I'm trying to train LSTM model in Keras using data of variable timestep, for example, the data looks like:
<tf.RaggedTensor [[[0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]],
[[1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
[1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]],
[[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0]], ...,
[[0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]],
[[1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0],
[1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0],
[1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0],
[1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0],
[1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]],
[[1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
[1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
[1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
[1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]]]>
and its corresponding label:
<tf.RaggedTensor [[6, 6], [7, 7], [8], ..., [6], [11, 11, 11, 11, 11], [24, 24, 24, 24, 24]]>
Each input data have 13 features, so for each time step, the model receives a 1 x 13 vector. I wonder if it is possible to do so? I don't mind doing this on pytorch either.
I try to align them with no reshape layer.
However, my input for each time step in the LSTM layer is a vector of dimension 13. And each sample has variable-length of these vectors, which means the time step is not constant for each sample. Can you show me a code example of how to train such model? –
TurquoiseJ
First of all, the concept of windows length and time steps is they take the same amount of the input with a higher number of length and time.
We assume the input to extract features can be divide by multiple times of windows travels along with axis, please see the attached for idea.
[Codes]:
batched_features = tf.constant( [ [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ], [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ], ], shape=( 2, 1, 13 ) )
batched_labels = tf.constant( [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ], [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]], shape=( 2, 13 ) )
dataset = tf.data.Dataset.from_tensor_slices((batched_features, batched_labels))
dataset = dataset.batch(10)
batched_features = dataset
[Sample]:
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
bidirectional (Bidirectiona (None, 1, 64) 11776
l)
bidirectional_1 (Bidirectio (None, 64) 24832
nal)
dense (Dense) (None, 13) 845
=================================================================
Total params: 37,453
Trainable params: 37,453
Non-trainable params: 0
_________________________________________________________________
<BatchDataset element_spec=(TensorSpec(shape=(None, 1, 13), dtype=tf.int32, name=None), TensorSpec(shape=(None, 13), dtype=tf.int32, name=None))>
Epoch 1/100
2022-03-28 05:19:04.116345: I tensorflow/stream_executor/cuda/cuda_dnn.cc:368] Loaded cuDNN version 8100
1/1 [==============================] - 8s 8s/step - loss: 0.0000e+00 - accuracy: 1.0000 - val_loss: 0.0000e+00 - val_accuracy: 1.0000
Epoch 2/100
1/1 [==============================] - 0s 38ms/step - loss: 0.0000e+00 - accuracy: 1.0000 - val_loss: 0.0000e+00 - val_accuracy: 1.0000
Assume each windows consume about 13 level of the input :
batched_features = tf.constant( [ [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ], [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ], ], shape=( 2, 1, 13 ) )
