updated readme

readme.md

@@ -1,13 +1,17 @@
-# Machine learning for material recognition with a TENG
-(Bi)LSTM for name classification.  
-More information on the project are [on my website](https://quintern.xyz/en/teng.html).
+# Machine learning for material recognition with a triboelectric nanogenerator (TENG)
+This project was written for my bachelor's thesis.
+
+It was written to classify TENG voltage output from pressing it against different materials.
+Contents:
+- Data preparation/plotting/loading utilites
+- (Bi)LSTM + fully connected + softmax model for name classifiying TENG output
+- Progress tracking utilities to easily find the best parameters
 
 ## Model training
-Adjust the parameters in `main.py` and run it.  
+Adjust the parameters in `main.py` and run it.
 All models and the settings they were trained with are automatically serialized with pickle and stored in a subfolder
 of the `<model_dir>` that was set in `main.py`.
 
 
 ## Model evaluation
 Run `find_best_model.py <model_dir>` with the `<model_dir>` specified in `main.py` during training.
-

teng_ml/main.py

@@ -42,9 +42,10 @@ def test_interpol():
 
 
 if __name__ == "__main__":
-    labels = LabelConverter(["foam_PDMS_white", "foam_PDMS_black", "foam_PDMS_TX100", "foam_PE", "antistatic_foil", "cardboard", "glass", "kapton", "bubble_wrap_PE", "fabric_PP", ])
-    # labels = LabelConverter(["foam_PDMS_white", "foam_PDMS_black", "foam_PDMS_TX100", "foam_PE", "kapton", "bubble_wrap_PE", "fabric_PP", ])
-    models_dir = "/home/matth/Uni/TENG/teng_2/models_gen_12"  # where to save models, settings and results
+    # labels = LabelConverter(["foam_PDMS_white", "foam_PDMS_black", "foam_PDMS_TX100", "foam_PE", "antistatic_foil", "cardboard", "glass", "kapton", "bubble_wrap_PE", "fabric_PP" ])
+    labels = LabelConverter(["foam_PDMS_white", "foam_PDMS_black", "foam_PDMS_TX100", "foam_PE", "antistatic_foil", "cardboard", "kapton", "bubble_wrap_PE", "fabric_PP" ])
+    # labels = LabelConverter(["foam_PDMS_white", "foam_PDMS_black", "foam_PDMS_TX100", "foam_PE", "kapton", "bubble_wrap_PE", "fabric_PP" ])
+    models_dir = "/home/matth/Uni/TENG/teng_2/models_gen_15"  # where to save models, settings and results
     if not path.isdir(models_dir):
         makedirs(models_dir)
     data_dir = "/home/matth/Uni/TENG/teng_2/sorted_data"
@@ -53,18 +54,18 @@ if __name__ == "__main__":
     # gen_6 best options: no glass, cardboard and antistatic_foil, not bidirectional, lr=0.0007, no datasplitter, 2 layers n_hidden = 10
 
     # Test with
-    num_layers = [ 2, 3 ]
-    hidden_size = [ 21, 28 ]
-    bidirectional = [ False, True ]
+    num_layers = [ 4, 5 ]
+    hidden_size = [ 28, 36 ]
+    bidirectional = [ True ]
     t_const_int = ConstantInterval(0.01)  # TODO check if needed: data was taken at equal rate, but it isnt perfect -> maybe just ignore?
     t_norm = Normalize(-1, 1)
-    transforms = [[ t_norm ]]  #, [ t_norm, t_const_int ]]
+    transforms = [[]] #, [ t_norm, t_const_int ]]
     batch_sizes = [ 4 ]
     splitters = [ DataSplitter(50, drop_if_smaller_than=30) ]  # smallest file has length 68 TODO: try with 0.5-1second snippets
     num_epochs = [ 80 ]
     # (epoch, min_accuracy)
-    training_cancel_points = [(15, 20),  (40, 25)]
-    # training_cancel_points = []
+    # training_cancel_points = [(15, 20),  (40, 25)]
+    training_cancel_points = []
 
     args = [num_layers, hidden_size, bidirectional, [None], [None], [None], transforms, splitters, num_epochs, batch_sizes]
 
