{"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"},"kaggle":{"accelerator":"nvidiaTeslaT4","dataSources":[{"sourceId":20270,"databundleVersionId":1222630,"sourceType":"competition"}],"dockerImageVersionId":30380,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":true}},"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')\n#plt.style.use('seaborn-notebook')","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:05:36.94413Z","iopub.execute_input":"2025-10-29T16:05:36.94487Z","iopub.status.idle":"2025-10-29T16:05:36.951096Z","shell.execute_reply.started":"2025-10-29T16:05:36.944835Z","shell.execute_reply":"2025-10-29T16:05:36.95001Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import gc\nimport tensorflow as tf\n\n# Clear memory\ngc.collect()\ntf.keras.backend.clear_session()\n\n# For PyTorch models\nimport torch\ntorch.cuda.empty_cache()  # If using GPU","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:14:09.358074Z","iopub.execute_input":"2025-10-29T16:14:09.358646Z","iopub.status.idle":"2025-10-29T16:14:09.536269Z","shell.execute_reply.started":"2025-10-29T16:14:09.358616Z","shell.execute_reply":"2025-10-29T16:14:09.535269Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:14:11.457543Z","iopub.execute_input":"2025-10-29T16:14:11.458279Z","iopub.status.idle":"2025-10-29T16:14:11.543537Z","shell.execute_reply.started":"2025-10-29T16:14:11.458244Z","shell.execute_reply":"2025-10-29T16:14:11.542788Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train.head()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:14:13.76006Z","iopub.execute_input":"2025-10-29T16:14:13.76073Z","iopub.status.idle":"2025-10-29T16:14:13.772838Z","shell.execute_reply.started":"2025-10-29T16:14:13.7607Z","shell.execute_reply":"2025-10-29T16:14:13.77185Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train.shape","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:14:17.637353Z","iopub.execute_input":"2025-10-29T16:14:17.638113Z","iopub.status.idle":"2025-10-29T16:14:17.643865Z","shell.execute_reply.started":"2025-10-29T16:14:17.638076Z","shell.execute_reply":"2025-10-29T16:14:17.642681Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train.info()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:19.470555Z","iopub.execute_input":"2025-10-29T15:33:19.470899Z","iopub.status.idle":"2025-10-29T15:33:19.516412Z","shell.execute_reply.started":"2025-10-29T15:33:19.470868Z","shell.execute_reply":"2025-10-29T15:33:19.515002Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train.describe()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:19.51845Z","iopub.execute_input":"2025-10-29T15:33:19.51879Z","iopub.status.idle":"2025-10-29T15:33:19.547996Z","shell.execute_reply.started":"2025-10-29T15:33:19.518761Z","shell.execute_reply":"2025-10-29T15:33:19.546761Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:19.549407Z","iopub.execute_input":"2025-10-29T15:33:19.549735Z","iopub.status.idle":"2025-10-29T15:33:19.823885Z","shell.execute_reply.started":"2025-10-29T15:33:19.549703Z","shell.execute_reply":"2025-10-29T15:33:19.822647Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:19.82561Z","iopub.execute_input":"2025-10-29T15:33:19.82595Z","iopub.status.idle":"2025-10-29T15:33:20.02171Z","shell.execute_reply.started":"2025-10-29T15:33:19.825905Z","shell.execute_reply":"2025-10-29T15:33:20.020535Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:20.022999Z","iopub.execute_input":"2025-10-29T15:33:20.023378Z","iopub.status.idle":"2025-10-29T15:33:20.301708Z","shell.execute_reply.started":"2025-10-29T15:33:20.023345Z","shell.execute_reply":"2025-10-29T15:33:20.300375Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:20.303666Z","iopub.execute_input":"2025-10-29T15:33:20.304193Z","iopub.status.idle":"2025-10-29T15:33:20.517052Z","shell.execute_reply.started":"2025-10-29T15:33:20.304133Z","shell.execute_reply":"2025-10-29T15:33:20.515959Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:20.518307Z","iopub.execute_input":"2025-10-29T15:33:20.518621Z","iopub.status.idle":"2025-10-29T15:33:20.781687Z","shell.execute_reply.started":"2025-10-29T15:33:20.518591Z","shell.execute_reply":"2025-10-29T15:33:20.779849Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:20.786264Z","iopub.execute_input":"2025-10-29T15:33:20.786666Z","iopub.status.idle":"2025-10-29T15:33:20.810493Z","shell.execute_reply.started":"2025-10-29T15:33:20.786633Z","shell.execute_reply":"2025-10-29T15:33:20.809246Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:06:09.348593Z","iopub.execute_input":"2025-10-29T16:06:09.349307Z","iopub.status.idle":"2025-10-29T16:06:09.370575Z","shell.execute_reply.started":"2025-10-29T16:06:09.349275Z","shell.execute_reply":"2025-10-29T16:06:09.369786Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:14:32.872946Z","iopub.execute_input":"2025-10-29T16:14:32.873931Z","iopub.status.idle":"2025-10-29T16:14:32.884865Z","shell.execute_reply.started":"2025-10-29T16:14:32.873892Z","shell.execute_reply":"2025-10-29T16:14:32.883675Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"for