Added rnn

teng-ml/prepare.py

@@ -1,145 +0,0 @@
-import pandas as pd
-import numpy as np
-import scipy.signal as signal
-import matplotlib.pyplot as plt
-from time import sleep
-from random import choice as r_choice
-from sys import exit
-
-from .util.transform import Normalize
-
-if __name__ == "__main__":
-    if __package__ is None:
-        # make relative imports work as described here: https://peps.python.org/pep-0366/#proposed-change
-        __package__ = "teng-ml"
-        import sys
-        from os import path
-        filepath = path.realpath(path.abspath(__file__))
-        sys.path.insert(0, path.dirname(path.dirname(filepath)))
-from .utility.data import load_dataframe
-
-file = "/home/matth/data/2023-04-25_kapton_8.2V_179mm002.csv"
-
-class PeakInfo:
-    """
-    Helper class for "iterating" through selected peaks.
-    """
-    def __init__(self):
-        self.reset()
-
-    def reset(self):
-        self._peak_names = [ "first", "second", "last", "lowest" ]
-        self._peaks = { p: None for p in self._peak_names }
-        self._iter = 0
-
-    def current(self):
-        # return (self._peak_names[self._iter]), self._peaks[self._peak_names[self._iter]]
-        return self._peaks[self._peak_names[self._iter]]
-    def name(self):
-        return self._peak_names[self._iter]
-
-    def next(self):
-        if self._iter < len(self._peak_names) - 1: self._iter += 1
-        return self.current()
-    def prev(self):
-        if self._iter > 0: self._iter -= 1
-        return self.current()
-
-    def set(self, value):
-        """Assign a value to the current peak"""
-        self._peaks[self._peak_names[self._iter]] = value
-    def is_done(self):
-        for peak in self._peaks.values():
-            if peak is None: return False
-        return True
-
-    def __getitem__(self, key):
-        return self._peaks[key]
-    def __setitem__(self, key, value):
-        self._peaks[key] =  value
-    def __repr__(self):
-        return f"{self._peak_names[self._iter]} peak"
-
-
-def find_peaks(a):
-    peaks = signal.find_peaks(a)
-
-
-def on_click(fig, ax, peaks, event):
-    """
-    Let the user select first, second and last peak by clicking on them in this order.
-    Right click undos the last selection
-    """
-    select = None
-    if event.button == 1:  # left click
-        peaks.set((event.xdata, event.ydata))
-        print(f"{peaks}: {event.xdata} - {event.ydata}")
-        ax.set_title(f"{peaks}: {event.xdata} - {event.ydata}")
-        peaks.next()
-    elif event.button == 3:  # right click
-        ax.set_title(f"Undo {peaks.name()}")
-        if not peaks.is_done():
-            peaks.prev()
-        peaks.set(None)
-    if peaks.is_done(): message = "Close window when done"
-    else: message = f"Click on {peaks}"
-    fig.suptitle(message)
-    fig.canvas.draw()
-    # fig1.canvas.flush_events()
-
-def calc_peaks(peaks):
-    # get the peak points from the information of a Peaks object
-    # 90% distance between first and second
-    min_distance = max(1, (peaks["second"][0] - peaks["first"][0]) * 0.9)
-    min_height = peaks["lowest"][1] * 0.99
-    vpeaks = signal.find_peaks(vdata, height=min_height, distance=min_distance)
-    return vpeaks
-
-
-if __name__ == "__main__":
-    """
-    Peak identification:
-    plot, let user choose first, second, last and lowest peak for identification
-    """
-    df = load_dataframe(file)
-    a = df.to_numpy()
-
-    # a2 = interpolate_to_linear_time()
-    # print(a2)
-    # exit()
-
-    vdata = Normalize(0, 1)(a[:,2])
-    plt.ion()
-    # vpeaks[0] is the list of the peaks
-    vpeaks = signal.find_peaks(vdata)[0]
-    fig, ax = plt.subplots()
-    ax.plot(vdata)
-    peak_lines = ax.vlines(vpeaks, 0, 1, colors="r")
-    ax.grid(True)
-    fig.suptitle("Click on first peak")
-    peak_info = PeakInfo()
-    # handle clicks
-    fig.canvas.mpl_connect("button_press_event", lambda ev: on_click(fig, ax, peak_info, ev))
-    # run until user closes, events are handled with on_click function
-    print(vdata.size)
-    while plt.fignum_exists(fig.number):
-        plt.pause(0.01)
-        if (peak_info.is_done()):
-            vpeaks = calc_peaks(peak_info)[0]
-            x_margin = (a[-1,0] - a[0,0]) * 0.05   # allow some margin if user clicked not close enough on peak
-            vpeaks = vpeaks[(vpeaks >= peak_info["first"][0] - x_margin) & (vpeaks <= peak_info["last"][0] + x_margin)]   # remove peaks before first and after last
-            peak_lines.remove()
-            peak_lines = ax.vlines(vpeaks, 0, 1, colors="r")
-            peak_info.reset()
-    print(a[:,0], vpeaks)
-
-    # separate peaks
-    indices = np.arange(0, a[:,0].size)
-    peak_datas = []
-    for i in range(len(vpeaks) - 1):
-        # TODO: user <= or <
-        peak_datas.append(vdata[(indices >= vpeaks[i]) & (indices < vpeaks[i+1])])
-        plt.plot(peak_datas[i])
-    print(peak_datas)
-    plt.pause(20)
-

teng-ml/rnn.py

@@ -0,0 +1,39 @@
+import torch
+import torch.nn as nn
+
+# BiLSTM Model
+
+class RNN(nn.Module):
+  def __init__(self, input_size, hidden_size, num_layers, num_classes, if_bidirectional):
+    super(RNN, self).__init__()
+    self.num_layers = num_layers
+    self.hidden_size = hidden_size
+    self.if_bidirectional = if_bidirectional
+    self.lstm = nn.LSTM(input_size, hidden_size, num_layers, batch_first=True, bidirectional=if_bidirectional)
+
+    if if_bidirectional == True:
+      self.fc = nn.Linear(hidden_size * 2, num_classes)
+    else:
+      self.fc = nn.Linear(hidden_size, num_classes)
+
+
+  def forward(self, x):
+    D = 2 if self.if_bidirectional == True else 1
+    Batch = x.batch_sizes[0]
+
+    h0 = torch.zeros(D * self.num_layers, Batch, self.hidden_size).to(device)
+    c0 = torch.zeros(D * self.num_layers, Batch, self.hidden_size).to(device)
+    x.to(device)
+    _, (h_n, _) = self.lstm(x, (h0, c0))
+    final_state  = h_n.view(self.num_layers, D, Batch, self.hidden_size)[-1]     # num_layers, num_directions, batch, hidden_size
+
+    if D == 1:
+      X = final_state.squeeze()
+    elif D == 2:
+      h_1, h_2 = final_state[0], final_state[1]  # forward & backward pass
+      #X = h_1 + h_2                # Add both states
+      X = torch.cat((h_1, h_2), 1)         # Concatenate both states, X-size: (Batch, hidden_size * 2）
+
+    output = self.fc(X) # fully-connected layer
+
+    return output