Adds exploratory script for SF and Pennylane. #2

say4n · 2021-09-18T10:37:24Z

No description provided.

FIXME: Training fails with TypeError: Dimension value must be integer or None or have an __index__ method, got value '1.0' with type '<class 'float'>'

FIXME: Traceback is as follows: TypeError: Exception encountered when calling layer "keras_layer" (type KerasLayer). Cannot interpret '2' as a data type Call arguments received: • inputs=tf.Tensor(shape=(100, 2), dtype=int64) > /Users/Sayan/Desktop/Projects/coherent-state-discrimination/examples/explore_pennylane.py(85)<module>() -> model.fit(X_batch, y_batch, epochs=5, batch_size=BATCH_SIZE)

say4n · 2021-10-28T05:25:13Z

Training presently. Logs as shown below.

❯ python3 explore_pennylane.py
2021-10-28 10:37:11.587161: I tensorflow/core/platform/cpu_feature_guard.cc:151] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
Iteration 1 of 10.
Epoch 1/50
1/1 [==============================] - 51s 51s/step - loss: 0.3759 - accuracy: 0.2540
Epoch 2/50
1/1 [==============================] - 47s 47s/step - loss: 0.3729 - accuracy: 0.2540
Epoch 3/50
1/1 [==============================] - 48s 48s/step - loss: 0.3687 - accuracy: 0.2540
Epoch 4/50
1/1 [==============================] - 48s 48s/step - loss: 0.3583 - accuracy: 0.2540
Epoch 5/50
1/1 [==============================] - 48s 48s/step - loss: 0.3430 - accuracy: 0.2540
Epoch 6/50
1/1 [==============================] - 51s 51s/step - loss: 0.3276 - accuracy: 0.5330
Epoch 7/50
1/1 [==============================] - 56s 56s/step - loss: 0.3072 - accuracy: 0.5330
Epoch 8/50
1/1 [==============================] - 56s 56s/step - loss: 0.2901 - accuracy: 0.5330
Epoch 9/50
1/1 [==============================] - 54s 54s/step - loss: 0.2806 - accuracy: 0.5330
Epoch 10/50
1/1 [==============================] - 53s 53s/step - loss: 0.2712 - accuracy: 0.5330
Epoch 11/50
1/1 [==============================] - 56s 56s/step - loss: 0.2606 - accuracy: 0.5330
Epoch 12/50
1/1 [==============================] - 52s 52s/step - loss: 0.2519 - accuracy: 0.5330
Epoch 13/50
1/1 [==============================] - 47s 47s/step - loss: 0.2478 - accuracy: 0.5330
Epoch 14/50
1/1 [==============================] - 47s 47s/step - loss: 0.2473 - accuracy: 0.5330
Epoch 15/50
1/1 [==============================] - 47s 47s/step - loss: 0.2441 - accuracy: 0.5330
Epoch 16/50
1/1 [==============================] - 47s 47s/step - loss: 0.2405 - accuracy: 0.5330
Epoch 17/50
1/1 [==============================] - 47s 47s/step - loss: 0.2392 - accuracy: 0.5330
Epoch 18/50
1/1 [==============================] - 49s 49s/step - loss: 0.2385 - accuracy: 0.5330
Epoch 19/50
1/1 [==============================] - 49s 49s/step - loss: 0.2377 - accuracy: 0.5330
Epoch 20/50
--- 8< ---

say4n · 2021-10-28T07:51:20Z

Here is the complete console log on training.