col in malignant.columns:\n    print(col + ' missing values: ' + str(malignant[col].isna().sum()))\n","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:06:19.680229Z","iopub.execute_input":"2025-10-29T16:06:19.680849Z","iopub.status.idle":"2025-10-29T16:06:19.689201Z","shell.execute_reply.started":"2025-10-29T16:06:19.680786Z","shell.execute_reply":"2025-10-29T16:06:19.688191Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:14:42.602364Z","iopub.execute_input":"2025-10-29T16:14:42.602708Z","iopub.status.idle":"2025-10-29T16:14:42.864683Z","shell.execute_reply.started":"2025-10-29T16:14:42.602677Z","shell.execute_reply":"2025-10-29T16:14:42.863869Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# **Test Set exploration**","metadata":{}},{"cell_type":"code","source":"test.head()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:06:25.188208Z","iopub.execute_input":"2025-10-29T16:06:25.188571Z","iopub.status.idle":"2025-10-29T16:06:25.202246Z","shell.execute_reply.started":"2025-10-29T16:06:25.188537Z","shell.execute_reply":"2025-10-29T16:06:25.201025Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"test.shape","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:21.134814Z","iopub.execute_input":"2025-10-29T15:33:21.135322Z","iopub.status.idle":"2025-10-29T15:33:21.147772Z","shell.execute_reply.started":"2025-10-29T15:33:21.135278Z","shell.execute_reply":"2025-10-29T15:33:21.146633Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:21.149169Z","iopub.execute_input":"2025-10-29T15:33:21.149516Z","iopub.status.idle":"2025-10-29T15:33:21.162961Z","shell.execute_reply.started":"2025-10-29T15:33:21.149484Z","shell.execute_reply":"2025-10-29T15:33:21.16152Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#how many images are associated with each patient?\nindividuals_count = train.groupby('patient_id').count()\n\nindividuals_count.head()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:14:49.449521Z","iopub.execute_input":"2025-10-29T16:14:49.450529Z","iopub.status.idle":"2025-10-29T16:14:49.478523Z","shell.execute_reply.started":"2025-10-29T16:14:49.450482Z","shell.execute_reply":"2025-10-29T16:14:49.477553Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:21.202173Z","iopub.execute_input":"2025-10-29T15:33:21.202479Z","iopub.status.idle":"2025-10-29T15:33:21.584952Z","shell.execute_reply.started":"2025-10-29T15:33:21.202452Z","shell.execute_reply":"2025-10-29T15:33:21.583896Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:14:54.185649Z","iopub.execute_input":"2025-10-29T16:14:54.186521Z","iopub.status.idle":"2025-10-29T16:14:54.387991Z","shell.execute_reply.started":"2025-10-29T16:14:54.186485Z","shell.execute_reply":"2025-10-29T16:14:54.387018Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:21.586829Z","iopub.execute_input":"2025-10-29T15:33:21.587304Z","iopub.status.idle":"2025-10-29T15:33:21.596344Z","shell.execute_reply.started":"2025-10-29T15:33:21.58726Z","shell.execute_reply":"2025-10-29T15:33:21.595267Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:15:03.178164Z","iopub.execute_input":"2025-10-29T16:15:03.179243Z","iopub.status.idle":"2025-10-29T16:15:03.191835Z","shell.execute_reply.started":"2025-10-29T16:15:03.179204Z","shell.execute_reply":"2025-10-29T16:15:03.190672Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:15:05.866009Z","iopub.execute_input":"2025-10-29T16:15:05.866351Z","iopub.status.idle":"2025-10-29T16:15:05.881306Z","shell.execute_reply.started":"2025-10-29T16:15:05.866325Z","shell.execute_reply":"2025-10-29T16:15:05.880293Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"print('percentage of missing patient data= {}%'.format(round((3+48+17)/len(train)*100,2)))","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:21.930199Z","iopub.execute_input":"2025-10-29T15:33:21.930522Z","iopub.status.idle":"2025-10-29T15:33:21.936817Z","shell.execute_reply.started":"2025-10-29T15:33:21.930494Z","shell.execute_reply":"2025-10-29T15:33:21.935248Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:15:15.777063Z","iopub.execute_input":"2025-10-29T16:15:15.777395Z","iopub.status.idle":"2025-10-29T16:15:15.794486Z","shell.execute_reply.started":"2025-10-29T16:15:15.777366Z","shell.execute_reply":"2025-10-29T16:15:15.793568Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:06:50.801563Z","iopub.execute_input":"2025-10-29T16:06:50.802421Z","iopub.status.idle":"2025-10-29T16:06:50.820156Z","shell.execute_reply.started":"2025-10-29T16:06:50.802386Z","shell.execute_reply":"2025-10-29T16:06:50.819215Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"**Relationships between positive cases and other variables.