Python AI, ML, DL, NLP and GAN exploration playground.
pipenv run python -m <module>- Some keras models are saved and loaded from
models/folder.
pipenv run tensorboard --logdir <path>
- Use
ResNet152V2pretrained model. - Test dataset accuracy:
16/16 ━━━━━━━━━━━━━━━━━━━━ 11s 366ms/step - accuracy: 0.9020 - loss: 0.7031 - Example predictions:
1/1 ━━━━━━━━━━━━━━━━━━━━ 6s 6s/step Class prediction vector [p(0), p(1), p(2), p(3), p(4), p(5)] = [[-2.0099866 -0.56881446 4.5948634 -2.6432185 2.2302372 1.7344184 0.23657088 0.4649722 1.4646876 -0.11805859]] Truth: 2, Class: 2 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 42ms/step Class prediction vector [p(0), p(1), p(2), p(3), p(4), p(5)] = [[ 1.1785551 -1.3516254 0.6351655 -2.5380015 -1.3923934 -0.04097138 0.8621799 -0.7493809 1.1925981 -0.79342335]] Truth: 1, Class: 8 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 41ms/step Class prediction vector [p(0), p(1), p(2), p(3), p(4), p(5)] = [[ 1.0378284 -3.540225 -1.0644982 -4.938359 -1.5945294 1.5830379 -0.3573578 -2.0449076 1.9372692 -0.02878545]] Truth: 3, Class: 8 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 41ms/step Class prediction vector [p(0), p(1), p(2), p(3), p(4), p(5)] = [[ 0.39373627 -3.8968973 -0.22976398 -4.179752 -0.29858652 3.7758834 0.9506325 -1.949536 1.0997412 0.07072492]] Truth: 5, Class: 5 1/1 ━━━━━━━━━━━━━━━━━━━━ 0s 45ms/step Class prediction vector [p(0), p(1), p(2), p(3), p(4), p(5)] = [[-1.0685589 -2.327231 -0.17754719 -3.8118033 3.1382742 3.5833392 0.10239653 -0.71539956 1.3708843 -0.6684125 ]] Truth: 5, Class: 5
- Bidirection LSTM
- Translate from human-readable date locale string to ISO-format date string.
- Architecture:
- Attention Map:
- Unidirection LSTM
- Identify name entities in input seqeunce
- Example:
- Inpuut: "Peter Parker , the White House director of trade and manufacturing policy of U.S , said in an interview on Sunday morning that the White House was working to prepare for the possibility of a second wave of the coronavirus in the fall , though he said it wouldn ’t necessarily come"
- Output:
Peter: B-per Parker: I-per White: B-org House: I-org Sunday: B-tim morning: I-tim White: B-org House: I-org coronavirus: B-org ’t: I-per necessarily: I-per
- Sentiment analysis of input sentence with an emoji output using GloVe vectors.
- Examples (Some of them are mislabelled):
The meal was great!: 🍴 I had a tough day!: ❤️ The job looks interesting!: 😄 I had a great trip!: 🍴 I learnt something new today!: ❤️
- Text similarity classifier
- Examples:
- "When will I see you?", "When can I see you again?" Prediction: True
- "Do they enjoy eating the desert?", "Do they like hiking in the desert?" Prediction: False
- Encoder-Decoder Transformer architecture
- Features:
'Age', 'RestingBP', 'Cholesterol', 'FastingBS', 'MaxHR', 'Oldpeak', 'Sex_F', 'Sex_M', 'ChestPainType_ASY', 'ChestPainType_ATA', 'ChestPainType_NAP', 'ChestPainType_TA', 'RestingECG_LVH', 'RestingECG_Normal', 'RestingECG_ST', 'ExerciseAngina_N', 'ExerciseAngina_Y', 'ST_Slope_Down', 'ST_Slope_Flat', 'ST_Slope_Up'
- Model explaination of contributing factors to heart disease:
- The red sections on the left are features which push the model towards the final prediction in the positive direction (i.e. a higher Age increases the predicted risk).
- The blue sections on the right are features that push the model towards the final prediction in the negative direction (if an increase in a feature leads to a lower risk, it will be shown in blue).
- Predicting the 10-year risk of death of individuals from the NHANES | epidemiology dataset.
- Model: Random Forest
- Features:
'Age', 'Diastolic BP', 'Poverty index', 'Race', 'Red blood cells', 'Sedimentation rate', 'Serum Albumin', 'Serum Cholesterol', 'Serum Iron', 'Serum Magnesium', 'Serum Protein', 'Sex', S'ystolic BP', 'TIBC', 'TS', 'White blood cells', 'BMI', 'Pulse pressure' - Model explaination of contributing factors to 10-year risk of death:
- The red sections on the left are features which push the model towards the final prediction in the positive direction (i.e. a higher Age increases the predicted risk).
- The blue sections on the right are features that push the model towards the final prediction in the negative direction (if an increase in a feature leads to a lower risk, it will be shown in blue).
-
Visualize the impact of each region of an image on a specific output for a Convolutional Neural Network model, DenseNet121 in this case with a pretrained model.
-
Generate a heatmap by computing gradients of the specific class scores we are interested in visualizing.
-
Features:
'Age', 'RestingBP', 'Cholesterol', 'FastingBS', 'MaxHR', 'Oldpeak', 'Sex_F', 'Sex_M', 'ChestPainType_ASY', 'ChestPainType_ATA', 'ChestPainType_NAP', 'ChestPainType_TA', 'RestingECG_LVH', 'RestingECG_Normal', 'RestingECG_ST', 'ExerciseAngina_N', 'ExerciseAngina_Y', 'ST_Slope_Down', 'ST_Slope_Flat', 'ST_Slope_Up'Image: 00016650_000.png Ground Truth: 0, Cardiomegaly Generating heatmap for class Cardiomegaly, prediction: Cardiomegaly (p=0.9047) Generating heatmap for class Mass, prediction: Cardiomegaly (p=0.2098) Generating heatmap for class Edema, prediction: Cardiomegaly (p=0.0489)
Image: 00005410_000.png Ground Truth: 5, Mass Generating heatmap for class Cardiomegaly, prediction: Mass (p=0.0091) Generating heatmap for class Mass, prediction: Mass (p=0.9355) Generating heatmap for class Edema, prediction: Mass (p=0.0848)
Image: 00004090_002.png Ground Truth: 12, Edema Generating heatmap for class Cardiomegaly, prediction: Edema (p=0.3357) Generating heatmap for class Mass, prediction: Edema (p=0.1737) Generating heatmap for class Edema, prediction: Edema (p=0.8023)