❯ python3 explore_pennylane.py
2021-10-28 11:44:22.764471: I tensorflow/core/platform/cpu_feature_guard.cc:151] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
Iteration 1 of 5.
Epoch 1/20
1/1 [==============================] - 56s 56s/step - loss: 0.3742 - accuracy: 0.2380
Epoch 2/20
1/1 [==============================] - 52s 52s/step - loss: 0.3711 - accuracy: 0.2710
Epoch 3/20
1/1 [==============================] - 56s 56s/step - loss: 0.3684 - accuracy: 0.2710
Epoch 4/20
1/1 [==============================] - 54s 54s/step - loss: 0.3645 - accuracy: 0.2710
Epoch 5/20
1/1 [==============================] - 48s 48s/step - loss: 0.3576 - accuracy: 0.5310
Epoch 6/20
1/1 [==============================] - 47s 47s/step - loss: 0.3480 - accuracy: 0.2600
Epoch 7/20
1/1 [==============================] - 48s 48s/step - loss: 0.3371 - accuracy: 0.4980
Epoch 8/20
1/1 [==============================] - 48s 48s/step - loss: 0.3256 - accuracy: 0.4980
Epoch 9/20
1/1 [==============================] - 47s 47s/step - loss: 0.3114 - accuracy: 0.4980
Epoch 10/20
1/1 [==============================] - 48s 48s/step - loss: 0.2973 - accuracy: 0.4980
Epoch 11/20
1/1 [==============================] - 48s 48s/step - loss: 0.2872 - accuracy: 0.4980
Epoch 12/20
1/1 [==============================] - 48s 48s/step - loss: 0.2809 - accuracy: 0.4980
Epoch 13/20
1/1 [==============================] - 47s 47s/step - loss: 0.2752 - accuracy: 0.4980
Epoch 14/20
1/1 [==============================] - 47s 47s/step - loss: 0.2679 - accuracy: 0.4980
Epoch 15/20
1/1 [==============================] - 48s 48s/step - loss: 0.2595 - accuracy: 0.4980
Epoch 16/20
1/1 [==============================] - 47s 47s/step - loss: 0.2503 - accuracy: 0.4980
Epoch 17/20
1/1 [==============================] - 47s 47s/step - loss: 0.2397 - accuracy: 0.7290
Epoch 18/20
1/1 [==============================] - 48s 48s/step - loss: 0.2284 - accuracy: 0.7290
Epoch 19/20
1/1 [==============================] - 47s 47s/step - loss: 0.2175 - accuracy: 0.7290
Epoch 20/20
1/1 [==============================] - 47s 47s/step - loss: 0.2087 - accuracy: 0.7290
Iteration 2 of 5.
Epoch 1/20
1/1 [==============================] - 48s 48s/step - loss: 0.1895 - accuracy: 0.7670
Epoch 2/20
1/1 [==============================] - 48s 48s/step - loss: 0.1825 - accuracy: 0.7670
Epoch 3/20
1/1 [==============================] - 51s 51s/step - loss: 0.1749 - accuracy: 0.7670
Epoch 4/20
1/1 [==============================] - 50s 50s/step - loss: 0.1685 - accuracy: 0.7670
Epoch 5/20
1/1 [==============================] - 48s 48s/step - loss: 0.1637 - accuracy: 0.7670
Epoch 6/20
1/1 [==============================] - 49s 49s/step - loss: 0.1590 - accuracy: 0.7670
Epoch 7/20
1/1 [==============================] - 47s 47s/step - loss: 0.1553 - accuracy: 0.7670
Epoch 8/20
1/1 [==============================] - 47s 47s/step - loss: 0.1525 - accuracy: 0.7670
Epoch 9/20
1/1 [==============================] - 48s 48s/step - loss: 0.1494 - accuracy: 0.7670
Epoch 10/20
1/1 [==============================] - 47s 47s/step - loss: 0.1459 - accuracy: 0.7670
Epoch 11/20
1/1 [==============================] - 48s 48s/step - loss: 0.1429 - accuracy: 0.7670
Epoch 12/20
1/1 [==============================] - 48s 48s/step - loss: 0.1405 - accuracy: 0.7670
Epoch 13/20
1/1 [==============================] - 48s 48s/step - loss: 0.1385 - accuracy: 0.7670
