total
-    acc_test = sum([acc_test[x] for x in acc_test.keys()]) / test_total
-
-    return acc_train, acc_test
-#+end_src
-
 #+begin_src python
-train = [x for x in vp.keys() if "train" in x]
-test = [x for x in vp.keys() if "test" in x]
-
 condition = "random"
-df = pd.DataFrame([total_accuracy(data[condition][vp]) for vp in data[condition].keys()], index=data[condition].keys(), columns=["train", "test"])
+df = pd.DataFrame([tools.total_accuracy(data[condition][vp], procedures) for vp in data[condition].keys()], index=data[condition].keys(), columns=["train", "test"])
 df
 #+end_src
 
@@ -116,7 +75,7 @@ To investigate, we look at the per procedure accuracy per subject.
 #+begin_src python
 condition = "random"
 proc_accs = [
-    count_correct(data[condition][vp], data[condition][vp].keys())
+    tools.count_correct(data[condition][vp], data[condition][vp].keys(), procedures)
     for vp in data[condition].keys()
 ]
 for vp in proc_accs:
-- 
cgit v1.2.3