@@ -81,7 +82,7 @@ if __name__ == "__main__":
         None,
         # lambda optimizer, st: torch.optim.lr_scheduler.ExponentialLR(optimizer, gamma=0.9),
         # lambda optimizer, st: torch.optim.lr_scheduler.ExponentialLR(optimizer, gamma=0.5),
-        lambda optimizer, st: torch.optim.lr_scheduler.StepLR(optimizer, step_size=st.num_epochs // 8, gamma=0.60, verbose=False),
+        # lambda optimizer, st: torch.optim.lr_scheduler.StepLR(optimizer, step_size=st.num_epochs // 8, gamma=0.60, verbose=False),
         # lambda optimizer, st: torch.optim.lr_scheduler.StepLR(optimizer, step_size=st.num_epochs // 10, gamma=0.75, verbose=False),
     ]
 

teng_ml/util/data_loader.py

@@ -10,8 +10,13 @@ import threading
 
 from sklearn.model_selection import train_test_split
 
+from teng_ml.util.transform import Multiply
+
 # groups: date, name, n_object, voltage, distance, index
 # re_filename = r"(\d{4}-\d{2}-\d{2})_([a-zA-Z_]+)_(\d{1,2}(?:\.\d*)?)V_(\d+(?:\.\d*)?)mm(\d+).csv"
+# for teng_1
+# re_filename = r"(\d{4}-\d{2}-\d{2})_([a-zA-Z_]+)_()(\d{1,2}(?:\.\d*)?)V_(\d+(?:\.\d*)?)mm(\d+).csv"
+# for teng_2
 re_filename = r"(\d{4}-\d{2}-\d{2})_([a-zA-Z0-9_]+)_(\d+)_(\d{1,2}(?:\.\d*)?)V_(\d+(?:\.\d*)?)mm(\d+).csv"
 
 class LabelConverter:
@@ -51,7 +56,7 @@ class Datasample:
     def __init__(self, date: str, label: str, n_object: str, voltage: str, distance: str, index: str, label_vec, datapath: str, init_data=False):
         self.date = date
         self.label = label
-        self.n_object = int(n_object)
+        self.n_object = 0 if n_object == "" else int(n_object)
         self.voltage = float(voltage)
         self.distance = float(distance)
         self.index = int(index)
@@ -86,6 +91,19 @@ class Dataset:
         """
         self.transforms = transforms
         self.data = []  # (data, label)
+
+        # NORMALIZE ALL DATA WITH THE SAME FACTOR
+        # sup = 0
+        # inf = 0
+        # for sample in datasamples:
+        #     data = sample.get_data()
+        #     max_ = np.max(data[:,2])
+        #     min_ = np.min(data[:,2])
+        #     if max_ > sup: sup = max_
+        #     if min_ < inf: inf = min_
+        # multiplier = 1 / max(sup, abs(inf))
+        # self.transforms.append(Multiply(multiplier))
+
         for sample in datasamples:
             data = self.apply_transforms(sample.get_data())
             if split_function is None:
@@ -128,7 +146,7 @@ def get_datafiles(datadir, labels: LabelConverter, exclude_n_object=None, filter
         label = match.groups()[1]
         if label not in labels: continue
 
-        sample_n_object = float(match.groups()[2])
+        sample_n_object = 0 if match.groups()[2] == "" else int(match.groups()[2])
         if exclude_n_object and exclude_n_object == sample_n_object: continue
         sample_voltage = float(match.groups()[3])
         if filter_voltage and filter_voltage != sample_voltage: continue

teng_ml/util/transform.py

@@ -1,5 +1,6 @@
 import numpy as np
 from scipy.interpolate import interp1d
+from torch import mul
 
 class Normalize:
     """
@@ -41,6 +42,15 @@ class NormalizeAmplitude:
         return f"NormalizeAmplitude(high={self.high})"
 
 
+class Multiply:
+    def __init__(self, multiplier):
+        self.multiplier = multiplier
+    def __call__(self, data):
+        return data * self.multiplier
+    def __repr__(self):
+        return f"Multiply(multiplier={self.multiplier})"
+
+
 class ConstantInterval:
     """
     Interpolate the data to have a constant interval / sample rate,