Epoch 14/20
1/1 [==============================] - 47s 47s/step - loss: 0.1364 - accuracy: 0.7670
Epoch 15/20
1/1 [==============================] - 48s 48s/step - loss: 0.1345 - accuracy: 0.7670
Epoch 16/20
1/1 [==============================] - 47s 47s/step - loss: 0.1329 - accuracy: 0.7670
Epoch 17/20
1/1 [==============================] - 47s 47s/step - loss: 0.1316 - accuracy: 0.7670
Epoch 18/20
1/1 [==============================] - 49s 49s/step - loss: 0.1305 - accuracy: 0.7670
Epoch 19/20
1/1 [==============================] - 48s 48s/step - loss: 0.1295 - accuracy: 0.7670
Epoch 20/20
1/1 [==============================] - 49s 49s/step - loss: 0.1286 - accuracy: 0.7670
Iteration 3 of 5.
Epoch 1/20
1/1 [==============================] - 47s 47s/step - loss: 0.1473 - accuracy: 0.7270
Epoch 2/20
1/1 [==============================] - 48s 48s/step - loss: 0.1467 - accuracy: 0.7270
Epoch 3/20
1/1 [==============================] - 48s 48s/step - loss: 0.1461 - accuracy: 0.7270
Epoch 4/20
1/1 [==============================] - 48s 48s/step - loss: 0.1456 - accuracy: 0.7270
Epoch 5/20
1/1 [==============================] - 48s 48s/step - loss: 0.1450 - accuracy: 0.7270
Epoch 6/20
1/1 [==============================] - 48s 48s/step - loss: 0.1446 - accuracy: 0.7270
Epoch 7/20
1/1 [==============================] - 48s 48s/step - loss: 0.1441 - accuracy: 0.7270
Epoch 8/20
1/1 [==============================] - 48s 48s/step - loss: 0.1437 - accuracy: 0.7270
Epoch 9/20
1/1 [==============================] - 48s 48s/step - loss: 0.1434 - accuracy: 0.7270
Epoch 10/20
1/1 [==============================] - 48s 48s/step - loss: 0.1431 - accuracy: 0.7270
Epoch 11/20
1/1 [==============================] - 47s 47s/step - loss: 0.1428 - accuracy: 0.7270
Epoch 12/20
1/1 [==============================] - 47s 47s/step - loss: 0.1426 - accuracy: 0.7270
Epoch 13/20
1/1 [==============================] - 47s 47s/step - loss: 0.1423 - accuracy: 0.7270
Epoch 14/20
1/1 [==============================] - 47s 47s/step - loss: 0.1421 - accuracy: 0.7270
Epoch 15/20
1/1 [==============================] - 48s 48s/step - loss: 0.1419 - accuracy: 0.7270
Epoch 16/20
1/1 [==============================] - 47s 47s/step - loss: 0.1417 - accuracy: 0.7270
Epoch 17/20
1/1 [==============================] - 47s 47s/step - loss: 0.1415 - accuracy: 0.7270
Epoch 18/20
1/1 [==============================] - 47s 47s/step - loss: 0.1413 - accuracy: 0.7270
Epoch 19/20
1/1 [==============================] - 47s 47s/step - loss: 0.1412 - accuracy: 0.7270
Epoch 20/20
1/1 [==============================] - 47s 47s/step - loss: 0.1410 - accuracy: 0.7270
Iteration 4 of 5.
Epoch 1/20
1/1 [==============================] - 48s 48s/step - loss: 0.1207 - accuracy: 0.7680
Epoch 2/20
1/1 [==============================] - 48s 48s/step - loss: 0.1206 - accuracy: 0.7680
Epoch 3/20
1/1 [==============================] - 48s 48s/step - loss: 0.1204 - accuracy: 0.7680
Epoch 4/20
1/1 [==============================] - 48s 48s/step - loss: 0.1203 - accuracy: 0.7680
Epoch 5/20
1/1 [==============================] - 48s 48s/step - loss: 0.1202 - accuracy: 0.7680
Epoch 6/20
1/1 [==============================] - 47s 47s/step - loss: 0.1201 - accuracy: 0.7680
Epoch 7/20
1/1 [==============================] - 47s 47s/step - loss: 0.1200 - accuracy: 0.7680
Epoch 8/20
