00:23:54 Jennifer Woodward-Greene: There are 6 classified correctly in 'emu'. There are 7 classified correctly in 'flamingo'. There were not false negatives or false positives. 00:24:14 ARS - Kossi Nouwakpo: perfect match 00:27:11 Jennifer Woodward-Greene: it is! 00:27:28 ARS - Kossi Nouwakpo: When I run the prediction on X_train, there is one mismatch. 00:29:30 Jennifer Woodward-Greene: it is calling an emu a flamingo 00:31:00 Lucas Heintzman: How does this "accuracy" correspond to a "Kappa Value"? 00:32:44 Lucas Heintzman: Just a thought... no reason to spend time on it. 00:35:44 ARS - Kossi Nouwakpo: Are the number of features limited by the size of the training set? i.e. can I have more features than training samples? 00:36:39 ARS - Kossi Nouwakpo: Thank you 00:37:28 Matthew.McEntire: Is there an "easy" way to identify which specific picture is being miss identified? 00:40:58 Andrew Russ: Emu calssified as hedgehog 00:44:42 Amy Hudson (she/her): i’m getting an error when I plug airplanes in … anyone else? 00:45:26 Amy Hudson (she/her): ok, thanks! 00:45:39 Andrew Russ: Cap A 00:57:31 Laura Boucheron: filenames_train = list() filenames_test = list() 00:58:07 Laura Boucheron: filenames_train.append(im_filename) filenames_test.append(im_filename) 01:00:32 Matthew.McEntire: Thank you 01:03:33 Amy Hudson (she/her): would we be able to weight our feature matrix? like value shape more over color? or add a feature vector of weights? 01:07:12 Laura Boucheron: misclass_idx = np.where(y_train!=y_train_hat) for i in misclass_idx[0]: print('The image '+filenames_train[i]+' is misclassified') plt.figure(figsize=(5,5)) I = imageio.imread(filenames_train[i]) plt.imshow(I) plt.show() 01:07:46 Lucas Heintzman: The code errors when you attempt to discriminate the same category twice. 01:08:10 Elizabeth Chin: I got the following error when using cougar_face and cougar_body: /anaconda3/envs/aiworkshop/lib/python3.7/site-packages/ipykernel_launcher.py:10: RuntimeWarning: Mean of empty slice. 01:08:18 Lucas Heintzman: Yes 01:09:28 YAKOV PACHEPSKY: Some objects may be classified better than others. What may be surmised from this info? 01:12:22 Elizabeth Chin: ok thanks for checking! good to know for the future 01:14:32 YAKOV PACHEPSKY: Thank you 01:14:58 Jennifer Woodward-Greene: How hard to plot the decision boundary? 01:16:53 Lucas Heintzman: Is there a means to produce a figure similar to the PowerPoint slide that shows the decision boundary and the filenames of which images fall out? 01:19:13 Elizabeth Chin: is it recommended to do feature selection for image features? 01:20:25 Elizabeth Chin: by feature selection i mean something like Recursive Feature Selection, or removing correlated features (I’m not sure if extracting the rgb features is also considered ‘feature selection’) 01:22:30 YAKOV PACHEPSKY: CAn one use the edge extraction? 01:27:56 Elizabeth Chin: yes! thank you 01:33:38 Kerrie Geil: Laura that answer was super helpful for me in thinking about how to approach my research. Thanks! 01:35:56 Kevin Cole: empty 01:38:11 Kerrie Geil: can you type this image preprocessing software or a link to it in the chat? 01:38:34 Laura Boucheron: ImageJ 01:38:50 Laura Boucheron: FIJI 01:39:28 Elizabeth Chin: this?: https://imagej.net/Fiji 01:39:50 Laura Boucheron: Yes--that's it 01:40:28 Kerrie Geil: I guess this is the general link https://imagej.nih.gov/ij/index.html 01:43:15 Jerry M: ImageJ + $200 desktop scanner makes a great particle size analysis tool. I've measured the sizes of seed, dust particles, and pellets this way. 01:43:21 Laura Boucheron: One plugin for interactive object segmentation: https://imagej.net/SIOX:_Simple_Interactive_Object_Extraction 01:49:54 ARS - Kossi Nouwakpo: Thank you 01:49:59 Elizabeth Chin: Thank you! 01:50:03 Lucas Heintzman: Any special data/formatting for Friday? 01:50:17 Maximilian Feldman: Did we talk about how to see the importance of each feature for classification? 01:50:38 YAKOV PACHEPSKY: Thank you very much 01:50:59 Laura Boucheron: A feature selectio method would give you some insight into importance of each feature for classificaiton. 01:52:49 Matthew.McEntire: thanks