Upload 6 files

WiCount.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d578ba4ec4d64ef7089e0fa5c8b47498a76fce5f7d32fb7c7f7f994235915ae1
+size 9389189

data_process_example/README.md

@@ -0,0 +1,21 @@
+# How to Run
+
+1. Execute `process1.py` to convert the `csv` file into a `pkl` file.
+
+2. Run one of the `process2` scripts based on your requirements:
+
+    
+
+   **Note:** We use `-1000` to represent the position of package loss. You can apply various interpolation methods to fill these gaps. We highly encourage you to try our CSI-BERT model to recover the lost packages. ([CSI-BERT](https://github.com/RS2002/CSI-BERT), [CSI-BERT2](https://github.com/RS2002/CSI-BERT2))
+
+    
+
+   (1) If you want to process each record into a long sequence, run `process2.py`. You can refer to `dataset.py` in [CSI-BERT2](https://github.com/RS2002/CSI-BERT2) for guidance.
+
+    
+
+   (2) If you prefer to split each record into multiple fixed-length samples, run `process2-split.py` and modify the `length` parameter in the code to your desired length. You can refer to `dataset.py` in [CSI-BERT](https://github.com/RS2002/CSI-BERT), [CrossFi](https://github.com/RS2002/CrossFi), [KNN-MMD](https://github.com/RS2002/CrossFi), and [LoFi](https://github.com/RS2002/LoFi/tree/main/network_examples) for usage instructions.
+
+    
+
+   (3) `process2-squeeze-split.py` functions similarly to `process2-split.py`, but it excludes all lost packages.

data_process_example/process1.py

@@ -0,0 +1,50 @@
+import numpy as np
+import pickle
+import os
+import pandas as pd
+
+root="./data/"
+data=[]
+csi_vaid_subcarrier_index = range(0, 52)
+
+def handle_complex_data(x, valid_indices):
+    real_parts = []
+    imag_parts = []
+    for i in valid_indices:
+        real_parts.append(x[i * 2])
+        imag_parts.append(x[i * 2 - 1])
+    return np.array(real_parts) + 1j * np.array(imag_parts)
+
+
+for people_num in os.listdir(root):
+    if len(people_num)>1:
+        continue
+    print(people_num)
+    path=os.path.join(root,people_num)
+
+    for file in os.listdir(path):
+        if file[-3:] != "csv":
+            continue
+        print(file)
+        df = pd.read_csv(os.path.join(path,file))
+        df.dropna(inplace=True)
+        df['data'] = df['data'].apply(lambda x: eval(x))
+        complex_data = df['data'].apply(lambda x: handle_complex_data(x, csi_vaid_subcarrier_index))
+        magnitude = complex_data.apply(lambda x: np.abs(x))
+        phase = complex_data.apply(lambda x: np.angle(x, deg=True))
+        time = np.array(df['timestamp'])
+        local_time = np.array(df['local_timestamp'])
+
+        data.append({
+            'csi_time':time,
+            'csi_local_time':local_time,
+            'people_num': eval(people_num),
+            'magnitude': np.array([np.array(a) for a in magnitude]),
+            'phase': np.array([np.array(a) for a in phase]),
+            'CSI': np.array([np.array(a) for a in complex_data])
+        })
+
+# 保存全局字典为一个pickle文件
+output_file = './csi_data.pkl'
+with open(output_file, 'wb') as f:
+    pickle.dump(data, f)

data_process_example/process2-split.py

@@ -0,0 +1,97 @@
+import numpy as np
+import pickle
+import copy
+
+def get_time(s):
+    s=s.split()[-1]
+    s=s.split(":")
+    h=float(s[0])
+    m=float(s[1])
+    t=float(s[2])
+    total=h*3600+m*60+t
+    return h,m,t,total
+
+gap=1
+length=gap*100
+pad=[-1000]*52
+action_list=[]
+people_list=[]
+timestamp=[]
+magnitudes=[]
+phases=[]
+loacl_gap=10000
+
+
+with open("./csi_data.pkl", 'rb') as f:
+    csi = pickle.load(f)
+
+for data in csi:
+    csi_time=data['csi_time']
+    local_time=data['csi_local_time']
+    magnitude=data['magnitude']
+    phase=data['phase']
+    people=data['people_num']
+    action=people
+    start_time=None
+    last_local=None
+    current_magnitude=[]
+    current_phase=[]
+    current_timestamp=[]
+    for i in range(len(csi_time)):
+        _, _, _, current_time = get_time(csi_time[i])
+        if start_time is None or current_time-start_time>gap:
+            if start_time is not None:
+                if len(current_magnitude)>=length:
+                    current_magnitude=current_magnitude[:length]
+                    current_phase=current_phase[:length]
+                    current_timestamp=current_timestamp[:length]
+                else:
+                    add=length-len(current_magnitude)
+                    delta=(current_timestamp[0]+length*loacl_gap-current_timestamp[-1])/add
+                    for j in range(add):
+                        current_magnitude.append(pad)
+                        current_phase.append(pad)
+                        current_timestamp.append(current_timestamp[-1]+delta)
+                magnitudes.append(copy.deepcopy(current_magnitude))
+                phases.append(copy.deepcopy(current_phase))
+                timestamp.append(copy.deepcopy(current_timestamp))
+                action_list.append(action)
+                people_list.append(people)
+                current_magnitude = []
+                current_phase = []
+                current_timestamp = []
+            start_time=current_time
+            last_local=local_time[i]
+            current_magnitude.append(magnitude[i])
+            current_phase.append(phase[i])
+            current_timestamp.append(local_time[i])
+        else:
+            local = local_time[i]
+            num=round((local-last_local-loacl_gap)/loacl_gap)
+            if num>0:
+                delta=(local-last_local)/(num+1)
+                for j in range(num):
+                    current_magnitude.append(pad)
+                    current_phase.append(pad)
+                    current_timestamp.append(current_timestamp[-1] + delta)
+            current_magnitude.append(magnitude[i])
+            current_phase.append(phase[i])
+            current_timestamp.append(local_time[i])
+            last_local=local
+
+
+action_list=np.array(action_list)
+people_list=np.array(people_list)
+timestamp=np.array(timestamp)
+magnitudes=np.array(magnitudes)
+phases=np.array(phases)
+print(action_list.shape)
+print(people_list.shape)
+print(timestamp.shape)
+print(magnitudes.shape)
+print(phases.shape)
+np.save("./magnitude.npy", np.array(magnitudes))
+np.save("./phase.npy", np.array(phases))
+np.save("./action.npy", np.array(action_list))
+np.save("./people.npy", np.array(people_list))
+np.save("./timestamp.npy", np.array(timestamp))