1/1 [==============================] - 47s 47s/step - loss: 0.1199 - accuracy: 0.7680
Epoch 9/20
1/1 [==============================] - 48s 48s/step - loss: 0.1199 - accuracy: 0.7680
Epoch 10/20
1/1 [==============================] - 48s 48s/step - loss: 0.1198 - accuracy: 0.7680
Epoch 11/20
1/1 [==============================] - 47s 47s/step - loss: 0.1197 - accuracy: 0.7680
Epoch 12/20
1/1 [==============================] - 48s 48s/step - loss: 0.1196 - accuracy: 0.7680
Epoch 13/20
1/1 [==============================] - 47s 47s/step - loss: 0.1195 - accuracy: 0.7680
Epoch 14/20
1/1 [==============================] - 47s 47s/step - loss: 0.1195 - accuracy: 0.7680
Epoch 15/20
1/1 [==============================] - 48s 48s/step - loss: 0.1194 - accuracy: 0.7680
Epoch 16/20
1/1 [==============================] - 47s 47s/step - loss: 0.1193 - accuracy: 0.7680
Epoch 17/20
1/1 [==============================] - 47s 47s/step - loss: 0.1193 - accuracy: 0.7680
Epoch 18/20
1/1 [==============================] - 48s 48s/step - loss: 0.1192 - accuracy: 0.7680
Epoch 19/20
1/1 [==============================] - 47s 47s/step - loss: 0.1192 - accuracy: 0.7680
Epoch 20/20
1/1 [==============================] - 47s 47s/step - loss: 0.1191 - accuracy: 0.7680
Iteration 5 of 5.
Epoch 1/20
1/1 [==============================] - 47s 47s/step - loss: 0.1344 - accuracy: 0.7370
Epoch 2/20
1/1 [==============================] - 48s 48s/step - loss: 0.1344 - accuracy: 0.7370
Epoch 3/20
1/1 [==============================] - 47s 47s/step - loss: 0.1343 - accuracy: 0.7370
Epoch 4/20
1/1 [==============================] - 48s 48s/step - loss: 0.1343 - accuracy: 0.7370
Epoch 5/20
1/1 [==============================] - 47s 47s/step - loss: 0.1342 - accuracy: 0.7370
Epoch 6/20
1/1 [==============================] - 48s 48s/step - loss: 0.1342 - accuracy: 0.7370
Epoch 7/20
1/1 [==============================] - 48s 48s/step - loss: 0.1342 - accuracy: 0.7370
Epoch 8/20
1/1 [==============================] - 47s 47s/step - loss: 0.1341 - accuracy: 0.7370
Epoch 9/20
1/1 [==============================] - 47s 47s/step - loss: 0.1341 - accuracy: 0.7370
Epoch 10/20
1/1 [==============================] - 48s 48s/step - loss: 0.1340 - accuracy: 0.7370
Epoch 11/20
1/1 [==============================] - 48s 48s/step - loss: 0.1340 - accuracy: 0.7370
Epoch 12/20
1/1 [==============================] - 47s 47s/step - loss: 0.1340 - accuracy: 0.7370
Epoch 13/20
1/1 [==============================] - 48s 48s/step - loss: 0.1339 - accuracy: 0.7370
Epoch 14/20
1/1 [==============================] - 47s 47s/step - loss: 0.1339 - accuracy: 0.7370
Epoch 15/20
1/1 [==============================] - 48s 48s/step - loss: 0.1339 - accuracy: 0.7370
Epoch 16/20
1/1 [==============================] - 47s 47s/step - loss: 0.1338 - accuracy: 0.7370
Epoch 17/20
1/1 [==============================] - 47s 47s/step - loss: 0.1338 - accuracy: 0.7370
Epoch 18/20
1/1 [==============================] - 48s 48s/step - loss: 0.1338 - accuracy: 0.7370
Epoch 19/20
1/1 [==============================] - 48s 48s/step - loss: 0.1337 - accuracy: 0.7370
Epoch 20/20
1/1 [==============================] - 47s 47s/step - loss: 0.1337 - accuracy: 0.7370
Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #
=================================================================
 keras_layer (KerasLayer)    (1000, 2)                 16

