I have tried to profile layer-by-layer of DenseNet in Pytorch as caffe-time tool.
First trial : using autograd.profiler like below
...
model = models.__dict__['densenet121'](pretrained=True)
model.to(device)
with torch.autograd.profiler.profile(use_cuda=True) as prof:
model.eval()
print(prof)
...
But the any results are shown except for this message :
<unfinished torch.autograd.profile>
Ultimately, I want to profile network archtiectures(i.g.DenseNet) to check where bottlenecks happen.
Could anyone do this?
To run profiler you have do some operations, you have to input some tensor into your model.
Change your code as following.
import torch
import torchvision.models as models
model = models.densenet121(pretrained=True)
x = torch.randn((1, 3, 224, 224), requires_grad=True)
with torch.autograd.profiler.profile(use_cuda=True) as prof:
model(x)
print(prof)
This is the sample of the output I got:
----------------------------------- --------------- --------------- --------------- --------------- ---------------
Name CPU time CUDA time Calls CPU total CUDA total
----------------------------------- --------------- --------------- --------------- --------------- ---------------
conv2d 9976.544us 9972.736us 1 9976.544us 9972.736us
convolution 9958.778us 9958.400us 1 9958.778us 9958.400us
_convolution 9946.712us 9947.136us 1 9946.712us 9947.136us
contiguous 6.692us 6.976us 1 6.692us 6.976us
empty 11.927us 12.032us 1 11.927us 12.032us
mkldnn_convolution 9880.452us 9889.792us 1 9880.452us 9889.792us
batch_norm 1214.791us 1213.440us 1 1214.791us 1213.440us
native_batch_norm 1190.496us 1193.056us 1 1190.496us 1193.056us
threshold_ 158.258us 159.584us 1 158.258us 159.584us
max_pool2d_with_indices 28837.682us 28836.834us 1 28837.682us 28836.834us
max_pool2d_with_indices_forward 28813.804us 28822.530us 1 28813.804us 28822.530us
batch_norm 1780.373us 1778.690us 1 1780.373us 1778.690us
native_batch_norm 1756.774us 1759.327us 1 1756.774us 1759.327us
threshold_ 64.665us 66.368us 1 64.665us 66.368us
conv2d 6103.544us 6102.142us 1 6103.544us 6102.142us
convolution 6089.946us 6089.600us 1 6089.946us 6089.600us
_convolution 6076.506us 6076.416us 1 6076.506us 6076.416us
contiguous 7.306us 7.938us 1 7.306us 7.938us
empty 9.037us 8.194us 1 9.037us 8.194us
mkldnn_convolution 6015.653us 6021.408us 1 6015.653us 6021.408us
batch_norm 700.129us 699.394us 1 700.129us 699.394us
There are many rows below this.
I have used (1,3,224,224) tensor as densenet only accepts 224x224 images. In the future change tensor size according to the network.
Related
this is what i have (a 3D list) :
**
[ STOCK NAME
last_price price2 price3
0 0.00 0.0 0.0
1 870.95 7650.0 2371500.0
2 870.95 7650.0 2371500.0
3 870.95 7650.0 2371500.0
4 877.30 7650.0 2371500.0
5 879.20 6800.0 2381700.0]
**
I want to create a dataframe exactly like the list that I have above. how do I do so? thank you very much.. i tried pd.DataFrame(the_list) but it gave me this error: ValueError: Must pass 2-d input. shape=(190, 6, 3).. thanks
I've a dataframe which looks like this:
wave mean median mad
0 4050.32 -0.016182 -0.011940 0.008885
1 4208.98 0.023707 0.007189 0.032585
2 4508.28 3.662293 0.001414 7.193139
3 4531.62 -15.459313 -0.001523 30.408377
4 4551.65 0.009028 0.007581 0.005247
5 4554.46 0.001861 0.010692 0.027969
6 6828.60 -10.604568 -0.000590 21.084799
7 6839.84 -0.003466 -0.001870 0.010169
8 6842.04 -32.751551 -0.002514 65.118329
9 6842.69 18.293519 -0.002158 36.385884
10 6843.66 0.006386 -0.002468 0.034995
11 6855.72 0.020803 0.000886 0.040529
As it's clearly evident in the above table that some of the values in the column mad and median are very big(outliers). So i want to remove the rows which have these very big values.
