{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.7.12","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"import pandas as pd\nimport numpy as np\nimport seaborn as sns\nimport matplotlib.pyplot as plt\n%matplotlib inline\nimport os\n\nsns.set_style('darkgrid')\nplt.style.use('seaborn-notebook')","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:48.214140Z","iopub.execute_input":"2023-09-08T17:55:48.215141Z","iopub.status.idle":"2023-09-08T17:55:49.369081Z","shell.execute_reply.started":"2023-09-08T17:55:48.215044Z","shell.execute_reply":"2023-09-08T17:55:49.368153Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#read in train and test csv files to calculate descriptive stats and characteristics\n\ntrain = pd.read_csv('../input/siim-isic-melanoma-classification/train.csv')\n\ntest = pd.read_csv('../input/siim-isic-melanoma-classification/test.csv')","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:49.370841Z","iopub.execute_input":"2023-09-08T17:55:49.371351Z","iopub.status.idle":"2023-09-08T17:55:49.497243Z","shell.execute_reply.started":"2023-09-08T17:55:49.371320Z","shell.execute_reply":"2023-09-08T17:55:49.495976Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train.head()","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:49.499088Z","iopub.execute_input":"2023-09-08T17:55:49.499809Z","iopub.status.idle":"2023-09-08T17:55:49.537751Z","shell.execute_reply.started":"2023-09-08T17:55:49.499764Z","shell.execute_reply":"2023-09-08T17:55:49.535795Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train.shape","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:49.546032Z","iopub.execute_input":"2023-09-08T17:55:49.549914Z","iopub.status.idle":"2023-09-08T17:55:49.562075Z","shell.execute_reply.started":"2023-09-08T17:55:49.549851Z","shell.execute_reply":"2023-09-08T17:55:49.560505Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train.info()","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:49.567238Z","iopub.execute_input":"2023-09-08T17:55:49.568250Z","iopub.status.idle":"2023-09-08T17:55:49.634979Z","shell.execute_reply.started":"2023-09-08T17:55:49.568195Z","shell.execute_reply":"2023-09-08T17:55:49.633453Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train.describe()","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:49.636875Z","iopub.execute_input":"2023-09-08T17:55:49.637741Z","iopub.status.idle":"2023-09-08T17:55:49.678203Z","shell.execute_reply.started":"2023-09-08T17:55:49.637693Z","shell.execute_reply":"2023-09-08T17:55:49.676484Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"ax1 = sns.countplot(data=train, x='anatom_site_general_challenge')\nax1.set_title('Distribution of Anatomical Sites Training Set')\nax1.tick_params(axis='x', labelrotation = 45, labelsize = 12)\n\n#note: largest category by far is the torso, oral/genital is the least common category","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:49.680882Z","iopub.execute_input":"2023-09-08T17:55:49.682435Z","iopub.status.idle":"2023-09-08T17:55:50.161090Z","shell.execute_reply.started":"2023-09-08T17:55:49.682343Z","shell.execute_reply":"2023-09-08T17:55:50.160250Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"ax2 = sns.countplot(data=train, x='sex')\nax2.set_title('Distribution of Gender')\n\n#note: gender is pretty evenly distributed amongst the image dataset","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:50.162423Z","iopub.execute_input":"2023-09-08T17:55:50.163133Z","iopub.status.idle":"2023-09-08T17:55:50.438589Z","shell.execute_reply.started":"2023-09-08T17:55:50.163099Z","shell.execute_reply":"2023-09-08T17:55:50.437572Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"ax3 = sns.countplot(data=train, x='diagnosis')\nax3.tick_params(axis='x', labelrotation = 45, labelsize = 12)\nax3.set_title('Distribution of Diagnosis')\n\n#note: the vast majority of the images have no associated diagnosis, some have the nevus diagnosis, and very few have the melanoma diagnosis. \n#Melanoma is the target we're actually looking for.","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:50.440253Z","iopub.execute_input":"2023-09-08T17:55:50.440961Z","iopub.status.idle":"2023-09-08T17:55:50.834712Z","shell.execute_reply.started":"2023-09-08T17:55:50.440916Z","shell.execute_reply":"2023-09-08T17:55:50.833446Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"ax4 = sns.countplot(data=train, x='benign_malignant')\nax4.tick_params(axis='x', labelrotation = 45, labelsize = 12)\nax4.set_title('Distribution of Diagnosis')\n\n#note: malignant here is defined as a have a \"melanoma\" disease label","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:50.840495Z","iopub.execute_input":"2023-09-08T17:55:50.841894Z","iopub.status.idle":"2023-09-08T17:55:51.127240Z","shell.execute_reply.started":"2023-09-08T17:55:50.841842Z","shell.execute_reply":"2023-09-08T17:55:51.126272Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"ax5 = sns.histplot(data=train, x='age_approx', bins=18)\nax5.tick_params(axis='x', labelrotation = 45, labelsize = 12)\nax5.set_title('Distribution of Age')\n\n#age follows a normal distribution","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:51.128962Z","iopub.execute_input":"2023-09-08T17:55:51.129359Z","iopub.status.idle":"2023-09-08T17:55:51.522215Z","shell.execute_reply.started":"2023-09-08T17:55:51.129323Z","shell.execute_reply":"2023-09-08T17:55:51.521244Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print('n rows where target != benign_malignant: {}'.format(len(train.loc[(train['target'] == 1) & (train['benign_malignant'] != 'malignant')])))\nprint('n rows where benign_malignant != melanoma diagnosis: {}'.format(len(train.loc[(train['diagnosis'] != 'melanoma') & (train['benign_malignant'] == 'malignant')])))\nprint('n rows where target != benign_malignant: {}'.format(len(train.loc[(train['target'] == 1) & (train['benign_malignant'] != 'malignant')])))\n\n#my assumptions hold--this dataset is labeled based off of the 'melanoma' diagnosis, \n#and a positive value in the target column indicates a malignant melanoma tumor","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:51.523896Z","iopub.execute_input":"2023-09-08T17:55:51.524257Z","iopub.status.idle":"2023-09-08T17:55:51.560472Z","shell.execute_reply.started":"2023-09-08T17:55:51.524223Z","shell.execute_reply":"2023-09-08T17:55:51.559185Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#print the number of missing values from each col in train\nfor col in train.columns:\n    print(col + ' missing values: ' + str(train[col].isna().sum()))\n    ","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:51.562031Z","iopub.execute_input":"2023-09-08T17:55:51.565752Z","iopub.status.idle":"2023-09-08T17:55:51.601059Z","shell.execute_reply.started":"2023-09-08T17:55:51.565701Z","shell.execute_reply":"2023-09-08T17:55:51.599802Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#Pull out just the observations that are positive for the target condition\nmalignant = train[train['target'] == 1]\nprint(malignant.head())\nprint(malignant.shape)","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:51.602770Z","iopub.execute_input":"2023-09-08T17:55:51.604259Z","iopub.status.idle":"2023-09-08T17:55:51.616167Z","shell.execute_reply.started":"2023-09-08T17:55:51.604218Z","shell.execute_reply":"2023-09-08T17:55:51.614443Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"for col in malignant.columns:\n    print(col + ' missing values: ' + str(malignant[col].isna().sum()))\n","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:51.617665Z","iopub.execute_input":"2023-09-08T17:55:51.618004Z","iopub.status.idle":"2023-09-08T17:55:51.632459Z","shell.execute_reply.started":"2023-09-08T17:55:51.617972Z","shell.execute_reply":"2023-09-08T17:55:51.631431Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#fill missing anatom site values with \"unknown or other\"\ntrain.anatom_site_general_challenge.fillna('other or unknown', inplace=True)\n\n#verify there are no more missing values in that column\nprint('anatom_site_general_challenge' + ' missing values: ' + str(train['anatom_site_general_challenge'].isna().sum()))\n\n#plot the new distribution\nax = sns.countplot(data=train, x='anatom_site_general_challenge')\n#train.anatom_site_general_challenge.hist()\nax.set_title('Distribution of Anatomical Sites Training Set')\nax.tick_params(axis='x', labelrotation = 45, labelsize = 12)","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:51.634302Z","iopub.execute_input":"2023-09-08T17:55:51.635489Z","iopub.status.idle":"2023-09-08T17:55:52.019932Z","shell.execute_reply.started":"2023-09-08T17:55:51.635436Z","shell.execute_reply":"2023-09-08T17:55:52.019050Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# **Test