data_processing.py

# -*- coding: utf-8 -*-
"""data_processing.ipynb

Automatically generated by Colab.

Original file is located at
    https://colab.research.google.com/drive/1Oz1QL0mD9g3lVBgtmqHa-QiwwIJ2JaX5
"""

import pandas as pd
import numpy as np
import os
from zipfile import ZipFile
import re
import json
import io
from PIL import Image, ImageFile
ImageFile.LOAD_TRUNCATED_IMAGES = True

from google.colab import drive
drive.mount('/content/drive')

path = "/content/drive/MyDrive/Duke/aphantasia_drawing_project/"

data_path = os.path.join(path,"data",'drawing_experiment')



df = pd.read_excel(data_path+"/questionnaire-data.xlsx", header=2)

df["vviq_score"] = np.sum(df.filter(like = "vviq"), axis = 1)
df["osiq_score"] = np.sum(df.filter(like = "osiq"), axis = 1)
df["treatment"] = np.where(df.vviq_score > 40, "control", "aphantasia")

df = df.rename(columns={
    "Sub ID": "sub_id",
    df.columns[5]: "art_ability",
    df.columns[6]: "art_experience",
    df.columns[9]: "difficult",
    df.columns[10]: "diff_explanation"
})

df.columns = df.columns.str.lower()

df = df.drop(df.filter(like="unnamed").columns, axis = 1)
df = df.drop(df.filter(regex="(vviq|osiq)\d+").columns, axis = 1)

df[df.columns[df.dtypes == "object"]] = df[df.columns[df.dtypes == "object"]].astype("string")

df = df.replace([np.nan,pd.NA, "nan","na","NA","n/a","N/A","N/a"], None)

df.set_index('sub_id', inplace=True)



actual_image_path = os.path.join(data_path,"Stimuli","Images")



actual_images = {}
for image_file in os.listdir(actual_image_path):
  img_path = os.path.join(actual_image_path, image_file)
  actual_images[image_file.removesuffix(".jpg")] = Image.open(img_path)



key_map = {
    'high_sun_ajwbpqrwvknlvpeh': 'kitchen',
    'low_sun_acqsqjhtcbxeomux': 'bedroom',
    "low_sun_byqgoskwpvsbllvy":"livingroom"
    }

for old_key, new_key in key_map.items():
  actual_images[new_key] = actual_images.pop(old_key)

aphantasia_drawings_path = os.path.join(data_path,"Drawings","Aphantasia")
control_drawings_path = os.path.join(data_path,"Drawings","Control")



directories = {
    "Aphantasia": aphantasia_drawings_path,
    "Control": control_drawings_path
}



aphantasia_subs = {i: "Aphantasia" for i in os.listdir(directories["Aphantasia"]) if "sub" in i}
control_subs = {i: "Control" for i in os.listdir(directories["Control"]) if "sub" in i}
sub_treatment_key = {**aphantasia_subs, **control_subs}



def get_sub_files(sub):
  treatment_group = sub_treatment_key[sub]
  directory = directories[treatment_group]
  pattern = re.compile("^.*" + sub + "-[a-z]{3}\d-(kitchen|livingroom|bedroom).*")
  sub_files = os.listdir(os.path.join(directory, sub))
  files_needed = {'mem1',"mem2",'mem3','pic1','pic2','pic3'}

  sub_key = {}
  for f in sub_files:
    if pattern.match(f):
      main_path = os.path.join(directory, sub, f)
      draw_type = f.split("-")[1]
      label = f.split("-")[2].removesuffix(".jpg")
      alt_path = os.path.join(directory, sub, "-".join([sub, draw_type]) + ".jpg")
      try:
        img = Image.open(main_path)
      except:
        img = Image.open(alt_path)
      sub_key[draw_type] = {
          "label": label,
          "drawing": img
      }

  unknown_drawings = files_needed - sub_key.keys()
  if unknown_drawings:
    for unk in unknown_drawings:
      path = os.path.join(directory, sub, "-".join([sub, unk]) + ".jpg")
      try:
        img = Image.open(path)
      except:
        img = "No Image"
      sub_key[unk] = {
          "label": "unknown",
          "drawing": img
      }
  return sub_key





subject_data = {}
for sub in iter(sub_treatment_key):
  subject_data[sub] = get_sub_files(sub)

def is_image_blank(image):
  if image.mode != 'RGB':
    image = image.convert('RGB')
  pixels = list(image.getdata())
  return all(pixel == (255, 255, 255) for pixel in pixels)

for sub in iter(subject_data):
  dat = subject_data[sub]
  for key in dat.keys():
    if is_image_blank(dat[key]["drawing"]):
      dat[key]["label"] = "blank"



subs_missing_labels = {}
for sub in iter(subject_data):
  dat = subject_data[sub]
  for key in dat.keys():
    if "unknown" in dat[key]["label"]:
      if sub not in subs_missing_labels:
        subs_missing_labels[sub] = []
      subs_missing_labels[sub].append(key)

subs_missing_labels

subject_data["sub8"]["pic3"]["label"] = "livingroom"
subject_data["sub6"]["pic3"]["label"] = "bedroom"
subject_data["sub6"]["pic1"]["label"] = "kitchen"



def clean_sub_dat(sub):
  id = int(sub[3:])
  treatment_group = sub_treatment_key[sub]
  if id in df.index:
    demographics_dict = df.loc[id].to_dict()
  else:
    demographics_dict = {}
  demographics_dict.pop("treatment",None)
  drawings = {
    "bedroom": {},
    "kitchen": {},
    "livingroom": {}
    }
  for draw_type, draw_data in subject_data[sub].items():
    t = "memory" if draw_type[:-1] == "mem" else "perception"
    for d in drawings.keys():
      if draw_data["label"] == d:
        drawings[d][t] = draw_data["drawing"]

  return {
      "subject_id": id,
      "treatment": treatment_group,
      "demographics": demographics_dict,
      "drawings": drawings,
      "image": actual_images
  }



full_data = []
for s in subject_data.keys():
  full_data.append(clean_sub_dat(s))

"""160,161,162 removed, they dont have images"""



full_df = pd.json_normalize(full_data)



def image_to_byt(img, size=(224, 224)):
  if  pd.isna(img):
    return None
  img_resized = img.resize(size)
  img_byte_arr = io.BytesIO()
  img_resized.save(img_byte_arr, format='PNG')
  return img_byte_arr.getvalue()

drawing_columns = [col for col in full_df.columns if "drawings" in col or "image" in col]



for col in drawing_columns:
  full_df[col] = full_df[col].apply(image_to_byt)

def safe_convert_to_int(value):
  try:
    return int(value)
  except (ValueError, TypeError):
    return -99

col_to_process = [
    "demographics.age",
    "demographics.art_ability",
    "demographics.vviq_score",
    "demographics.osiq_score"
]

for col in col_to_process:
    full_df[col] = full_df[col].apply(safe_convert_to_int)

full_data_path = os.path.join(path, "data","aphantasia_data.parquet")

full_df.to_parquet(full_data_path, index=False)