 dense (Dense)               (1000, 4)                 12

=================================================================
Total params: 28
Trainable params: 28
Non-trainable params: 0
_________________________________________________________________

say4n · 2021-11-03T04:49:32Z

Summary

Input

A codeword that characterises a n-mode CV input. For example for the 2 mode case, the inputs can be 00, 01, 10, 11.

Output of the "Network"

A one hot output (ideally) that decodes the original signal. For example for the two mode case it will be as follows:

Codeword	One-hot vector
{-a, -a}	[1 0 0 0]
{-a, +a}	[0 1 0 0]
{+a, -a}	[0 0 1 0]
{+a, +a}	[0 0 0 1]

Architecture

CNNeuralNet layer with displacement embedding as keras layer followed by a fully connected layer. Recurrence needs to be added.

Loss Function

Optimisation is performed with the mean average error loss function over the target and the output vector across the entire training batch.

TODO

Ancilla Inputs
Recurrence

say4n · 2021-11-09T08:01:17Z

The following training logs are for the 2-mode case where the target is to predict {+a, +a} no matter what the input is. In other words, this can be thought of as learning displacement such that we are able to translate the input to {+a, +a} no matter what the input is (this needs some discussion).

❯ ❯ python3 explore_pennylane.py
2021-11-09 13:32:12.477830: I tensorflow/core/platform/cpu_feature_guard.cc:151] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
Iteration 1 of 5.
Epoch 1/5
1/1 [==============================] - 51s 51s/step - loss: 0.3802 - accuracy: 0.0000e+00
Epoch 2/5
1/1 [==============================] - 60s 60s/step - loss: 0.3524 - accuracy: 1.0000
Epoch 3/5
1/1 [==============================] - 52s 52s/step - loss: 0.3240 - accuracy: 1.0000
Epoch 4/5
1/1 [==============================] - 52s 52s/step - loss: 0.2934 - accuracy: 1.0000
Epoch 5/5
1/1 [==============================] - 51s 51s/step - loss: 0.2595 - accuracy: 1.0000
Iteration 2 of 5.
Epoch 1/5
1/1 [==============================] - 54s 54s/step - loss: 0.2220 - accuracy: 1.0000
Epoch 2/5
1/1 [==============================] - 50s 50s/step - loss: 0.1823 - accuracy: 1.0000
Epoch 3/5
1/1 [==============================] - 49s 49s/step - loss: 0.1447 - accuracy: 1.0000
Epoch 4/5
1/1 [==============================] - 50s 50s/step - loss: 0.1124 - accuracy: 1.0000
Epoch 5/5
1/1 [==============================] - 49s 49s/step - loss: 0.0861 - accuracy: 1.0000
Iteration 3 of 5.
Epoch 1/5
1/1 [==============================] - 49s 49s/step - loss: 0.0639 - accuracy: 1.0000
Epoch 2/5
1/1 [==============================] - 49s 49s/step - loss: 0.0458 - accuracy: 1.0000
Epoch 3/5
1/1 [==============================] - 50s 50s/step - loss: 0.0320 - accuracy: 1.0000
Epoch 4/5
1/1 [==============================] - 50s 50s/step - loss: 0.0223 - accuracy: 1.0000
Epoch 5/5
1/1 [==============================] - 50s 50s/step - loss: 0.0156 - accuracy: 1.0000
Iteration 4 of 5.
Epoch 1/5
1/1 [==============================] - 50s 50s/step - loss: 0.0112 - accuracy: 1.0000
Epoch 2/5
1/1 [==============================] - 50s 50s/step - loss: 0.0080 - accuracy: 1.0000
Epoch 3/5
1/1 [==============================] - 49s 49s/step - loss: 0.0058 - accuracy: 1.0000
Epoch 4/5
1/1 [==============================] - 50s 50s/step - loss: 0.0042 - accuracy: 1.0000
Epoch 5/5
1/1 [==============================] - 51s 51s/step - loss: 0.0031 - accuracy: 1.0000
Iteration 5 of 5.
Epoch 1/5
1/1 [==============================] - 50s 50s/step - loss: 0.0023 - accuracy: 1.0000
Epoch 2/5
1/1 [==============================] - 50s 50s/step - loss: 0.0017 - accuracy: 1.0000
Epoch 3/5
1/1 [==============================] - 51s 51s/step - loss: 0.0014 - accuracy: 1.0000
Epoch 4/5
1/1 [==============================] - 48s 48s/step - loss: 0.0011 - accuracy: 1.0000
Epoch 5/5
1/1 [==============================] - 48s 48s/step - loss: 9.0213e-04 - accuracy: 1.0000
Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #
=================================================================
 keras_layer (KerasLayer)    (1000, 2)                 16