Set exploration**","metadata":{}},{"cell_type":"code","source":"test.head()","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:52.021307Z","iopub.execute_input":"2023-09-08T17:55:52.022255Z","iopub.status.idle":"2023-09-08T17:55:52.037791Z","shell.execute_reply.started":"2023-09-08T17:55:52.022219Z","shell.execute_reply":"2023-09-08T17:55:52.036573Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test.shape","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:52.039078Z","iopub.execute_input":"2023-09-08T17:55:52.039438Z","iopub.status.idle":"2023-09-08T17:55:52.047057Z","shell.execute_reply.started":"2023-09-08T17:55:52.039378Z","shell.execute_reply":"2023-09-08T17:55:52.045815Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#print the number of missing values from each col in train\nfor col in test.columns:\n    print(col + ' missing values: ' + str(test[col].isna().sum()))\n","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:52.048763Z","iopub.execute_input":"2023-09-08T17:55:52.049223Z","iopub.status.idle":"2023-09-08T17:55:52.068041Z","shell.execute_reply.started":"2023-09-08T17:55:52.049176Z","shell.execute_reply":"2023-09-08T17:55:52.066799Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#how many images are associated with each patient?\nindividuals_count = train.groupby('patient_id').count()\n\nindividuals_count.head()","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:52.071923Z","iopub.execute_input":"2023-09-08T17:55:52.072306Z","iopub.status.idle":"2023-09-08T17:55:52.128047Z","shell.execute_reply.started":"2023-09-08T17:55:52.072273Z","shell.execute_reply":"2023-09-08T17:55:52.126907Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#show the disribution of image counts\nfig, ax = plt.subplots(figsize=(25,10))\nax = sns.histplot(data=individuals_count, x='image_name')\nax.set_title('Images per Patient')\nax.set_xlabel('Unique Images per Patient')\nax.set_ylabel('Number of Patients')\nax.set_xticks(ticks=range(0,120,10))\n\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:52.130441Z","iopub.execute_input":"2023-09-08T17:55:52.130791Z","iopub.status.idle":"2023-09-08T17:55:52.608608Z","shell.execute_reply.started":"2023-09-08T17:55:52.130759Z","shell.execute_reply":"2023-09-08T17:55:52.607508Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print('the mean number of images per patient is: {}'.format(round(individuals_count.image_name.mean(),2)))\nprint('the median number of images per patient is: {}'.format(round(individuals_count.image_name.median(),2)))\nprint('the standard deviation of the number of images per patient is: {}'.format(round(individuals_count.image_name.std(),2)))","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:52.610541Z","iopub.execute_input":"2023-09-08T17:55:52.611619Z","iopub.status.idle":"2023-09-08T17:55:52.620184Z","shell.execute_reply.started":"2023-09-08T17:55:52.611578Z","shell.execute_reply":"2023-09-08T17:55:52.619022Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#check out the distribution of positive cases grouped by patient\nindividuals_sum = train.groupby('patient_id').sum()\n\nprint(individuals_sum.target.describe())\n\n#fig, ax = plt.subplots(figsize=(18,10))\nax = sns.histplot(data=individuals_sum, x='target', bins=10)\nax.set_title('Quantity of Malignant Tumors per Patient')\nax.set_xlabel('Sum of Positives')\nax.set_ylabel('Number of Patients')\n#ax.set_xticks(ticks=range(0,20))\n\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:52.621644Z","iopub.execute_input":"2023-09-08T17:55:52.622194Z","iopub.status.idle":"2023-09-08T17:55:52.984689Z","shell.execute_reply.started":"2023-09-08T17:55:52.622155Z","shell.execute_reply":"2023-09-08T17:55:52.983730Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"**Another pass at missing sex and age values**","metadata":{}},{"cell_type":"code","source":"#let's pull out individuals from individuals_count who have fewer counts of sex or age_approx than image_name\nindividuals_count.loc[(individuals_count['sex'] < individuals_count['image_name']) | (individuals_count['age_approx'] < individuals_count['image_name'])]","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:52.986424Z","iopub.execute_input":"2023-09-08T17:55:52.987152Z","iopub.status.idle":"2023-09-08T17:55:53.004195Z","shell.execute_reply.started":"2023-09-08T17:55:52.987107Z","shell.execute_reply":"2023-09-08T17:55:53.002810Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"pats = ['IP_0550106','IP_5205991', 'IP_9835712']\nindividuals_sum.reset_index(inplace=True)\n\n\nindividuals_sum.query('patient_id in @pats')","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:53.006067Z","iopub.execute_input":"2023-09-08T17:55:53.006488Z","iopub.status.idle":"2023-09-08T17:55:53.027657Z","shell.execute_reply.started":"2023-09-08T17:55:53.006452Z","shell.execute_reply":"2023-09-08T17:55:53.026454Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print('percentage of missing patient data= {}%'.format(round((3+48+17)/len(train)*100,2)))","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:53.029237Z","iopub.execute_input":"2023-09-08T17:55:53.029631Z","iopub.status.idle":"2023-09-08T17:55:53.036141Z","shell.execute_reply.started":"2023-09-08T17:55:53.029596Z","shell.execute_reply":"2023-09-08T17:55:53.035114Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#drop all rows from the train dataframe that are associated with the 3 patients identified above.\ntrain.drop(train[(train['patient_id'] == 'IP_0550106') | (train['patient_id'] == 'IP_5205991') |(train['patient_id'] == 'IP_9835712')].index, inplace=True)","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:53.037495Z","iopub.execute_input":"2023-09-08T17:55:53.038434Z","iopub.status.idle":"2023-09-08T17:55:53.074102Z","shell.execute_reply.started":"2023-09-08T17:55:53.038368Z","shell.execute_reply":"2023-09-08T17:55:53.072646Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#check for any more missing values:\nfor col in train.columns:\n    print(col + ' missing values: ' + str(train[col].isna().sum()))","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:53.082859Z","iopub.execute_input":"2023-09-08T17:55:53.083266Z","iopub.status.idle":"2023-09-08T17:55:53.117553Z","shell.execute_reply.started":"2023-09-08T17:55:53.083227Z","shell.execute_reply":"2023-09-08T17:55:53.116233Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"**Relationships between positive cases and other variables.**","metadata":{}},{"cell_type":"code","source":"#make sure index is the patient id so we can join these aggregate dfs\n\nindividuals_sum.set_index('patient_id', inplace=True)","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:53.118904Z","iopub.execute_input":"2023-09-08T17:55:53.119267Z","iopub.status.idle":"2023-09-08T17:55:53.125001Z","shell.execute_reply.started":"2023-09-08T17:55:53.119233Z","shell.execute_reply":"2023-09-08T17:55:53.124117Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"patients = individuals_count.join(individuals_sum, lsuffix='_count', rsuffix='_sum')\npatients.drop(columns=['sex','age_approx_count','anatom_site_general_challenge','diagnosis', 'benign_malignant','target_count', 'age_approx_sum'], inplace=True)\npatients.rename(columns={'image_name':'n_images'}, inplace=True)\npatients.head()","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:53.126411Z","iopub.execute_input":"2023-09-08T17:55:53.127544Z","iopub.status.idle":"2023-09-08T17:55:53.154536Z","shell.execute_reply.started":"2023-09-08T17:55:53.127509Z","shell.execute_reply":"2023-09-08T17:55:53.153124Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#join with train df to include sex and age data for each patient\npatients = patients.join(train.set_index('patient_id'), on='patient_id')","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:53.155886Z","iopub.execute_input":"2023-09-08T17:55:53.157114Z","iopub.status.idle":"2023-09-08T17:55:53.188939Z","shell.execute_reply.started":"2023-09-08T17:55:53.157072Z","shell.execute_reply":"2023-09-08T17:55:53.187544Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#clean it up\npatients.reset_index(inplace=True)\npatients.drop_duplicates(subset='patient_id', inplace=True)\npatients.drop(columns=['image_name','anatom_site_general_challenge', 'diagnosis',   'benign_malignant', 'target'], inplace=True)\npatients.head()","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:53.190560Z","iopub.execute_input":"2023-09-08T17:55:53.191493Z","iopub.status.idle":"2023-09-08T17:55:53.226134Z","shell.execute_reply.started":"2023-09-08T17:55:53.191447Z","shell.execute_reply":"2023-09-08T17:55:53.225199Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#plot