**","metadata":{}},{"cell_type":"code","source":"if 'index' in individuals_sum.columns:\n    individuals_sum.drop('index', axis=1, inplace=True)\n    print(\"✓ Cleaned up - patient_id is already the index\")\n\n\nprint(\"Index name:\", individuals_sum.index.name)\nprint(\"Columns:\", individuals_sum.columns.tolist())\nprint(\"\\nFirst few rows:\")\nprint(individuals_sum.head())","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:12:26.822422Z","iopub.execute_input":"2025-10-29T16:12:26.822873Z","iopub.status.idle":"2025-10-29T16:12:26.832548Z","shell.execute_reply.started":"2025-10-29T16:12:26.822815Z","shell.execute_reply":"2025-10-29T16:12:26.83152Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:15:33.199982Z","iopub.execute_input":"2025-10-29T16:15:33.200319Z","iopub.status.idle":"2025-10-29T16:15:33.22213Z","shell.execute_reply.started":"2025-10-29T16:15:33.20029Z","shell.execute_reply":"2025-10-29T16:15:33.220878Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:15:40.845554Z","iopub.execute_input":"2025-10-29T16:15:40.846427Z","iopub.status.idle":"2025-10-29T16:15:40.858505Z","shell.execute_reply.started":"2025-10-29T16:15:40.846391Z","shell.execute_reply":"2025-10-29T16:15:40.857548Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:15:47.975715Z","iopub.execute_input":"2025-10-29T16:15:47.976091Z","iopub.status.idle":"2025-10-29T16:15:48.051177Z","shell.execute_reply.started":"2025-10-29T16:15:47.97606Z","shell.execute_reply":"2025-10-29T16:15:48.04996Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:15:52.623966Z","iopub.execute_input":"2025-10-29T16:15:52.624295Z","iopub.status.idle":"2025-10-29T16:15:52.645289Z","shell.execute_reply.started":"2025-10-29T16:15:52.624266Z","shell.execute_reply":"2025-10-29T16:15:52.644439Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:22.086366Z","iopub.execute_input":"2025-10-29T15:33:22.08669Z","iopub.status.idle":"2025-10-29T15:33:22.439487Z","shell.execute_reply.started":"2025-10-29T15:33:22.086659Z","shell.execute_reply":"2025-10-29T15:33:22.438342Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:22.440721Z","iopub.execute_input":"2025-10-29T15:33:22.441076Z","iopub.status.idle":"2025-10-29T15:33:22.754606Z","shell.execute_reply.started":"2025-10-29T15:33:22.441036Z","shell.execute_reply":"2025-10-29T15:33:22.753483Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"ax1 = sns.stripplot(data=patients, y='target_sum', x ='sex')\nax1.set_title('target_sum distribution by gender')","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:22.756118Z","iopub.execute_input":"2025-10-29T15:33:22.756615Z","iopub.status.idle":"2025-10-29T15:33:22.947661Z","shell.execute_reply.started":"2025-10-29T15:33:22.75657Z","shell.execute_reply":"2025-10-29T15:33:22.946541Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:22.949325Z","iopub.execute_input":"2025-10-29T15:33:22.950134Z","iopub.status.idle":"2025-10-29T15:33:23.298464Z","shell.execute_reply.started":"2025-10-29T15:33:22.950091Z","shell.execute_reply":"2025-10-29T15:33:23.296607Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:23.299972Z","iopub.execute_input":"2025-10-29T15:33:23.300413Z","iopub.status.idle":"2025-10-29T15:33:23.579882Z","shell.execute_reply.started":"2025-10-29T15:33:23.30037Z","shell.execute_reply":"2025-10-29T15:33:23.578746Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:23.581346Z","iopub.execute_input":"2025-10-29T15:33:23.581678Z","iopub.status.idle":"2025-10-29T15:33:27.044562Z","shell.execute_reply.started":"2025-10-29T15:33:23.581648Z","shell.execute_reply":"2025-10-29T15:33:27.043502Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:27.046107Z","iopub.execute_input":"2025-10-29T15:33:27.047007Z","iopub.status.idle":"2025-10-29T15:33:27.054334Z","shell.execute_reply.started":"2025-10-29T15:33:27.046964Z","shell.execute_reply":"2025-10-29T15:33:27.053185Z"}},