๐EDA ๐
For each of the label and text in the graphs its bounding box has been provided so, that we can train our object detection model on it.
","metadata":{"id":"467NdINKV4ia"}},{"cell_type":"code","source":"df = pd.DataFrame({'image_path':train_images})\n\ndf['train_annotations'] = df['image_path'].str.replace('images', 'annotations')\ndf['train_annotations'] = df['train_annotations'].str.replace('jpg', 'json')\n\ndf.head()","metadata":{"id":"WVdGq4dkRltB","outputId":"35e3dbae-f462-4afa-b358-b11f6d8c620f","execution":{"iopub.status.busy":"2023-03-26T14:35:35.096040Z","iopub.execute_input":"2023-03-26T14:35:35.096421Z","iopub.status.idle":"2023-03-26T14:35:35.281480Z","shell.execute_reply.started":"2023-03-26T14:35:35.096384Z","shell.execute_reply":"2023-03-26T14:35:35.279529Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"chart_type, source, height, width = [], [], [], []\nfor annotations in tqdm(df['train_annotations']):\n annot = open(annotations)\n annot = json.load(annot)\n \n chart_type.append(annot['chart-type'])\n source.append(annot['source'])\n height.append(annot['plot-bb']['height'])\n width.append(annot['plot-bb']['width'])\n\ndf['chart_type'] = chart_type\ndf['source'] = source\ndf['height'] = height\ndf['width'] = width","metadata":{"id":"6fYAGY2VR2Xh","outputId":"3ecf69a0-aedd-4e77-9469-bdebec980e36","execution":{"iopub.status.busy":"2023-03-26T14:35:35.282660Z","iopub.execute_input":"2023-03-26T14:35:35.285231Z","iopub.status.idle":"2023-03-26T14:39:36.695261Z","shell.execute_reply.started":"2023-03-26T14:35:35.285170Z","shell.execute_reply":"2023-03-26T14:39:36.693969Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df.head()","metadata":{"id":"thx5UCu1Wztm","outputId":"ff07b433-d9af-40f6-bc73-fa7c05639886","execution":{"iopub.status.busy":"2023-03-26T14:39:36.699071Z","iopub.execute_input":"2023-03-26T14:39:36.699486Z","iopub.status.idle":"2023-03-26T14:39:36.714237Z","shell.execute_reply.started":"2023-03-26T14:39:36.699451Z","shell.execute_reply":"2023-03-26T14:39:36.712973Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plt.subplots(figsize=(15, 10))\nfor i in range(16):\n plt.subplot(4, 4, i+1)\n\n # Selecting a random image\n # index from the dataframe.\n k = np.random.randint(0, len(df))\n img = cv2.imread(df.loc[k, 'image_path'])\n plt.imshow(img)\n plt.title(df.loc[k, 'chart_type'])\n # plt.axis('off')\nplt.show()","metadata":{"id":"_bpAEM_GTfZY","outputId":"968dd246-6c9a-43d4-ad9d-a085d608e87e","execution":{"iopub.status.busy":"2023-03-26T14:39:36.715735Z","iopub.execute_input":"2023-03-26T14:39:36.716093Z","iopub.status.idle":"2023-03-26T14:39:38.924385Z","shell.execute_reply.started":"2023-03-26T14:39:36.716060Z","shell.execute_reply":"2023-03-26T14:39:38.923155Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"hor_bar = df[df['chart_type']=='horizontal_bar']['image_path'].values\n\nplt.figure(figsize=(14, 7))\nimg = plt.imread(hor_bar[10])\nplt.imshow(img)\nplt.show()","metadata":{"id":"bpr1chT_RaSP","outputId":"95003225-4940-40b1-8831-50e85bc3a11f","execution":{"iopub.status.busy":"2023-03-26T14:39:38.925928Z","iopub.execute_input":"2023-03-26T14:39:38.926363Z","iopub.status.idle":"2023-03-26T14:39:39.403235Z","shell.execute_reply.started":"2023-03-26T14:39:38.926320Z","shell.execute_reply":"2023-03-26T14:39:39.401915Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plt.figure(figsize=(14, 10))\nimg_ver = cv2.rotate(img, cv2.ROTATE_90_COUNTERCLOCKWISE)\nplt.imshow(img_ver)\nplt.show()","metadata":{"id":"4RWHnAZ0SCzG","outputId":"9b3f2ca8-3de6-4d83-e7ae-02801e9c94eb","execution":{"iopub.status.busy":"2023-03-26T14:39:39.404878Z","iopub.execute_input":"2023-03-26T14:39:39.405333Z","iopub.status.idle":"2023-03-26T14:39:39.728750Z","shell.execute_reply.started":"2023-03-26T14:39:39.405286Z","shell.execute_reply":"2023-03-26T14:39:39.727466Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"So, by rotating a horizontal bar chart 90 degrees anit clockwise we can convert it into a vertical bar chart this can certainly help us solve the problem of data imbalance. But this would be a feasible solution if we were to output only bar_chart instead of vertical or horizontal bar chart so, we will have to train our model on this class as well.