n_images v. target_sum\nax1 = sns.scatterplot(data=patients, x='n_images', y ='target_sum', hue='sex')\nax1.set_title('n_images per patient v. target_sum')\n#ax1.tick_params(axis='x', labelrotation = 45, labelsize = 12)","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:53.228520Z","iopub.execute_input":"2023-09-08T17:55:53.228989Z","iopub.status.idle":"2023-09-08T17:55:53.670071Z","shell.execute_reply.started":"2023-09-08T17:55:53.228944Z","shell.execute_reply":"2023-09-08T17:55:53.668937Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"ax1 = sns.violinplot(data=patients, y='n_images', x ='target_sum')\nax1.set_title('n_images per patient v. target_sum')\n","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:53.671639Z","iopub.execute_input":"2023-09-08T17:55:53.672106Z","iopub.status.idle":"2023-09-08T17:55:54.069557Z","shell.execute_reply.started":"2023-09-08T17:55:53.672073Z","shell.execute_reply":"2023-09-08T17:55:54.068339Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"ax1 = sns.stripplot(data=patients, y='target_sum', x ='sex')\nax1.set_title('target_sum distribution by gender')","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:54.071135Z","iopub.execute_input":"2023-09-08T17:55:54.072335Z","iopub.status.idle":"2023-09-08T17:55:54.317610Z","shell.execute_reply.started":"2023-09-08T17:55:54.072290Z","shell.execute_reply":"2023-09-08T17:55:54.316353Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#plot age_approx v target_sum\nax1 = sns.stripplot(data=patients, x='age_approx', y ='target_sum')\nax1.set_title('age_approx v. target_sum')\n","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:54.319612Z","iopub.execute_input":"2023-09-08T17:55:54.320095Z","iopub.status.idle":"2023-09-08T17:55:54.766077Z","shell.execute_reply.started":"2023-09-08T17:55:54.320049Z","shell.execute_reply":"2023-09-08T17:55:54.764860Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#plot n_images v. target_sum\nax1 = sns.histplot(data=patients, x='age_approx', bins=16)\nax1.set_title('Ditribution of Patient Ages in Dataset')","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:54.767669Z","iopub.execute_input":"2023-09-08T17:55:54.768768Z","iopub.status.idle":"2023-09-08T17:55:55.109561Z","shell.execute_reply.started":"2023-09-08T17:55:54.768719Z","shell.execute_reply":"2023-09-08T17:55:55.108268Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"**Visualizing Images**\n","metadata":{}},{"cell_type":"code","source":"#open one image\n\n#import ImageIO\nimport imageio.v2 as imageio\n\n#load an image\nim = imageio.imread('../input/siim-isic-melanoma-classification/jpeg/train/ISIC_0015719.jpg')\n\n#plot image\nplt.imshow(im)\nplt.axis('off')\n#plt.show()\n\n#print metadata\nprint(im.meta.keys())","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:55.110947Z","iopub.execute_input":"2023-09-08T17:55:55.111318Z","iopub.status.idle":"2023-09-08T17:55:59.250987Z","shell.execute_reply.started":"2023-09-08T17:55:55.111286Z","shell.execute_reply":"2023-09-08T17:55:59.249864Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"**Plot malignant lesions**","metadata":{}},{"cell_type":"code","source":"#Now, let's see if we can show all the malignant images\n\n#make a list of image names that have the malignant target value = 1\nmal_ims = train[train['target']==1].image_name\n\n#convert those image names into path names to access each photo\n#mal_im_paths = ['../input/siim-isic-melanoma-classification/jpeg/train/' + str(im) + '.jpg' for im in mal_ims]","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:59.252645Z","iopub.execute_input":"2023-09-08T17:55:59.253863Z","iopub.status.idle":"2023-09-08T17:55:59.262537Z","shell.execute_reply.started":"2023-09-08T17:55:59.253805Z","shell.execute_reply":"2023-09-08T17:55:59.261071Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#plot all malignant images\n\nfig, axs = plt.subplots(nrows=146, ncols=4, tight_layout=True, figsize=(15,365))\nfig.suptitle('Malignant Lesions', fontsize=14)\n\nfor ax, image in zip(axs.ravel(), mal_ims):\n    path = '../input/siim-isic-melanoma-classification/jpeg/train/' + str(image) + '.jpg'\n    im = imageio.imread(path)\n    ax.imshow(im)\n    ax.axis('off')\n    