 dense (Dense)               (1000, 4)                 12

=================================================================
Total params: 28
Trainable params: 28
Non-trainable params: 0
_________________________________________________________________

say4n and others added 11 commits September 18, 2021 16:06

Adds exploratory script for SF.

3a8e6e8

Adds parameterization for SF program.

8206f8e

chore: add prob function

12d859b

Adds optimization, switches to tf backend.

ae337cf

Restructures cells, adds explicit loss function.

885e78f

Restructures cells to prepare for batched updates.

f25e653

Add batched update.

9cd68b2

Work in progress.

142694f

Adds CVLayer (WIP).

ccee3d1

Add keras stuff.

e0653af

Adds training and simulation.

fa25e0a

FIXME: Training fails with TypeError: Dimension value must be integer or None or have an __index__ method, got value '1.0' with type '<class 'float'>'

say4n changed the title ~~Adds exploratory script for SF.~~ Adds exploratory script for SF and Pennylane. Oct 25, 2021

say4n added 2 commits October 27, 2021 12:57

Fixes errors.

29c7640

say4n added 3 commits November 9, 2021 13:36

Predict {+a, +a} no matter what the input.

9b79eab

Adds logging, plotting.

c06541d

Push changes.

fefddda

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Adds exploratory script for SF and Pennylane. #2

Adds exploratory script for SF and Pennylane. #2

say4n commented Sep 18, 2021

say4n commented Oct 28, 2021 •

edited

Loading

say4n commented Oct 28, 2021

say4n commented Nov 3, 2021 •

edited by MattRos90

Loading

say4n commented Nov 9, 2021 •

edited

Loading

Adds exploratory script for SF and Pennylane. #2

Are you sure you want to change the base?

Adds exploratory script for SF and Pennylane. #2

Conversation

say4n commented Sep 18, 2021

say4n commented Oct 28, 2021 • edited Loading

say4n commented Oct 28, 2021

say4n commented Nov 3, 2021 • edited by MattRos90 Loading

Summary

Input

Output of the "Network"

Architecture

Loss Function

TODO

say4n commented Nov 9, 2021 • edited Loading

say4n commented Oct 28, 2021 •

edited

Loading

say4n commented Nov 3, 2021 •

edited by MattRos90

Loading

say4n commented Nov 9, 2021 •

edited

Loading