batched_labels = tf.constant( [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ], [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]], shape=( 2, 13 ) )
Adding more windows is easy by
batched_features = tf.constant( [ [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ], [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ], [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ], ], shape=( 3, 1, 13 ) )
batched_labels = tf.constant( [[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ], [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ], [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ]], shape=( 3, 13 ) )
dataset = tf.data.Dataset.from_tensor_slices((batched_features, batched_labels))
dataset = dataset.batch(10)
batched_features = dataset
At least you tell me what is the purpose they can use reverse windows to have certain results. ( Apmplitues frequency )
The results will look like these for each windows :
[ Output ] : 2 and 3 Windows
# Sequence types with timestep #1:
# <BatchDataset element_spec=(TensorSpec(shape=(None, 1, 13), dtype=tf.int32, name=None), TensorSpec(shape=(None, 13), dtype=tf.int32, name=None))>
# Sequence types with timestep #2:
# <BatchDataset element_spec=(TensorSpec(shape=(None, 1, 13), dtype=tf.int32, name=None), TensorSpec(shape=(None, 13), dtype=tf.int32, name=None))>
[ Result ]:
Related
import numpy as np
from matplotlib import pyplot as plt
data = np.array([[0.8, 2.4, 2.5, 3.9, 0.0, 4.0, 0.0],
[2.4, 0.0, 4.0, 1.0, 2.7, 0.0, 0.0],
[1.1, 2.4, 0.8, 4.3, 1.9, 4.4, 0.0],
[0.6, 0.0, 0.3, 0.0, 3.1, 0.0, 0.0],
[0.7, 1.7, 0.6, 2.6, 2.2, 6.2, 0.0],
[1.3, 1.2, 0.0, 0.0, 0.0, 3.2, 5.1],
[0.1, 2.0, 0.0, 1.4, 0.0, 1.9, 6.3]])
plt.figure(figsize=(6, 4))
im = plt.imshow(data, cmap="YlGn")
linewidth = 2
for axis in ['top', 'bottom', 'left', 'right']:
plt.gca().spines[axis].set_linewidth(linewidth)
plt.gca().set_xticks(np.arange(data.shape[1] + 1) - .5, minor=True)
plt.gca().set_yticks(np.arange(data.shape[0] + 1) - .5, minor=True)
plt.gca().grid(which="minor", color="black", linewidth=linewidth)
plt.gca().tick_params(which="minor", bottom=False, left=False)
plt.tight_layout()
plt.gca().set_xticks(ticks=[])
plt.gca().set_yticks(ticks=[])
plt.savefig("test.pdf",
bbox_inches="tight",
transparent="True",
pad_inches=1.0/72.0 * linewidth / 2.0)
This code will output the following pdf, but you can see that there are white borders on the left and bottom, so the pdf is not centered after being inserted into LaTex. How to solve this problem?
plt result:
import numpy as np
from matplotlib import pyplot as plt
data = np.array([[0.8, 2.4, 2.5, 3.9, 0.0, 4.0, 0.0],
[2.4, 0.0, 4.0, 1.0, 2.7, 0.0, 0.0],
[1.1, 2.4, 0.8, 4.3, 1.9, 4.4, 0.0],
[0.6, 0.0, 0.3, 0.0, 3.1, 0.0, 0.0],
[0.7, 1.7, 0.6, 2.6, 2.2, 6.2, 0.0],
[1.3, 1.2, 0.0, 0.0, 0.0, 3.2, 5.1],
[0.1, 2.0, 0.0, 1.4, 0.0, 1.9, 6.3]])
plt.figure(figsize=(6, 4))
im = plt.imshow(data, cmap="YlGn")
linewidth = 2
for axis in ['top', 'bottom', 'left', 'right']:
plt.gca().spines[axis].set_linewidth(linewidth)
plt.gca().set_xticks(np.arange(data.shape[1] + 1) - .5, minor=True)
plt.gca().set_yticks(np.arange(data.shape[0] + 1) - .5, minor=True)
plt.gca().grid(which="minor", color="black", linewidth=linewidth)
plt.gca().tick_params(which="minor", bottom=False, left=False)
plt.tight_layout()
plt.gca().set_xticks(ticks=[])
plt.gca().set_yticks(ticks=[])
plt.gca().tick_params(axis="both",
which="major",
left=False,
bottom=False,
labelleft=False,
labelbottom=False)
plt.savefig("test.pdf",
bbox_inches="tight",
transparent="True",
pad_inches=1.0 / 72.0 * linewidth / 2.0)
It was an issue with ticks, solved now.