ax.set_title(image)","metadata":{"execution":{"iopub.status.busy":"2023-09-08T17:55:59.264156Z","iopub.execute_input":"2023-09-08T17:55:59.265030Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import random\nrandom.seed(42)\n\n#make a list of image names that have the malignant target value = 0\nben_ims = train[train['target']==0].image_name\n\n#take a random sample of 100 of these images\nrand_ben_ims = random.sample(list(ben_ims),100)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#plot the subset of 100 benign images\n\nfig, axs = plt.subplots(nrows=25, ncols=4, tight_layout=True, figsize=(15,62.5))\nfig.suptitle('Subset of Benign Lesions', fontsize=14, y=0.98)\n\nfor ax, image in zip(axs.ravel(), rand_ben_ims):\n    path = '../input/siim-isic-melanoma-classification/jpeg/train/' + str(image) + '.jpg'\n    im = imageio.imread(path)\n    ax.imshow(im)\n    ax.axis('off')\n    ax.set_title(image)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# System\nimport cv2\nimport os, os.path\nfrom PIL import Image             \nimport gc\nimport time\nimport datetime\n\n# Basics\nimport pandas as pd\nimport numpy as np\nimport random\nimport seaborn as sns\nimport matplotlib\nimport matplotlib.pyplot as plt\n%matplotlib inline\n\nfrom tqdm.notebook import tqdm     \n\n# SKlearn\nfrom sklearn.model_selection import StratifiedKFold, GroupKFold\nfrom sklearn.metrics import accuracy_score, roc_auc_score, confusion_matrix\nfrom sklearn.preprocessing import LabelEncoder\nfrom sklearn.preprocessing import OneHotEncoder\nfrom sklearn import preprocessing\n\nfrom skimage import io\n\n# PyTorch\nimport torch\nimport torchvision\nimport torch.nn as nn\nimport torch.nn.functional as F\nfrom torch.utils.data import Dataset, DataLoader, Subset\n\nfrom torchvision.models import resnet34\nfrom torchvision import transforms\n\nimport warnings\nwarnings.filterwarnings(\"ignore\")\n\nprint(\"PyTorch Version: \",torch.__version__)\nprint(\"Torchvision Version: \",torchvision.__version__)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# For reproducibility\nseed = 1234\n\nnp.random.seed(seed)\nrandom.seed(seed)\ntorch.manual_seed(seed)\ntorch.cuda.manual_seed(seed)\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\nprint('Device available now:', device)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_df = pd.read_csv('../input/siim-isic-melanoma-classification/train.csv')[1:1000]\n# train_df = train_df.head(1000)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_df.target.value_counts()","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"len(train_df)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class MelanomaDataset(Dataset):\n    \n    def __init__(self, dataframe,is_train=True, is_valid=False, is_test=False):\n        self.dataframe, self.is_train, self.is_valid = dataframe, is_train, is_valid\n        \n        # Data Augmentation\n        if is_train or is_test:\n            self.transform = transforms.Compose([transforms.ToPILImage(),\n                                                 transforms.RandomResizedCrop((224,224), scale=(0.4, 1.0)),\n                                                 transforms.RandomHorizontalFlip(p = 0.3),\n                                                 transforms.RandomVerticalFlip(p = 0.3),\n                                                 transforms.ToTensor(),\n                                                 transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))])\n        else:\n            self.transform = transforms.Compose([transforms.ToPILImage(),\n                                                 transforms.Resize((224,224)),\n                                                 transforms.ToTensor(),\n                                                 transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))])\n            \n    def __len__(self):\n        return len(self.dataframe)\n    \n    def __getitem__(self, index):\n        # Select path and read image\n        img_name = self.dataframe['image_name'][index]\n        image_path = f'../input/siim-isic-melanoma-classification/jpeg/train/{img_name}.jpg'\n        image = io.imread(image_path)\n        \n        # Apply transforms\n        image = self.transform(image)\n\n        \n        # If train/valid: image + class | If test: only image\n        if self.is_train or self.is_valid:\n            return (image, self.dataframe['target'][index])\n        else:\n            return (image)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(resnet34(pretrained=True))","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class ResNet34Network(nn.Module):\n    def __init__(self):\n        super().