For example in row3 the value of mad is 30.408377 which very big so i want to drop this row. I know that i can use one line
to remove these values from the columns but it doesn't removes the complete row
df[np.abs(df.mad-df.mad.mean()) <= (3*df.mad.std())]
But i want to remove the complete row.
How can i do that?
Predicates like what you've given will remove entire rows. But none of your data is outside of 3 standard deviations. If you tone it down to just one standard deviation, rows are removed with your example data.
Here's an example using your data:
import pandas as pd
import numpy as np
columns = ["wave", "mean", "median", "mad"]
data = [
[4050.32, -0.016182, -0.011940, 0.008885],
[4208.98, 0.023707, 0.007189, 0.032585],
[4508.28, 3.662293, 0.001414, 7.193139],
[4531.62, -15.459313, -0.001523, 30.408377],
[4551.65, 0.009028, 0.007581, 0.005247],
[4554.46, 0.001861, 0.010692, 0.027969],
[6828.60, -10.604568, -0.000590, 21.084799],
[6839.84, -0.003466, -0.001870, 0.010169],
[6842.04, -32.751551, -0.002514, 65.118329],
[6842.69, 18.293519, -0.002158, 36.385884],
[6843.66, 0.006386, -0.002468, 0.034995],
[6855.72, 0.020803, 0.000886, 0.040529],
]
df = pd.DataFrame(np.array(data), columns=columns)
print("ORIGINAL: ")
print(df)
print()
res = df[np.abs(df['mad']-df['mad'].mean()) <= (df['mad'].std())]
print("REMOVED: ")
print(res)
this outputs:
ORIGINAL:
wave mean median mad
0 4050.32 -0.016182 -0.011940 0.008885
1 4208.98 0.023707 0.007189 0.032585
2 4508.28 3.662293 0.001414 7.193139
3 4531.62 -15.459313 -0.001523 30.408377
4 4551.65 0.009028 0.007581 0.005247
5 4554.46 0.001861 0.010692 0.027969
6 6828.60 -10.604568 -0.000590 21.084799
7 6839.84 -0.003466 -0.001870 0.010169
8 6842.04 -32.751551 -0.002514 65.118329
9 6842.69 18.293519 -0.002158 36.385884
10 6843.66 0.006386 -0.002468 0.034995
11 6855.72 0.020803 0.000886 0.040529
REMOVED:
wave mean median mad
0 4050.32 -0.016182 -0.011940 0.008885
1 4208.98 0.023707 0.007189 0.032585
2 4508.28 3.662293 0.001414 7.193139
3 4531.62 -15.459313 -0.001523 30.408377
4 4551.65 0.009028 0.007581 0.005247
5 4554.46 0.001861 0.010692 0.027969
6 6828.60 -10.604568 -0.000590 21.084799
7 6839.84 -0.003466 -0.001870 0.010169
10 6843.66 0.006386 -0.002468 0.034995
11 6855.72 0.020803 0.000886 0.040529
Observe that rows indexed 8 and 9 are now gone.
Be sure you're reassigning the output of df[np.abs(df['mad']-df['mad'].mean()) <= (df['mad'].std())] as shown above. The operation is not done in place.
Doing df[np.abs(df.mad-df.mad.mean()) <= (3*df.mad.std())] will not change the dataframe.