I have a large dataset with thousands of rows though fewer columns, i have ordered them by row values so that each of the 'objects' are grouped together, just like the dataset in Table1 below:
#Table1 :
data = [['ALFA', 351740.00, 0.31, 0.22, 0.44, 0.19, 0.05],
['ALFA', 401740.00, 0.43, 0.26, 0.23, 0.16, 0.09],
['ALFA', 892350.00, 0.58, 0.24, 0.05, 0.07, 0.4],
['Bravo', 511830.00, 0.52, 0.16, 0.08, 0.26, 0],
['Charlie', 590030.00, 0.75, 0.2, 0.14, 0.37, 0.06],
['Charlie', 590030.00, 0.75, 0.2, 0.27, 0.2, 0.01],
['Charlie', 590030.00, 0.75, 0.2, 0.29, 0.11, 0.04],
['Charlie', 590030.00, 0.75, 0.2, 0.27, 0.2, 0.01],
['Charlie', 401740.00, 0.43, 0.26, 0.14, 0.37, 0.06],
['Charlie', 511830.00, 0.52, 0.16, 0.13, 0.22, 0.01],
['Delta', 590030.00, 0.75, 0.2, 0.34, 0.3, 0],
['Delta', 590030.00, 0.75, 0.2, 0, 0.28, 0],
['Delta', 351740.00, 0.31, 0.22, 0.44, 0.19, 0.05],
['Echo', 892350.00, 0.58, 0.24, 0.23, 0.16, 0.09],
['Echo', 590030.00, 0.75, 0.2, 0.05, 0.07, 0.4],
['Echo', 590030.00, 0.75, 0.2, 0.08, 0.26, 0],
['Echo', 590030.00, 0.75, 0.2, 0.14, 0.37, 0.06],
['Foxtrot', 401740.00, 0.43, 0.26, 0.27, 0.2, 0.01],
['Foxtrot', 511830.00, 0.52, 0.16, 0.29, 0.11, 0.04],
['Golf', 590030.00, 0.75, 0.2, 0.27, 0.2, 0.01],
['Golf', 590030.00, 0.75, 0.2, 0.14, 0.37, 0.06],
['Golf', 351740.00, 0.31, 0.22, 0.13, 0.22, 0.01],
['Hotel', 892350.00, 0.58, 0.24, 0.34, 0.3, 0],
['Hotel', 590030.00, 0.75, 0.2, 0, 0.28, 0],
['Hotel', 590030.00, 0.75, 0.2, 0.29, 0.11, 0.04]]
df = pd.DataFrame(data, columns= ['Objects', 'Column1', 'Column2', 'Column3', 'Column4', 'Column5', 'Column6'])
df
However i would like to write a query to go through the dataset, partition the data by these objects and get only the averages for all the columns (for each object) in a separate table much like the Table2 below:
#Table2:
data2 = [['ALFA', 548610.00, 0.44, 0.24, 0.24, 0.14, 0.18],
['Bravo', 511830.00, 0.52, 0.16, 0.08, 0.26, 0],
['Charlie', 545615.00, 0.66, 0.20, 0.21, 0.25, 0.03],
['Delta', 510600.00, 0.60, 0.21, 0.26, 0.26, 0.02],
['Echo', 665610.00, 0.71, 0.21, 0.13, 0.22, 0.14],
['Foxtrot', 456785.00, 0.48, 0.21, 0.28, 0.16, 0.03],
['Golf', 510600.00, 0.60, 0.21, 0.18, 0.26, 0.03],
['Hotel', 690803.33, 0.69, 0.21, 0.21, 0.23, 0.01]]
df2 = pd.DataFrame(data, columns= ['Objects', 'Column1', 'Column2', 'Column3', 'Column4', 'Column5', 'Column6'])
df2
Please note that the number of the objects vary across the dataset so the query should be able to count the number of objects and use that number to get the average of all the columns for each object and then present all these values in a new table (much Like what partition windows function does).