","metadata":{"id":"mvc-tw_6VJEn"}},{"cell_type":"code","source":"df['chart_type'].value_counts()","metadata":{"id":"FWz3MYg2W4uN","outputId":"7f49e03f-f93a-4224-ac8d-d6c413530520","execution":{"iopub.status.busy":"2023-03-26T14:39:39.730248Z","iopub.execute_input":"2023-03-26T14:39:39.730622Z","iopub.status.idle":"2023-03-26T14:39:39.749998Z","shell.execute_reply.started":"2023-03-26T14:39:39.730588Z","shell.execute_reply":"2023-03-26T14:39:39.748759Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df.describe()[['height', 'width']].T","metadata":{"id":"Ck3KchrtQfOm","outputId":"3acf2d5a-71d2-4391-acfd-4464bd88ca8e","execution":{"iopub.status.busy":"2023-03-26T14:39:39.751586Z","iopub.execute_input":"2023-03-26T14:39:39.752024Z","iopub.status.idle":"2023-03-26T14:39:39.792074Z","shell.execute_reply.started":"2023-03-26T14:39:39.751986Z","shell.execute_reply":"2023-03-26T14:39:39.790791Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plt.subplots(figsize=(10, 5))\n\nplt.subplot(1, 2, 1)\nsb.boxplot(df['height'])\n\nplt.subplot(1, 2, 2)\nsb.boxplot(df['width'])\n\nplt.show()","metadata":{"id":"4z2n6AMnXXhq","outputId":"75539ebf-175e-4c3f-d6bf-80d350d0d7f3","execution":{"iopub.status.busy":"2023-03-26T14:39:39.793904Z","iopub.execute_input":"2023-03-26T14:39:39.794392Z","iopub.status.idle":"2023-03-26T14:39:40.057449Z","shell.execute_reply.started":"2023-03-26T14:39:39.794342Z","shell.execute_reply":"2023-03-26T14:39:40.056012Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"ss = pd.read_csv(f'{PATH}/sample_submission.csv')\nss","metadata":{"id":"OsqsQ1cbZ5IO","outputId":"223515aa-92f3-409f-b37b-7587952e6175","execution":{"iopub.status.busy":"2023-03-26T14:39:40.059537Z","iopub.execute_input":"2023-03-26T14:39:40.060054Z","iopub.status.idle":"2023-03-26T14:39:40.080817Z","shell.execute_reply.started":"2023-03-26T14:39:40.060003Z","shell.execute_reply":"2023-03-26T14:39:40.079810Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Here we can see that for each image id we can predict the chart type with the help of a image classification model but for the data series we need to train a object detection for which we have been provided with the annotations as well.
","metadata":{"id":"hdi-6vS8UqVl"}},{"cell_type":"code","source":"plt.figure(figsize=(14, 6))\nimage = plt.imread(f'{PATH}/train/images/2501ab594a7c.jpg')\nplt.imshow(image)\nplt.show()","metadata":{"id":"q7M2J4ZYaQ4l","outputId":"c5be43ca-c256-4bcc-be48-2f6c7408a0ce","execution":{"iopub.status.busy":"2023-03-26T14:39:40.081928Z","iopub.execute_input":"2023-03-26T14:39:40.082882Z","iopub.status.idle":"2023-03-26T14:39:40.374808Z","shell.execute_reply.started":"2023-03-26T14:39:40.082842Z","shell.execute_reply":"2023-03-26T14:39:40.373786Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"1. 2501ab594a7c_x - Holt;Hooker;Howard;Jefferson;...\n2. 2501ab594a7c_y - 20;73;51;63;....\n\nSo, this will be the contents of two rows in the sample submission dataframe in the column data_series.
","metadata":{"id":"KXPtQsRSUB30"}}]}