"outputs":[],"execution_count":null},{"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":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T15:33:27.055916Z","iopub.execute_input":"2025-10-29T15:33:27.056378Z","iopub.status.idle":"2025-10-29T15:43:07.566207Z","shell.execute_reply.started":"2025-10-29T15:33:27.056338Z","shell.execute_reply":"2025-10-29T15:43:07.565009Z"}},"outputs":[],"execution_count":null},{"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":{"iopub.status.busy":"2025-10-29T15:43:07.567754Z","iopub.execute_input":"2025-10-29T15:43:07.568159Z","iopub.status.idle":"2025-10-29T15:43:07.586167Z","shell.execute_reply.started":"2025-10-29T15:43:07.568124Z","shell.execute_reply":"2025-10-29T15:43:07.584702Z"}},"outputs":[],"execution_count":null},{"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":{"iopub.status.busy":"2025-10-29T15:43:07.588052Z","iopub.execute_input":"2025-10-29T15:43:07.588617Z"}},"outputs":[],"execution_count":null},{"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":{"iopub.status.busy":"2025-10-29T16:16:13.860156Z","iopub.execute_input":"2025-10-29T16:16:13.86049Z","iopub.status.idle":"2025-10-29T16:16:14.865865Z","shell.execute_reply.started":"2025-10-29T16:16:13.86046Z","shell.execute_reply":"2025-10-29T16:16:14.864843Z"}},"outputs":[],"execution_count":null},{"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":{"iopub.status.busy":"2025-10-29T16:16:22.202579Z","iopub.execute_input":"2025-10-29T16:16:22.202959Z","iopub.status.idle":"2025-10-29T16:16:22.208647Z","shell.execute_reply.started":"2025-10-29T16:16:22.202925Z","shell.execute_reply":"2025-10-29T16:16:22.207723Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\nprint('Device available now:', device)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:16:25.008541Z","iopub.execute_input":"2025-10-29T16:16:25.009463Z","iopub.status.idle":"2025-10-29T16:16:25.091485Z","shell.execute_reply.started":"2025-10-29T16:16:25.009425Z","shell.execute_reply":"2025-10-29T16:16:25.090374Z"}},"outputs":[],"execution_count":null},{"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":{"iopub.status.busy":"2025-10-29T16:16:30.226546Z","iopub.execute_input":"2025-10-29T16:16:30.226902Z","iopub.status.idle":"2025-10-29T16:16:30.273107Z","shell.execute_reply.started":"2025-10-29T16:16:30.226874Z","shell.execute_reply":"2025-10-29T16:16:30.272335Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_df.target.value_counts()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:16:32.510761Z","iopub.execute_input":"2025-10-29T16:16:32.511103Z","iopub.status.idle":"2025-10-29T16:16:32.520554Z","shell.execute_reply.started":"2025-10-29T16:16:32.511077Z","shell.execute_reply":"2025-10-29T16:16:32.519447Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"len(train_df)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:16:35.080139Z","iopub.execute_input":"2025-10-29T16:16:35.080938Z","iopub.status.idle":"2025-10-29T16:16:35.086858Z","shell.execute_reply.started":"2025-10-29T16:16:35.080903Z","shell.execute_reply":"2025-10-29T16:16:35.085802Z"}},"outputs":[],"execution_count":null},{"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":{"iopub.status.busy":"2025-10-29T16:16:37.409983Z","iopub.execute_input":"2025-10-29T16:16:37.410329Z","iopub.status.idle":"2025-10-29T16:16:37.42011Z","shell.execute_reply.started":"2025-10-29T16:16:37.4103Z","shell.execute_reply":"2025-10-29T16:16:37.419211Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"print(resnet34(pretrained=True))","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:16:40.862351Z","iopub.execute_input":"2025-10-29T16:16:40.862697Z","iopub.status.idle":"2025-10-29T16:16:41.895347Z","shell.execute_reply.started":"2025-10-29T16:16:40.862667Z","shell.execute_reply":"2025-10-29T16:16:41.894292Z"}},"outputs":[],"execution_count":null},{"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":{"iopub.status.busy":"2025-10-29T16:16:47.721559Z","iopub.execute_input":"2025-10-29T16:16:47.722391Z","iopub.status.idle":"2025-10-29T16:16:47.728016Z","shell.execute_reply.started":"2025-10-29T16:16:47.722355Z","shell.execute_reply":"2025-10-29T16:16:47.72703Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model = ResNet34Network()\nmodel = model.to(device)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:16:51.177919Z","iopub.execute_input":"2025-10-29T16:16:51.178882Z","iopub.status.idle":"2025-10-29T16:16:57.285634Z","shell.execute_reply.started":"2025-10-29T16:16:51.178844Z","shell.execute_reply":"2025-10-29T16:16:57.284771Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"print(model)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:16:57.287213Z","iopub.execute_input":"2025-10-29T16:16:57.287486Z","iopub.status.idle":"2025-10-29T16:16:57.294001Z","shell.execute_reply.started":"2025-10-29T16:16:57.287457Z","shell.execute_reply":"2025-10-29T16:16:57.292926Z"}},"outputs":[],"execution_count":null},{"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":{"iopub.status.busy":"2025-10-29T16:17:04.190747Z","iopub.execute_input":"2025-10-29T16:17:04.191126Z","iopub.status.idle":"2025-10-29T16:17:04.196618Z","shell.execute_reply.started":"2025-10-29T16:17:04.191096Z","shell.execute_reply":"2025-10-29T16:17:04.195562Z"}},"outputs":[],"execution_count":null},{"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":{"iopub.status.busy":"2025-10-29T16:17:07.22084Z","iopub.execute_input":"2025-10-29T16:17:07.221651Z","iopub.status.idle":"2025-10-29T16:17:07.921659Z","shell.execute_reply.started":"2025-10-29T16:17:07.221619Z","shell.execute_reply":"2025-10-29T16:17:07.920608Z"}},"outputs":[],"execution_count":null},{"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":{"iopub.status.busy":"2025-10-29T16:17:10.267088Z","iopub.execute_input":"2025-10-29T16:17:10.267451Z","iopub.status.idle":"2025-10-29T16:17:10.273255Z","shell.execute_reply.started":"2025-10-29T16:17:10.267416Z","shell.execute_reply":"2025-10-29T16:17:10.272328Z"}},"outputs":[],"execution_count":null},{"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":{"iopub.status.busy":"2025-10-29T16:17:16.29552Z","iopub.execute_input":"2025-10-29T16:17:16.295873Z","iopub.status.idle":"2025-10-29T16:17:16.304367Z","shell.execute_reply.started":"2025-10-29T16:17:16.295843Z","shell.execute_reply":"2025-10-29T16:17:16.303579Z"}},"outputs":[],"execution_count":null},{"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":{"iopub.status.busy":"2025-10-29T16:17:19.059347Z","iopub.execute_input":"2025-10-29T16:17:19.059722Z","iopub.status.idle":"2025-10-29T16:17:19.06518Z","shell.execute_reply.started":"2025-10-29T16:17:19.059689Z","shell.execute_reply":"2025-10-29T16:17:19.06422Z"}},"outputs":[],"execution_count":null},{"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":{"iopub.status.busy":"2025-10-29T16:17:22.860777Z","iopub.execute_input":"2025-10-29T16:17:22.861167Z","iopub.status.idle":"2025-10-29T16:17:22.870368Z","shell.execute_reply.started":"2025-10-29T16:17:22.861127Z","shell.execute_reply":"2025-10-29T16:17:22.869177Z"}},"outputs":[],"execution_count":null},{"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":{"iopub.status.busy":"2025-10-29T16:17:26.47056Z","iopub.execute_input":"2025-10-29T16:17:26.470928Z","iopub.status.idle":"2025-10-29T16:17:26.47697Z","shell.execute_reply.started":"2025-10-29T16:17:26.470895Z","shell.execute_reply":"2025-10-29T16:17:26.476001Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"check_accuracy(test_loader,model)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:17:29.677411Z","iopub.execute_input":"2025-10-29T16:17:29.677821Z","iopub.status.idle":"2025-10-29T16:19:16.956953Z","shell.execute_reply.started":"2025-10-29T16:17:29.677776Z","shell.execute_reply":"2025-10-29T16:19:16.955942Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# **Building CNN Sequential Model**","metadata":{}},{"cell_type":"code","source":"import tensorflow as tf \nimport tensorflow_addons as tfa \nimport numpy as np \nimport pandas as pd \nimport matplotlib.pyplot as plt\nfrom sklearn.model_selection import train_test_split","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:19:17.089477Z","iopub.execute_input":"2025-10-29T16:19:17.089768Z","iopub.status.idle":"2025-10-29T16:19:17.094332Z","shell.execute_reply.started":"2025-10-29T16:19:17.089734Z","shell.execute_reply":"2025-10-29T16:19:17.093422Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_img_dir = '../input/siim-isic-melanoma-classification/jpeg/train/'\ntest_img_dir = '../input/siim-isic-melanoma-classification/jpeg/test/'\ntrain = pd.read_csv('../input/siim-isic-melanoma-classification/train.csv')\ntest = pd.read_csv('../input/siim-isic-melanoma-classification/test.csv')\ntrain1= pd.read_csv('../input/siim-isic-melanoma-classification/train.csv')[1:1000]\ntest1 = pd.read_csv('../input/siim-isic-melanoma-classification/test.csv')","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:19:17.096687Z","iopub.execute_input":"2025-10-29T16:19:17.096986Z","iopub.status.idle":"2025-10-29T16:19:17.208991Z","shell.execute_reply.started":"2025-10-29T16:19:17.096963Z","shell.execute_reply":"2025-10-29T16:19:17.208279Z"}},"outputs":[],"execution_count":null},{"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":{"iopub.status.busy":"2025-10-29T16:19:17.210139Z","iopub.execute_input":"2025-10-29T16:19:17.210407Z","iopub.status.idle":"2025-10-29T16:43:46.927745Z","shell.execute_reply.started":"2025-10-29T16:19:17.210382Z","shell.execute_reply":"2025-10-29T16:43:46.926844Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train.target.value_counts()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:43:46.929174Z","iopub.execute_input":"2025-10-29T16:43:46.929864Z","iopub.status.idle":"2025-10-29T16:43:46.937243Z","shell.execute_reply.started":"2025-10-29T16:43:46.929826Z","shell.execute_reply":"2025-10-29T16:43:46.936244Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train1.target.value_counts()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:43:46.938889Z","iopub.execute_input":"2025-10-29T16:43:46.939126Z","iopub.status.idle":"2025-10-29T16:43:46.948974Z","shell.execute_reply.started":"2025-10-29T16:43:46.939104Z","shell.execute_reply":"2025-10-29T16:43:46.94806Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"# loading the data\ntrain_car = []\nfor image_name in train1.image_name:\n    train_car.append(tf.keras.preprocessing.image.img_to_array(\n        tf.keras.preprocessing.image.load_img(path = (train_img_dir + image_name + \".jpg\"), color_mode = \"rgba\", target_size = (128,128)), dtype=\"float32\")\n                    )","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-10-29T16:43:46.950105Z","iopub.execute_input":"2025-10-29T16:43:46.950373Z"}},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train1[\"img\"] = train_car\ntest[\"path\"] = [(test_img_dir + img_name + \".jpg\") for img_name in test.image_name]","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"# split the data\nx_train, x_cross_validation, y_train, y_cross_validation = train_test_split(train1.img, train1.target, test_size=0.2, shuffle = True, random_state=42)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"# scale the pixels between 0 and 1\nx_train /= 255\nx_cross_validation /= 255\n# redefine dtypes\ny_train = np.array(y_train, dtype = \"float32\")\ny_cross_validation = np.array(y_cross_validation, dtype = \"float32\")\nx_train = np.array([np.array(val) for val in x_train])\nx_cross_validation = np.array([np.array(val) for val in x_cross_validation])","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model1 = tf.keras.models.Sequential()\n\n# Convolutional & Max Pooling layers\n\nmodel1.add(tf.keras.layers.Conv2D(16, kernel_size=(3, 3), activation='relu', input_shape=(128,128,4)))\nmodel1.add(tf.keras.layers.MaxPooling2D(pool_size=(2, 2)))\n\nmodel1.add(tf.keras.layers.Conv2D(64, kernel_size=(3, 3), activation='relu'))\nmodel1.add(tf.keras.layers.MaxPooling2D(pool_size=(2, 2)))\n\nmodel1.add(tf.keras.layers.Conv2D(128, kernel_size=(3, 3), activation='relu'))\nmodel1.add(tf.keras.layers.MaxPooling2D(pool_size=(2, 2)))\n\n# Flatten & Dense layers\n\nmodel1.add(tf.keras.layers.Flatten())\nmodel1.add(tf.keras.layers.Dense(512, activation='relu'))\n\n# performing binary classification\nmodel1.add(tf.keras.layers.Dense(1, activation='sigmoid'))\n\nmodel1.compile(loss = tfa.losses.SigmoidFocalCrossEntropy(),\n              optimizer = tf.keras.optimizers.Adam(),\n              metrics = ['binary_accuracy',\n                       tf.keras.metrics.FalsePositives(),\n                       tf.keras.metrics.FalseNegatives(), \n                       tf.keras.metrics.TruePositives(),\n                       tf.keras.metrics.TrueNegatives()\n                      ]\n             )","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model1.summary()","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"from tensorflow.keras.optimizers import Adam\n\nlearning_rate = .0001\noptimizer = Adam(learning_rate=learning_rate)\n\nmodel1.compile(loss=\"binary_crossentropy\", optimizer=optimizer, metrics=['accuracy'])","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"# saving the best model for our predictions\ncheckpointer = tf.keras.callbacks.ModelCheckpoint(filepath=\"weights.hdf5\", verbose=1, save_best_only=True)\n\nhistory = model1.fit(x = x_train,\n                    y = y_train,\n                    validation_data=(x_cross_validation, y_cross_validation),\n                    batch_size=10,\n                    epochs=10,\n                    verbose=1, \n                    