For instance note that the '548610.00' alues in Table2 for ALFA(column1) is merely an addition of Column1 values of ALFA in Table1 (351740.00 + 401740.00 + 401740.00) and divide by the count of ALFA being '3'
I believe a simple avg() function should answer your question
SELECT Objects,
AVG(Column1),
AVG(Column2),
AVG(Column3),
AVG(Column4),
AVG(Column5),
AVG(Column6)
FROM tableA
GROUP BY Objects
db fiddle link
I want create a list from 0 to 1 with step 0.05, the result will like this: [0, 0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.45, 0.5, 0.55, 0.6, 0.65, 0.7, 0.75, 0.8, 0.85, 0.9, 0.95, 1]
I try with following code, but the output seems not correct. Anyone could help? Thanks.
print(np.arange(0, 1, 0.05).tolist())
Output:
[0.0, 0.05, 0.1, 0.15000000000000002, 0.2, 0.25, 0.30000000000000004, 0.35000000000000003, 0.4, 0.45, 0.5, 0.55, 0.6000000000000001, 0.65, 0.7000000000000001, 0.75, 0.8, 0.8500000000000001, 0.9, 0.9500000000000001]
You want np.linspace()
np.linspace(0, 1, 21)
Out[]:
array([0. , 0.05, 0.1 , 0.15, 0.2 , 0.25, 0.3 , 0.35, 0.4 , 0.45, 0.5 ,
0.55, 0.6 , 0.65, 0.7 , 0.75, 0.8 , 0.85, 0.9 , 0.95, 1. ])
Its not necessary use .tolist().
Try this:
a = np.arange(0, 1, 0.05)
print (a)
Output:
[0. 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65
0.7 0.75 0.8 0.85 0.9 0.95]
This works:
print(np.arange(0, 1, 0.05).round(2).tolist())
Output:
[0.0, 0.05, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.45, 0.5, 0.55, 0.6, 0.65, 0.7, 0.75, 0.8, 0.85, 0.9, 0.95]
I'm using the tf.unsorted_segment_sum method of TensorFlow and it works.
For example:
tf.unsorted_segment_sum(tf.constant([0.2, 0.1, 0.5, 0.7, 0.8]),
tf.constant([0, 0, 1, 2, 2]), 3)
Gives the right result:
array([ 0.3, 0.5 , 1.5 ], dtype=float32)
I want to get:
array([0.3, 0.3, 0.5, 1.5, 1.5], dtype=float32)
I've solved it.
data = tf.constant([0.2, 0.1, 0.5, 0.7, 0.8])
gr_idx = tf.constant([0, 0, 1, 2, 2])
y, idx, count = tf.unique_with_count(gr_idx)
group_sum = tf.segment_sum(data, gr_idx)
group_sup = tf.gather(group_sum, idx)
answer:
array([0.3, 0.3, 0.5, 1.5, 1.5], dtype=float32)
Need to plot a confusion matrix with this script. By running it an empty plot appears. Seems I am close to solution. Any hint?
from numpy import *
import matplotlib.pyplot as plt
from pylab import *
conf_arr = [[50.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [3.0, 26.0, 0.0, 0.0, 0.0, 1.0, 0.0], [0.0, 0.0, 10.0, 0.0, 0.0, 0.0, 0.0], [4.0, 1.0, 0.0, 5.0, 0.0, 0.0, 0.0], [3.0, 0.0, 1.0, 0.0, 6.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 47.0, 0.0], [2.0, 0.0, 0.0, 0.0, 0.0, 0.0, 8.0]]
norm_conf = []
for i in conf_arr:
a = 0
tmp_arr = []
a = sum(i,0)
for j in i:
tmp_arr.append(float(j)/float(a))
norm_conf.append(tmp_arr)
plt.clf()
fig = plt.figure()
ax = fig.add_subplot(111)
res = ax.imshow(array(norm_conf), cmap=cm.jet, interpolation='nearest')
cb = fig.colorbar(res)
savefig("confmat.png", format="png")
Thanks, I have the plot. Now, the ticks in the x-axes are very small (the graph dimension is: 3 cm x 10 cm or so). How can I enlarge them in order to have a more proportioned graph, lets say 10cm x 10 cm plot? A possible reason is that I visualize the graph as a subplot? Was not able to find the suitable literature to adjust that.
You don't need to clear a current figure (plt.clf()) before adding a new one.
#plt.clf() # <<<<< here
fig = plt.figure()
ax = fig.add_subplot(111)