__init__()\n        \n        # Define Feature part (IMAGE)\n        self.features = resnet34(pretrained=True) # 1000 neurons out\n        \n        for param in self.features.parameters():\n            param.requires_grad = False\n  \n        # Define Classification part\n        self.classification = nn.Linear(1000, 1)\n        \n        \n    def forward(self, image):\n        # Image CNN\n        image = self.features(image)\n        \n        # Classifier\n        out = self.classification(image)\n        \n        return out","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model = ResNet34Network()\nmodel = model.to(device)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(model)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Data object and Loader\ndataset = MelanomaDataset(train_df, is_train=True, is_valid=False, is_test=False)\nloader = DataLoader(dataset,batch_size=3, shuffle=True)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Get a sample\nfor image, labels in loader:\n    image_example = image\n    labels_example = torch.tensor(labels, dtype=torch.float32)\n    break\n    \nprint('Data shape:', image_example.shape)\nprint('Label:', labels_example)\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"learning_rate = 0.0005\nepochs = 5\ndevice = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\nprint(device)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Initiate the model\nmodel = model\nmodel = model.to(device)\noptimizer = torch.optim.Adam(filter(lambda p: p.requires_grad, model.parameters()), lr = learning_rate)\ncriterion = nn.BCEWithLogitsLoss()  ","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_size = int(len(train_df) * 0.6)\nval_size = int(len(train_df) * 0.2)\ntest_size = int(len(train_df) * 0.2)\n\nprint(f'train size : {train_size}, val size : {val_size}, test size : {test_size}')","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# --- Read in Data ---\ntrain_data = train_df.iloc[:train_size].reset_index(drop=True)\nvalid_data = train_df.iloc[train_size:train_size + val_size].reset_index(drop=True)\ntest_data = train_df.iloc[train_size + val_size:].reset_index(drop=True)\n# Create Data instances\ntrain = MelanomaDataset(train_data, is_train=True, is_valid=False, is_test=False)\nvalid = MelanomaDataset(valid_data, is_train=False, is_valid=True, is_test=False)\ntest = MelanomaDataset(valid_data, is_train=False, is_valid=True, is_test=False)\n\n# Dataloaders\ntrain_loader = DataLoader(train, batch_size=16, shuffle=True)\nvalid_loader = DataLoader(valid, batch_size=8, shuffle=True)\ntest_loader = DataLoader(valid, batch_size=1, shuffle=True)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# === EPOCHS ===\nfor epoch in range(epochs):\n    print(f'epoch : {epoch} start!')\n    start_time = time.time()\n    correct = 0\n    train_losses = 0\n\n    # === TRAIN ===\n    # Sets the module in training mode.\n    model.train()\n\n    for images, labels in train_loader:\n        # Save them to device\n        images = torch.tensor(images, device=device, dtype=torch.float32)\n        labels = torch.tensor(labels, device=device, dtype=torch.float32)\n\n        # Clear gradients first; very important, usually done BEFORE prediction\n        optimizer.zero_grad()\n\n        # Log Probabilities & Backpropagation\n        out = model(images)\n        loss = criterion(out, labels.unsqueeze(1))\n        loss.backward()\n        optimizer.step()\n\n        train_losses += loss.item()\n        # From log probabilities to actual probabilities\n        train_preds = torch.round(torch.sigmoid(out)) # 0 and 1\n        # Number of correct predictions\n        correct += (train_preds.cpu() == labels.cpu().unsqueeze(1)).sum().item()\n\n    # Compute Train Accuracy\n    train_acc = correct*100 / train_size\n    print(f'Epoch :{epoch + 1} - train accuracy: {train_acc}')\n    \n    # === EVAL ===\n    model.eval()\n\n    # Create matrix to store evaluation predictions (for accuracy)\n    valid_preds = torch.zeros(size = (len(valid_data), 1), device=device, dtype=torch.float32)\n\n\n    # Disables gradients (we need to be sure no