callbacks=[checkpointer]\n                   )","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"# summarize history for accuracy\nplt.plot(history.history['accuracy'])\nplt.plot(history.history['val_accuracy'])\nplt.title('CNN Model Accuracy')\nplt.ylabel('accuracy')\nplt.xlabel('epoch')\nplt.legend(['train accuracy', ' Validation accuracy'], loc='upper left')\nplt.show()\n# summarize history for loss\nplt.plot(history.history['loss'])\nplt.plot(history.history['val_loss'])\nplt.title('CNN Model Loss')\nplt.ylabel('loss')\nplt.xlabel('epoch')\nplt.legend(['train loss', 'test loss'], loc='upper left')\nplt.show()\n","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"# input to be predicted\ntest = tf.keras.preprocessing.image.img_to_array(tf.keras.preprocessing.image.load_img(path = (\"../input/siim-isic-melanoma-classification/jpeg/test/ISIC_0052060.jpg\"), color_mode = \"rgba\", target_size = (128,128)), dtype=\"float32\")\ntest = test / 255\ntest=np.reshape(test,(1,128,128,4))\ntest = np.array(test)\n# load the best model\nmodel1.load_weights('weights.hdf5')\n\nmodel1.predict(test)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model1.compile(optimizer='Adam',\n              loss='categorical_crossentropy',\n              metrics=['accuracy'])","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"score = model1.evaluate(x_cross_validation, y_cross_validation, verbose=0)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"print('Test loss:', score[0])\nprint('Test accuracy:', score[1])","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# **Building ResNet18**","metadata":{}},{"cell_type":"code","source":"\nfrom torchvision.models import resnet18\nfrom torchvision import transforms\n\nimport warnings\nwarnings.filterwarnings(\"ignore\")","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"print(resnet18(pretrained=True))","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"class ResNet18Network(nn.Module):\n    def __init__(self):\n        super().__init__()\n        \n        # Define Feature part (IMAGE)\n        self.features = resnet18(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},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model5 = ResNet18Network()\nmodel5 = model5.to(device)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model5 = model5\nmodel5 = model5.to(device)\noptimizer = torch.optim.Adam(filter(lambda p: p.requires_grad, model5.parameters()), lr = learning_rate)\ncriterion = nn.BCEWithLogitsLoss() ","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"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    model5.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    model5.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 = model5(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(model5.state_dict(), './model5.pt')\n\nfrom IPython.display import FileLink\nFileLink(r'model5.pt')","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"def check_accuracy(loader, model5):\n    num_correct = 0\n    num_samples = 0\n    model5.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 = model5(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},"outputs":[],"execution_count":null},{"cell_type":"code","source":"check_accuracy(test_loader,model5)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# **Implemantation of VGG16**","metadata":{}},{"cell_type":"code","source":"import keras,os\nfrom keras.models import Sequential\nfrom keras.layers import Dense, Conv2D, MaxPool2D , Flatten\nfrom keras.preprocessing.image import ImageDataGenerator\nimport numpy as np","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model3 = Sequential()\nmodel3.add(Conv2D(input_shape=(224,224,3),filters=64,kernel_size=(3,3),padding=\"same\", activation=\"relu\"))\nmodel3.add(Conv2D(filters=64,kernel_size=(3,3),padding=\"same\", activation=\"relu\"))\nmodel3.add(MaxPool2D(pool_size=(2,2),strides=(2,2)))\nmodel3.add(Conv2D(filters=128, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\nmodel3.add(Conv2D(filters=128, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\nmodel3.add(MaxPool2D(pool_size=(2,2),strides=(2,2)))\nmodel3.add(Conv2D(filters=256, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\nmodel3.add(Conv2D(filters=256, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\nmodel3.add(Conv2D(filters=256, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\nmodel3.add(MaxPool2D(pool_size=(2,2),strides=(2,2)))\nmodel3.add(Conv2D(filters=512, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