But assign it back to df, so that:
df = df[np.abs(df.mad-df.mad.mean()) <= (3*df.mad.std())]
I have a two dimensional list of values:
[
[[12.2],[5325]],
[[13.4],[235326]],
[[15.9],[235326]],
[[17.7],[53521]],
[[21.3],[42342]],
[[22.6],[6546]],
[[25.9],[34634]],
[[27.2],[523523]],
[[33.4],[235325]],
[[36.2],[235352]]
]
I would like to get a list of averages defined by a given step so that for a step=10 it would like like this:
[
[[10],[average of all 10-19]],
[[20],[average of all 20-29]],
[[30],[average of all 30-39]]
]
How can I achieve that? Please note that the number of 10s, 20s, 30s and so on is not always the same.
import pandas as pd
df = pd.DataFrame((q[0][0], q[1][0]) for q in thelist)
df['group'] = (df[0] / 10).astype(int)
Now df is:
0 1 group
0 12.2 5325 1
1 13.4 235326 1
2 15.9 235326 1
3 17.7 53521 1
4 21.3 42342 2
5 22.6 6546 2
6 25.9 34634 2
7 27.2 523523 2
8 33.4 235325 3
9 36.2 235352 3
Then:
df.groupby('group').mean()
Gives you the answers you seek:
0 1
group
1 14.80 132374
2 24.25 151761
3 34.80 235338
I used the below code:
import pandas as pd
pandas_bigram = pd.DataFrame(bigram_data)
print pandas_bigram
I got output as below
0
0 ashoka -**0
1 - wikipedia,**1
2 wikipedia, the**2
3 the free**2
4 free encyclopedia**2
5 encyclopedia ashoka**1
6 ashoka from**2
7 from wikipedia,**1
8 wikipedia, the**2
9 the free**2
10 free encyclopedia**2
My question is How to split this data frame. So, that i will get data in two rows. the data here is separated by "**".
import pandas as pd
df= [" ashoka -**0","- wikipedia,**1","wikipedia, the**2"]
df=pd.DataFrame(df)
print(df)
0
0 ashoka -**0
1 - wikipedia,**1
2 wikipedia, the**2
Use split function: The method split() returns a list of all the words in the string, using str as the separator (splits on all whitespace if left unspecified), optionally limiting the number of splits to num.
df1 = pd.DataFrame(df[0].str.split('*',1).tolist(),
columns = ['0','1'])
print(df1)
0 1
0 ashoka - *0
1 - wikipedia, *1
2 wikipedia, the *2
I have a data frame that is composed of several datasets (about 146 and counting). two of my columns are labeled "start_time" and "stop_time," which represent the start and stop of a response (i.e., the total duration of the response).
I need to get the "inter-response time" or the start_time subtracted from the next corresponding value in start_time. Basically if:
start_time = [1,4,7]
stop_time = [2,5,8]
I need:
stop_time[0] - start_time[1]
stop_time[2] - start_time[3]
in order to get:
iri = [2,2]
My code looks like this:
iri_t = []
def grps():
for grp in lset2_name_grps.groups:
beg_eng_t = pd.DataFrame([lset2_name_grps.stop_time, lset2_name_grps.start_time], columns=['end_t','beg_t'])
end_t = [i for i in lset2_name_grps.stop_time]
beg_t = [i for i in lset2_name_grps.start_time]
beg_t = np.insert(beg_t, len(beg_t),0)
end_t = np.insert(end_t, 0,0)
iri_t.append(np.subtract(end_t, beg_t))
# for i,j in zip(end_t, beg_t):
# iri_t.append(np.subtract(i,j))
# lset2_name_grps['iri'] = iri_t
grps()
Essentially, it doesn't do anything close to what I'm trying to accomplish and the only out I get is either "Not Implemented" or an error.
How about something like this:
import pandas as pd
starts = pd.Series([1, 4, 7])
stops = pd.Series([2, 5, 8])
iri_t = [0]
for i in range(1, len(starts)):
iri_t.append(starts[i] - ends[i-1])
times_df = pd.concat([starts, stops, pd.Series(iri_t)], axis=1)
This creates the following data_frame:
0 1 2
0 1 2 0
1 4 5 2
2 7 8 2
I think what your asking (correct me if I'm wrong) is best accomplished by putting the two columns in a single dataframe, using shift to offset one of your columns, then doing an ordinary subtraction.
df = pd.DataFrame({'start_time':[1,4,7], 'stop_time':[2,5,8]})
df.stop_time - df.start_time.shift()
Out[5]:
0 NaN
1 4
2 4
dtype: float64