optimization happens)\n    with torch.no_grad():\n        for k, (images, labels) in enumerate(valid_loader):\n            images = torch.tensor(images, device=device, dtype=torch.float32)\n            labels = torch.tensor(labels, device=device, dtype=torch.float32)\n\n            out = model(images)\n            pred = torch.sigmoid(out)\n            valid_preds[k*images.shape[0] : k*images.shape[0] + images.shape[0]] = pred\n\n        # Compute accuracy\n        valid_acc = accuracy_score(valid_data['target'].values, \n                                           torch.round(valid_preds.cpu()))*100\n        \n        # Compute ROC\n        valid_roc = roc_auc_score(valid_data['target'].values, \n                                          valid_preds.cpu())\n\n        # Compute time on Train + Eval\n        duration = str(datetime.timedelta(seconds=time.time() - start_time))[:7]\n\n\n        # PRINT INFO\n        print('{} | Epoch: {}/{} | Loss: {:.4} | Train Acc: {:.3} | Valid Acc: {:.3} ROC: {:.3}'.\\\n                    format(duration, epoch+1, epochs, train_losses, train_acc, valid_acc,valid_roc))\n        \n        \ntorch.save(model.state_dict(), './model.pt')\n\nfrom IPython.display import FileLink\nFileLink(r'model.pt')","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def check_accuracy(loader, model):\n    num_correct = 0\n    num_samples = 0\n    model.eval()\n    \n    with torch.no_grad():\n        for x, y in loader:\n            x = x.to(device=device)\n            y = y.to(device=device)\n            \n            scores = model(x)\n            _, predictions = scores.max(1)\n            num_correct += (predictions == y).sum()\n            num_samples += predictions.size(0)\n        \n        print(f'Got {num_correct} / {num_samples} with Test Accuracy {float(num_correct)/float(num_samples)*100:.2f}') ","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"check_accuracy(test_loader,model)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"torch.save(model.state_dict(), './model.pt')","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Data object and Loader\ndataset = MelanomaDataset(train_df, is_train=True, is_valid=False, is_test=False)\nloader = DataLoader(dataset,batch_size=1, shuffle=True)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Get a sample\nfor image, label in test_loader:\n    image = torch.tensor(image, device=device, dtype=torch.float32)\n    label = torch.tensor(label, device=device, dtype=torch.float32)\n    out = model(image)\n    loss = criterion(out, label.unsqueeze(1))\n    pred = torch.sigmoid(out)\n    print('loss: ', loss)\n    print('Label:', label)\n    print('Pred:', pred)\n    break","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from IPython.display import FileLink\nFileLink(r'model.pt')","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model_lo = ResNet34Network()\nmodel_lo = model_lo.to(device)\nmodel_lo.load_state_dict(torch.load('model.pt'))\nmodel_lo.eval()","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(train_df.loc[train_df['target']==1].head().values)","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_df.columns","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Data object and Loader\ndf = pd.DataFrame({'image_name':['ISIC_2637011', 'ISIC_0149568'], 'patient_id':['IP_7279968', 'IP_0962375'], 'sex':['male', 'female'], 'age_approx':[45.0, 55.0],\n                  'anatom_site_general_challenge':['head/neck', 'upper extremity'], 'diagnosis':['unknown', 'melanoma'],\n                   'benign_malignant':['benign', 'malignant'], 'target':[0,1]})\n\n\ndataset = MelanomaDataset(df, is_train=True, is_valid=False, is_test=False)\nloader = DataLoader(dataset,batch_size=1, shuffle=False)\n# Get a sample\nfor image, label in loader:\n    image = torch.tensor(image, device=device, dtype=torch.float32)\n    label = torch.tensor(label, device=device, dtype=torch.float32)\n    out = model(image)\n    loss = criterion(out, label.unsqueeze(1))\n    pred = torch.sigmoid(out)\n    \n#     print('loss: ', loss)\n#     print('Label:', label)\n    print('Pred:', pred)\n    print('out:', torch.round(pred))\n    print('__________________________________________________________________')","metadata":{"trusted":true},"execution_count":null,"outputs":[]}],"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}}