Mpd_classification_reporterge branch '20201008_landmarkv2' of github.…

…com:3springs/deep_ml_curriculum into 20201008_landmarkv2
3springs · Oct 13, 2020 · 27fb846 · 27fb846
2 parents 7435a24 + 5a95008
commit 27fb846
Show file tree

Hide file tree

Showing 6 changed files with 1,707 additions and 463 deletions.
diff --git a/notebooks/c02_Intro_to_NN_Part_2/Intro_to_NN_Part_2.ipynb b/notebooks/c02_Intro_to_NN_Part_2/Intro_to_NN_Part_2.ipynb
diff --git a/notebooks/c02_Intro_to_NN_Part_2/Intro_to_NN_Part_2.py b/notebooks/c02_Intro_to_NN_Part_2/Intro_to_NN_Part_2.py
@@ -120,13 +120,14 @@
 # Credits to researchers at Georgia Tech, Agile Geoscience
 # License CCbySA
 #
-# In this notebook, we will be using the landmass dataset, which have been preprocessed already. In this dataset, we have images of 4 different types of landmass: 'Chaotic Horizon', 'Fault', 'Horizon', 'Salt Dome'.
+# In this notebook, we will be using the landmass dataset, which have been preprocessed already. In this dataset, we have images of 4 different types of landmass: ['Discontinuous', 'Faulted', 'Continuous', 'Salt'].
 #
 # This is an example of [seismic data](https://en.wikipedia.org/wiki/Reflection_seismology) which is a way of using seismic to image the structure of the Earth, below the surface. These waves are similar to sounds waves in air. The lines represent changes in density below the surface.
 #
 # We will train a CNN to learn how to classify images into those 4 groups.
 
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+device
 
 # +
 # Let's import the Patches
@@ -157,6 +158,10 @@
 print("Class:", landmassf3_train.classes[y])
 x
 
+# Note that this is an unbalanced dataset, so we expect an accuracy of at least 52%, this is out baseline
+labels = pd.Series(landmassf3_train.train_labels).replace(dict(enumerate(landmassf3_train.classes)))
+labels.value_counts() / len(landmassf3_train)
+
 landmassf3_train.classes
 
 # Source: [Neural Networks](https://pytorch.org/tutorials/beginner/blitz/neural_networks_tutorial.html#sphx-glr-beginner-blitz-neural-networks-tutorial-py)
@@ -400,8 +405,8 @@ def forward(self, x):
 
 convnet = BetterCNN().to(device)
 optimizer = torch.optim.Adam(convnet.parameters(), lr=learning_rate)
-model = train(convnet, x_train, y_train, criterion, optimizer)
-test(model, x_test, y_test)
+convnet = train(convnet, x_train, y_train, criterion, optimizer)
+test(convnet, x_test, y_test)
 
 from deep_ml_curriculum.torchsummaryX import summary
 # We can also summarise the number of parameters in each layer

diff --git a/notebooks/c03_Finetuning/Finetuning.ipynb b/notebooks/c03_Finetuning/Finetuning.ipynb