\nmodel3.add(Conv2D(filters=512, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\nmodel3.add(Conv2D(filters=512, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\nmodel3.add(MaxPool2D(pool_size=(2,2),strides=(2,2)))\nmodel3.add(Conv2D(filters=512, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\nmodel3.add(Conv2D(filters=512, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\nmodel3.add(Conv2D(filters=512, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\nmodel3.add(MaxPool2D(pool_size=(2,2),strides=(2,2)))","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model3.add(Flatten())\nmodel3.add(Dense(units=512,activation=\"relu\"))\nmodel3.add(Dense(units=2, activation=\"softmax\"))","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"from tensorflow.keras.optimizers import Adam\nopt = Adam(lr=0.001)\nmodel3.compile(optimizer=opt, loss=keras.losses.categorical_crossentropy, metrics=['accuracy'])","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model3.summary()","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"from keras.callbacks import ModelCheckpoint, EarlyStopping\ncheckpoint = ModelCheckpoint(\"vgg16_1.h5\", monitor='val_accuracy', verbose=1, save_best_only=True, save_weights_only=False, mode='auto', period=1)\nearly = EarlyStopping(monitor='val_accuracy', min_delta=0, patience=20, verbose=1, mode='auto')\nhistory = model3.fit(x = x_train,\n                    y = y_train,\n                    validation_data=(x_cross_validation, y_cross_validation),\n                    batch_size=50,\n                    epochs=50,\n                    verbose=1, \n                    callbacks=[checkpoint, early]\n                   )","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\nimport os,cv2\nfrom IPython.display import Image\nfrom keras.preprocessing import image\nfrom keras import optimizers\nfrom keras import layers,models\nfrom keras.applications.imagenet_utils import preprocess_input\nimport matplotlib.pyplot as plt\nimport seaborn as sns\nfrom keras import regularizers\nfrom keras.preprocessing.image import ImageDataGenerator\nfrom keras.applications.vgg16 import VGG16","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_img_dir = '../input/siim-isic-melanoma-classification/jpeg/train/'\ntest_img_dir = '../input/siim-isic-melanoma-classification/jpeg/test/'\ntrain = pd.read_csv('../input/siim-isic-melanoma-classification/train.csv')\ntest = pd.read_csv('../input/siim-isic-melanoma-classification/test.csv')\ntrain1= pd.read_csv('../input/siim-isic-melanoma-classification/train.csv')[1:1000]\ntest1 = pd.read_csv('../input/siim-isic-melanoma-classification/test.csv')","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train1.target.value_counts()","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train1['target'] = train1['target'].astype(str)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"datagen=ImageDataGenerator(rescale=1./255)\nbatch_size=50","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"train_generator=datagen.flow_from_dataframe(dataframe=train1[:901],directory=train_img_dir,x_col= 'image_name',\n                                            y_col='target',class_mode='raw',batch_size=batch_size,\n                                            target_size=(224,224))\n\n\nvalidation_generator=datagen.flow_from_dataframe(dataframe=train1[900:],directory=train_img_dir,x_col= 'image_name',\n                                                y_col='target',class_mode='raw',batch_size=10,\n                                                target_size=(224,224))","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"# **Building MobileNet** ","metadata":{}},{"cell_type":"code","source":"import tensorflow as tf\nfrom tensorflow.keras.applications.mobilenet import MobileNet\nfrom keras.layers import Dense, GlobalAveragePooling2D,Activation,Flatten\nmodel4 = tf.keras.applications.MobileNet()","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"model4.summary()","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"X=train1.drop('target',axis=1)\ny=train1['target']","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"from imblearn.over_sampling import RandomOverSampler\nros = RandomOverSampler(random_state=42)\nX_res, y_res = ros.fit_resample(X, y)","metadata":{"trusted":true},"outputs":[],"execution_count":null},{"cell_type":"code","source":"#mix X_res and y_res into train dataset again\nX_res['target']=y_res\ntrain_resampled=X_res","metadata":{"trusted":true},"outputs":[],"execution_count":null}]}