data_process_example/process2-squeeze-split.py

@@ -0,0 +1,54 @@
+import numpy as np
+import pickle
+import copy
+
+
+gap=1
+length=gap*100
+pad=[-1000]*52
+action_list=[]
+people_list=[]
+timestamp=[]
+magnitudes=[]
+phases=[]
+loacl_gap=10000
+
+
+with open("./csi_data.pkl", 'rb') as f:
+    csi = pickle.load(f)
+
+for data in csi:
+    csi_time=data['csi_time']
+    local_time=data['csi_local_time']
+    magnitude=data['magnitude']
+    phase=data['phase']
+    people=data['people_num']
+    action=people
+
+    index=0
+    while index<len(magnitude)-length:
+        current_magnitude=magnitude[index:index+length]
+        current_phase=phase[index:index+length]
+        current_timestamp=local_time[index:index+length]
+        index+=(length+gap-1)
+        magnitudes.append(copy.deepcopy(current_magnitude))
+        phases.append(copy.deepcopy(current_phase))
+        timestamp.append(copy.deepcopy(current_timestamp))
+        action_list.append(action)
+        people_list.append(people)
+
+action_list=np.array(action_list)
+people_list=np.array(people_list)
+timestamp=np.array(timestamp)
+magnitudes=np.array(magnitudes)
+phases=np.array(phases)
+print(action_list.shape)
+print(people_list.shape)
+print(timestamp.shape)
+print(magnitudes.shape)
+print(phases.shape)
+np.save("./squeeze_data/magnitude.npy", np.array(magnitudes))
+np.save("./squeeze_data/phase.npy", np.array(phases))
+np.save("./squeeze_data/action.npy", np.array(action_list))
+np.save("./squeeze_data/people.npy", np.array(people_list))
+np.save("./squeeze_data/timestamp.npy", np.array(timestamp))

data_process_example/process2.py

@@ -0,0 +1,57 @@
+import numpy as np
+import pickle
+
+
+result=[]
+
+pad=[-1000]*52
+loacl_gap=10000
+
+
+with open("./csi_data.pkl", 'rb') as f:
+    csi = pickle.load(f)
+
+for data in csi:
+    csi_time=data['csi_time']
+    local_time=data['csi_local_time']
+    magnitude=data['magnitude']
+    phase=data['phase']
+    people_num=data['people_num']
+
+    last_local=None
+    current_magnitude=[]
+    current_phase=[]
+    current_timestamp=[]
+    for i in range(len(csi_time)):
+        if last_local is None:
+            last_local=local_time[i]
+            current_magnitude.append(magnitude[i])
+            current_phase.append(phase[i])
+            current_timestamp.append(local_time[i])
+        else:
+            local = local_time[i]
+            num=round((local-last_local-loacl_gap)/loacl_gap)
+            if num>0:
+                delta=(local-last_local)/(num+1)
+                for j in range(num):
+                    current_magnitude.append(pad)
+                    current_phase.append(pad)
+                    current_timestamp.append(current_timestamp[-1] + delta)
+            current_magnitude.append(magnitude[i])
+            current_phase.append(phase[i])
+            current_timestamp.append(local_time[i])
+            last_local=local
+
+    print(len(current_magnitude))
+    result.append({
+        'time': np.array(current_timestamp),
+        'action': people_num,
+        'people': people_num,
+        'magnitude': np.array(current_magnitude),
+        'phase': np.array(current_phase)
+    })
+
+output_file = './data_sequence.pkl'
+with open(output_file, 'wb') as f:
+    pickle.dump(result, f)
+